diff --git a/.readthedocs.yaml b/.readthedocs.yaml
deleted file mode 100644
index 80148e0..0000000
--- a/.readthedocs.yaml
+++ /dev/null
@@ -1,18 +0,0 @@
-# Read the Docs configuration file
-# See https://docs.readthedocs.io/en/stable/config-file/v2.html for details
-
-version: 2
-
-build:
-  os: ubuntu-22.04
-  tools:
-    python: "3.10"
-
-python:
-  install:
-  - requirements: docs/sphinx/requirements.txt
-
-sphinx:
-  configuration: docs/conf.py
-
-formats: [htmlzip, pdf]
diff --git a/build_docs.py b/build_docs.py
new file mode 100644
index 0000000..5667b75
--- /dev/null
+++ b/build_docs.py
@@ -0,0 +1,198 @@
+#!/usr/bin/env python3
+"""Discover and build Sphinx docsets in this monorepo.
+
+A "docset" is any folder in the repo that contains a ``conf.py`` (searched
+recursively, so docsets may be nested e.g. ``reference-architecture/MI3XX``).
+This script finds them automatically, so adding a new docset folder requires
+no changes to tasks.json or to this script.
+
+Usage:
+    python build_docs.py --list
+        Print every discovered docset (name + path), one per line.
+
+    python build_docs.py --all [--clean]
+        Build every discovered docset.
+
+    python build_docs.py --docset <name> [--clean]
+        Build a single docset by folder name (e.g. "docs").
+
+    python build_docs.py --file <path> [--clean]
+        Build whichever docset the given file lives in. Used by the VS Code
+        "Build Docs" task with ${file} so Ctrl+B builds the docset you're
+        currently editing.
+
+Each docset builds into its own namespaced output + doctree cache under
+    <repo>/.vscode/build/<docset>/{html,doctrees}
+so docsets never clobber each other or share a (project-specific) doctree
+cache.
+
+Exit codes:
+    0  success
+    1  usage / discovery error (e.g. file not under any docset)
+    N  the underlying sphinx-build exit code on build failure
+"""
+
+from __future__ import annotations
+
+import argparse
+import os
+import shutil
+import subprocess
+import sys
+from pathlib import Path
+
+
+# Repo root = the directory containing this script.
+REPO_ROOT = Path(__file__).resolve().parent
+BUILD_ROOT = REPO_ROOT / ".vscode" / "build"
+
+# Folders that can never be docsets even if they somehow contain a conf.py.
+IGNORE_DIRS = {".git", ".vscode", ".venv", "venv", "node_modules", "__pycache__"}
+
+
+def _docset_name(src: Path, taken: set[str]) -> str:
+    """Pick a unique docset name for a source dir.
+
+    Prefer the folder's own basename (e.g. "docs", "MI3XX"). If that collides
+    with an already-claimed name, fall back to the repo-relative path with the
+    OS separator replaced by "-" (e.g. "reference-architecture-MI3XX") so every
+    docset stays addressable on the CLI and gets its own build output dir.
+    """
+    name = src.name
+    if name not in taken:
+        return name
+    return "-".join(src.relative_to(REPO_ROOT).parts)
+
+
+def discover_docsets() -> dict[str, Path]:
+    """Return {docset_name: source_dir} for every folder holding a conf.py.
+
+    Walks the repo recursively (not just top-level children) so docsets may be
+    nested, e.g. ``reference-architecture/MI3XX/conf.py``. Ignored folders are
+    pruned, and once a folder is identified as a docset we do not descend into
+    it (a docset never contains another docset). Sorted for stable ordering.
+    """
+    docsets: dict[str, Path] = {}
+    for dirpath, dirnames, filenames in os.walk(REPO_ROOT):
+        # Prune ignored dirs in place so os.walk never descends into them.
+        dirnames[:] = sorted(d for d in dirnames if d not in IGNORE_DIRS)
+        if "conf.py" in filenames:
+            src = Path(dirpath)
+            docsets[_docset_name(src, set(docsets))] = src
+            # Don't look for docsets nested inside this one.
+            dirnames[:] = []
+    return dict(sorted(docsets.items()))
+
+
+def docset_for_file(file_path: Path, docsets: dict[str, Path]) -> str | None:
+    """Return the docset name that ``file_path`` belongs to, or None."""
+    try:
+        resolved = file_path.resolve()
+    except OSError:
+        return None
+    for name, src in docsets.items():
+        try:
+            resolved.relative_to(src.resolve())
+            return name
+        except ValueError:
+            continue
+    return None
+
+
+def build_one(name: str, src: Path, clean: bool) -> int:
+    """Build a single docset with Sphinx. Returns the sphinx-build exit code."""
+    out_html = BUILD_ROOT / name / "html"
+    out_doctrees = BUILD_ROOT / name / "doctrees"
+
+    if clean:
+        target = BUILD_ROOT / name
+        if target.exists():
+            print(f"[clean] removing {target}")
+            shutil.rmtree(target, ignore_errors=True)
+
+    out_html.mkdir(parents=True, exist_ok=True)
+    out_doctrees.mkdir(parents=True, exist_ok=True)
+
+    cmd = [
+        sys.executable, "-m", "sphinx",
+        "-j", "auto",
+        "-T",
+        "-b", "html",
+        "-d", str(out_doctrees),
+        "-D", "language=en",
+        str(src),
+        str(out_html),
+    ]
+    print(f"[build] {name}: {' '.join(cmd)}")
+    result = subprocess.run(cmd, cwd=str(REPO_ROOT))
+    if result.returncode == 0:
+        print(f"[build] {name}: OK -> {out_html}")
+    else:
+        print(f"[build] {name}: FAILED (exit {result.returncode})", file=sys.stderr)
+    return result.returncode
+
+
+def main(argv: list[str]) -> int:
+    parser = argparse.ArgumentParser(description=__doc__,
+                                     formatter_class=argparse.RawDescriptionHelpFormatter)
+    group = parser.add_mutually_exclusive_group(required=True)
+    group.add_argument("--list", action="store_true", help="List discovered docsets and exit.")
+    group.add_argument("--list-names", action="store_true",
+                       help="Print 'all' then each docset name, one per line (for pickers).")
+    group.add_argument("--all", action="store_true", help="Build every docset.")
+    group.add_argument("--docset", metavar="NAME", help="Build a single docset by folder name.")
+    group.add_argument("--file", metavar="PATH", help="Build the docset containing this file.")
+    parser.add_argument("--clean", action="store_true", help="Remove the docset's build output first.")
+    args = parser.parse_args(argv)
+
+    docsets = discover_docsets()
+
+    if args.list_names:
+        print("all")
+        for name in docsets:
+            print(name)
+        return 0
+
+    if args.list:
+        if not docsets:
+            print("(no docsets found — no top-level folder contains a conf.py)")
+            return 0
+        for name, src in docsets.items():
+            print(f"{name}\t{src}")
+        return 0
+
+    if not docsets:
+        print("Error: no docsets found (no top-level folder contains a conf.py).", file=sys.stderr)
+        return 1
+
+    if args.all:
+        rc = 0
+        for name, src in docsets.items():
+            rc = build_one(name, src, args.clean) or rc
+        return rc
+
+    if args.docset:
+        if args.docset == "all":
+            rc = 0
+            for name, src in docsets.items():
+                rc = build_one(name, src, args.clean) or rc
+            return rc
+        if args.docset not in docsets:
+            print(f"Error: '{args.docset}' is not a docset. Known: {', '.join(docsets) or '(none)'}",
+                  file=sys.stderr)
+            return 1
+        return build_one(args.docset, docsets[args.docset], args.clean)
+
+    if args.file:
+        name = docset_for_file(Path(args.file), docsets)
+        if name is None:
+            print(f"Error: '{args.file}' is not inside any docset.\n"
+                  f"Open a file under one of: {', '.join(docsets)}", file=sys.stderr)
+            return 1
+        return build_one(name, docsets[name], args.clean)
+
+    return 0  # unreachable (group is required)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main(sys.argv[1:]))
\ No newline at end of file
diff --git a/docs/.readthedocs.yaml b/docs/.readthedocs.yaml
new file mode 100644
index 0000000..a49c8d8
--- /dev/null
+++ b/docs/.readthedocs.yaml
@@ -0,0 +1,30 @@
+# Read the Docs configuration file
+# See https://docs.readthedocs.io/en/stable/config-file/v2.html for details
+
+version: 2
+
+build:
+  os: ubuntu-22.04
+  tools:
+    python: "3.10"
+  jobs:
+    post_checkout:
+      # Monorepo guard: cancel the build unless this change touched docs/
+      - git fetch origin develop --depth 1 || true
+      - |
+        if [ "$READTHEDOCS_VERSION_TYPE" = "external" ]; then
+          BASE=origin/develop   # PR build: compare against the develop base
+        else
+          BASE=HEAD^            # branch/tag build (latest): compare against previous commit
+        fi
+        if git diff --quiet "$BASE" HEAD -- docs/; then
+          exit 183
+        fi
+python:
+  install:
+  - requirements: docs/sphinx/requirements.txt
+
+sphinx:
+  configuration: docs/conf.py
+
+formats: [htmlzip]
diff --git a/docs/_static/cluster/1024-8192-gpu-reference-cluster-design.pdf b/docs/_static/cluster/1024-8192-gpu-reference-cluster-design.pdf
deleted file mode 100644
index fb21479..0000000
Binary files a/docs/_static/cluster/1024-8192-gpu-reference-cluster-design.pdf and /dev/null differ
diff --git a/docs/_static/cluster/128-1024-gpu-reference-cluster-design.pdf b/docs/_static/cluster/128-1024-gpu-reference-cluster-design.pdf
deleted file mode 100644
index 54f9e28..0000000
Binary files a/docs/_static/cluster/128-1024-gpu-reference-cluster-design.pdf and /dev/null differ
diff --git a/docs/_static/css/custom.css b/docs/_static/css/custom.css
index 787dcb8..e7d3b4a 100644
--- a/docs/_static/css/custom.css
+++ b/docs/_static/css/custom.css
@@ -26,4 +26,20 @@ table.rdma-errors td:nth-child(1) {
 table.rdma-errors th:nth-child(2),
 table.rdma-errors td:nth-child(2) {
     width: 60% !important;
-}
\ No newline at end of file
+}
+
+/* Remove white background from topology diagrams in dark mode */
+html[data-theme=dark] .bd-content img {
+    background-color: transparent !important;
+}
+
+/* Restore normal brightness for images in dark mode */
+html[data-theme=dark] .bd-content img:not(.only-dark,.dark-light) {
+    filter: brightness(1) contrast(1) !important;
+}
+
+/* Add neutral grey background for images in light mode */
+html:not([data-theme=dark]) .bd-content img {
+    background-color: #777 !important;
+}
+
diff --git a/docs/_static/network/355-4MW-PPT-reference-network-design.pdf b/docs/_static/network/355-4MW-PPT-reference-network-design.pdf
deleted file mode 100644
index f396517..0000000
Binary files a/docs/_static/network/355-4MW-PPT-reference-network-design.pdf and /dev/null differ
diff --git a/docs/_static/network/3XX-2K-reference-network-design.pdf b/docs/_static/network/3XX-2K-reference-network-design.pdf
deleted file mode 100644
index 555743a..0000000
Binary files a/docs/_static/network/3XX-2K-reference-network-design.pdf and /dev/null differ
diff --git a/docs/_static/network/3XX-4K-reference-network-design.pdf b/docs/_static/network/3XX-4K-reference-network-design.pdf
deleted file mode 100644
index c589463..0000000
Binary files a/docs/_static/network/3XX-4K-reference-network-design.pdf and /dev/null differ
diff --git a/docs/_static/network/3XX-6K-reference-network-design.pdf b/docs/_static/network/3XX-6K-reference-network-design.pdf
deleted file mode 100644
index be6c126..0000000
Binary files a/docs/_static/network/3XX-6K-reference-network-design.pdf and /dev/null differ
diff --git a/docs/_static/network/3XX-8K-reference-network-design.pdf b/docs/_static/network/3XX-8K-reference-network-design.pdf
deleted file mode 100644
index 97bccbf..0000000
Binary files a/docs/_static/network/3XX-8K-reference-network-design.pdf and /dev/null differ
diff --git a/docs/index.rst b/docs/index.rst
index d10b619..cd07104 100644
--- a/docs/index.rst
+++ b/docs/index.rst
@@ -6,18 +6,14 @@
 Cluster network performance validation for AMD Instinct accelerators
 ********************************************************************
 
-When running HPC and AI applications in a cluster network environment,
-performance is only as fast as the slowest individual node in the network. To
-achieve optimal performance, each server must be configured for maximum data
-transfer rates and bandwidth utilization based on the available hardware. It is
-crucial to validate both host and device performance in single-node and
-multi-node environments using the appropriate benchmarking tools.
-
-Refer to the relevant networking guides for step-by-step instructions on
-validating network configurations in single-node and multi-node environments.
-These guides cover system settings, device configurations, networking tools, and
-performance tests to ensure AMD Instinct™-powered GPU clusters achieve
-optimal speed and bandwidth during operation.
+When running HPC and AI applications in a cluster network environment, performance is only as fast as the slowest
+individual node in the network. To achieve optimal performance, each server must be configured for maximum data transfer
+rates and bandwidth utilization based on the available hardware. It is crucial to validate both host and device
+performance in single-node and multi-node environments using the appropriate benchmarking tools.
+
+Refer to the relevant networking guides for step-by-step instructions on validating network configurations in
+single-node and multi-node environments. These guides cover system settings, device configurations, networking tools,
+and performance tests to ensure AMD Instinct™-powered GPU clusters achieve optimal speed and bandwidth during operation.
 
 .. grid:: 2
    :gutter: 3
@@ -32,15 +28,13 @@ optimal speed and bandwidth during operation.
 
    .. grid-item-card:: Reference
 
-      * :doc:`Hardware support <reference/hardware-support>`
-      * :doc:`Cluster design <reference/cluster-design>`
+      * :doc:`Hardware support <reference/hardware-support>` 
+
 
 .. note::
 
-   AMD Instinct systems vary in form and configuration, and cluster design
-   adds additional layers of complexity. The guidelines in this
-   documentation are written at a high level for broad applicability across
-   diverse environments. While certain scenarios may include specific hardware
-   examples, your setup will likely differ in terms of GPUs and CPUs per server,
-   firmware versions, and network interconnects. Adjustments might be necessary
-   to align with your particular configuration.
+   AMD Instinct systems vary in form and configuration, and cluster design adds additional layers of complexity. The
+   guidelines in this documentation are written at a high level for broad applicability across diverse environments.
+   While certain scenarios may include specific hardware examples, your setup will likely differ in terms of GPUs and
+   CPUs per server, firmware versions, and network interconnects. Adjustments might be necessary to align with your
+   particular configuration.
diff --git a/docs/reference/cluster-design.rst b/docs/reference/cluster-design.rst
index c92c356..fb00656 100644
--- a/docs/reference/cluster-design.rst
+++ b/docs/reference/cluster-design.rst
@@ -6,18 +6,5 @@
 Cluster architecture and network design
 ************************************************************************************************************************
 
-.. grid:: 2
-   :gutter: 2
-
-   .. grid-item-card:: Reference Cluster Design Architecture MI300X/MI325
-
-      * `128-1028 GPU Reference Cluster Design <../_static/cluster/128-1024-gpu-reference-cluster-design.pdf>`_ 
-      * `1024-8192 GPU Reference Cluster Design <../_static/cluster/1024-8192-gpu-reference-cluster-design.pdf>`_
-
-   .. grid-item-card:: Reference Network Design Architecture 3XX
-
-      * `3XX-2K Reference Network Design <../_static/network/3XX-2K-reference-network-design.pdf>`_
-      * `3XX-4K Reference Network Design <../_static/network/3XX-4K-reference-network-design.pdf>`_
-      * `3XX-6K Reference Network Design <../_static/network/3XX-6K-reference-network-design.pdf>`_
-      * `3XX-8K Reference Network Design <../_static/network/3XX-8K-reference-network-design.pdf>`_
-      * `355-4MW Reference Network Design <../_static/network/355-4MW-PPT-reference-network-design.pdf>`_
+Cluster architecture and network design materials have been collated and moved to a dedicated page. Please see the  `AMD
+Instinct MI3XX Reference Design <https://instinct.docs.amd.com/projects/MI3XX-reference/latest/overview.html>`_.
\ No newline at end of file
diff --git a/docs/sphinx/_toc.yml b/docs/sphinx/_toc.yml
index 51bfc4f..052ef65 100644
--- a/docs/sphinx/_toc.yml
+++ b/docs/sphinx/_toc.yml
@@ -17,7 +17,6 @@ subtrees:
     title: RoCE network configuration  
   - file: how-to/troubleshooting
     title: Network troubleshooting
-
 - caption: Reference
   entries:
   - file: reference/hardware-support
diff --git a/docs/sphinx/_toc.yml.in b/docs/sphinx/_toc.yml.in
index 51bfc4f..052ef65 100644
--- a/docs/sphinx/_toc.yml.in
+++ b/docs/sphinx/_toc.yml.in
@@ -17,7 +17,6 @@ subtrees:
     title: RoCE network configuration  
   - file: how-to/troubleshooting
     title: Network troubleshooting
-
 - caption: Reference
   entries:
   - file: reference/hardware-support
diff --git a/docs/sphinx/requirements.in b/docs/sphinx/requirements.in
index 5d981c8..d964434 100644
--- a/docs/sphinx/requirements.in
+++ b/docs/sphinx/requirements.in
@@ -1 +1 @@
-rocm-docs-core==1.35.0
\ No newline at end of file
+rocm-docs-core==1.35.0
diff --git a/reference-architecture/MI3XX/.readthedocs.yaml b/reference-architecture/MI3XX/.readthedocs.yaml
new file mode 100644
index 0000000..731cdab
--- /dev/null
+++ b/reference-architecture/MI3XX/.readthedocs.yaml
@@ -0,0 +1,31 @@
+# Read the Docs configuration file
+# See https://docs.readthedocs.io/en/stable/config-file/v2.html for details
+
+version: 2
+
+build:
+  os: ubuntu-22.04
+  tools:
+    python: "3.10"
+  jobs:
+    post_checkout:
+      # Monorepo guard: cancel the build unless this change touched reference-architecture/MI3XX/
+      - git fetch origin develop --depth 1 || true
+      - |
+        if [ "$READTHEDOCS_VERSION_TYPE" = "external" ]; then
+          BASE=origin/develop   # PR build: compare against the develop base
+        else
+          BASE=HEAD^            # branch/tag build (latest): compare against previous commit
+        fi
+        if git diff --quiet "$BASE" HEAD -- reference-architecture/MI3XX/; then
+          exit 183
+        fi
+python:
+  install:
+  # Shared across all docsets; dependabot tracks this single file (.github/dependabot.yml)
+  - requirements: docs/sphinx/requirements.txt
+
+sphinx:
+  configuration: reference-architecture/MI3XX/conf.py
+
+formats: [htmlzip]
diff --git a/reference-architecture/MI3XX/_static/css/custom.css b/reference-architecture/MI3XX/_static/css/custom.css
new file mode 100644
index 0000000..e7d3b4a
--- /dev/null
+++ b/reference-architecture/MI3XX/_static/css/custom.css
@@ -0,0 +1,45 @@
+/* Network troubleshooting page */
+
+/* Custom width for rccl errors table */
+table.rccl-errors {
+    table-layout: fixed !important;
+    width: 100% !important;
+}
+table.rccl-errors th:nth-child(1),
+table.rccl-errors td:nth-child(1) {
+    width: 40% !important;
+}
+table.rccl-errors th:nth-child(2),
+table.rccl-errors td:nth-child(2) {
+    width: 60% !important;
+}
+
+/* Custom width for rdma errors table */
+table.rdma-errors {
+    table-layout: fixed !important;
+    width: 100% !important;
+}
+table.rdma-errors th:nth-child(1),
+table.rdma-errors td:nth-child(1) {
+    width: 40% !important;
+}
+table.rdma-errors th:nth-child(2),
+table.rdma-errors td:nth-child(2) {
+    width: 60% !important;
+}
+
+/* Remove white background from topology diagrams in dark mode */
+html[data-theme=dark] .bd-content img {
+    background-color: transparent !important;
+}
+
+/* Restore normal brightness for images in dark mode */
+html[data-theme=dark] .bd-content img:not(.only-dark,.dark-light) {
+    filter: brightness(1) contrast(1) !important;
+}
+
+/* Add neutral grey background for images in light mode */
+html:not([data-theme=dark]) .bd-content img {
+    background-color: #777 !important;
+}
+
diff --git a/reference-architecture/MI3XX/conf.py b/reference-architecture/MI3XX/conf.py
new file mode 100644
index 0000000..2bf69d3
--- /dev/null
+++ b/reference-architecture/MI3XX/conf.py
@@ -0,0 +1,258 @@
+"""Configuration file for the Sphinx documentation builder."""
+import os
+import re
+from pathlib import Path
+
+html_baseurl = os.environ.get("READTHEDOCS_CANONICAL_URL", "instinct.docs.amd.com")
+html_context = {}
+if os.environ.get("READTHEDOCS", "") == "True":
+    html_context["READTHEDOCS"] = True
+
+project = "AMD Instinct Hub"
+html_title = "AMD Instinct MI3XX Reference Design"
+author = "Advanced Micro Devices, Inc."
+copyright = "Copyright (c) 2026 Advanced Micro Devices, Inc. All rights reserved."
+version = "1.0"
+release = version
+setting_all_article_info = False
+
+external_toc_path = "./sphinx/_toc.yml"
+
+external_projects_current_project = "MI3XX-reference"
+
+html_theme = "rocm_docs_theme"
+html_theme_options = {
+    "flavor": "instinct",
+    "show_toc_level": 1,
+    "navbar_align": "content",
+    "link_main_doc": True,
+    "use_download_button": True,
+}
+extensions = ["rocm_docs"]
+
+html_static_path = ['_static']
+
+html_extra_path = ["llms.txt"]
+
+EXCLUDED_DIRS = {
+    "_build",
+    "_templates",
+    "_static",
+    ".git",
+    ".venv",
+}
+
+MARKUP_PREFIXES = (
+    ":::",
+    "```{",
+    "```",
+    ":img-top:",
+    ":class",
+    ":link:",
+    ":link-type:",
+    ":shadow:",
+    ":columns:",
+    ":padding:",
+    ":gutter:",
+    ":open:",
+    ":name:",
+    ":header-rows:",
+    ":alt:",
+    "+++",
+    "-->",
+    "{bdg-",
+)
+
+# Matches lines like "align: center", "alt:", "name: foo" (directive options
+# not starting with a colon, common in MyST figure/table fences)
+_BARE_DIRECTIVE_RE = re.compile(r"^[a-z][a-z_-]*:\s*\S*$")
+
+# Matches MyST/RST anchor labels like "(some-label)="
+_ANCHOR_LABEL_RE = re.compile(r"^\(\w[\w-]*\)=$")
+
+# Matches RST section underlines (e.g. "====", "----", "~~~~")
+_RST_UNDERLINE_RE = re.compile(r"^[=\-~^\"\'#*+]{3,}$")
+
+# Matches RST code block directives (e.g. ".. code-block:: cpp", ".. code:: sh")
+_RST_CODE_BLOCK_RE = re.compile(r"^\.\.\s+(code-block|code|sourcecode)::")
+
+# Matches markdown table separator rows (e.g. "|---|---|", "| :--- | ---: |").
+_MD_TABLE_SEP_RE = re.compile(r"^\|[\s|:\-]+\|$")
+
+# Matches RST directives whose indented body should be discarded (e.g. raw HTML).
+_RST_SKIP_BLOCK_RE = re.compile(r"^\.\.\s+raw::")
+
+# Matches HTML tags (e.g. "<div>", "</p>", "<!--") but NOT RST hyperlink URL
+# continuation lines (e.g. "<https://...>`_").  The negative lookahead excludes
+# URL schemes so that multi-line RST inline hyperlinks are preserved.
+_HTML_TAG_RE = re.compile(r"^<(?!https?://|ftp://|mailto:)[a-zA-Z/!]")
+
+# Matches trailing HTML close tags at the end of a prose line
+# (e.g. "Browse blogs.</p>", "See the guide.</li></ul>").
+_TRAILING_HTML_CLOSE_RE = re.compile(r"(</[a-zA-Z]+>)+\s*$")
+
+MIN_PROSE_LINES = 10
+
+
+def should_skip(path: Path) -> bool:
+    return any(part in EXCLUDED_DIRS for part in path.parts)
+
+
+def is_prose_line(line: str) -> bool:
+    stripped = line.strip()
+    if not stripped:
+        return False
+    if stripped.startswith(MARKUP_PREFIXES):
+        return False
+    # Drop bare directive-option lines (e.g. "align: center", "alt:")
+    if _BARE_DIRECTIVE_RE.match(stripped):
+        return False
+    # Drop MyST/RST anchor labels (e.g. "(some-label)=")
+    if _ANCHOR_LABEL_RE.match(stripped):
+        return False
+    # Drop markdown table separator rows (e.g. "|---|---|", "| :--- | ---: |")
+    if _MD_TABLE_SEP_RE.match(stripped):
+        return False
+    # Drop HTML tags (e.g. "<div>", "</p>") but keep RST hyperlink URL
+    # continuation lines (e.g. "<https://rocm.docs.amd.com/...>`_")
+    if _HTML_TAG_RE.match(stripped):
+        return False
+    # Drop RST directives, comments, hyperlink targets, and substitution definitions
+    if stripped.startswith(".."):
+        return False
+    # Drop YAML frontmatter key-value pairs (e.g. "description lang=en": "text")
+    if stripped.startswith('"') and re.match(r'^"[^"]+"\s*:', stripped):
+        return False
+    # Drop RST field list items (e.g. ":type: int") and extended RST meta
+    # options (e.g. ":description lang=en: text"). Excludes inline roles at line
+    # start (e.g. ":cpp:func:`hipMalloc` returns..." or ":ref:`foo <bar>` describes...")
+    # because those are followed by a backtick, not a space or end-of-line.
+    if re.match(r"^:[A-Za-z][A-Za-z0-9_ =-]*:(\s|$)", stripped):
+        return False
+    # Drop RST section underlines (e.g. "====", "----", "~~~~")
+    if _RST_UNDERLINE_RE.match(stripped):
+        return False
+    return True
+
+
+def generate_combined_markdown(app, exception):
+    if exception:
+        return
+
+    docs_root = Path(app.srcdir)
+    output_file = Path(app.outdir) / "llms-full.txt"
+    base_file = docs_root / "llms.txt"
+
+    combined = []
+
+    if base_file.exists():
+        base_text = base_file.read_text(encoding="utf-8").rstrip().rstrip("-").rstrip()
+        combined.append(base_text)
+    else:
+        combined.append("# AMD Instinct MI3XX Reference Design")
+
+    all_files = sorted(
+        list(docs_root.rglob("*.md")) + list(docs_root.rglob("*.rst"))
+    )
+
+    for doc_file in all_files:
+        if should_skip(doc_file):
+            continue
+
+        if doc_file == base_file:
+            continue
+
+        try:
+            content = doc_file.read_text(encoding="utf-8")
+        except Exception:
+            continue
+
+        lines = content.splitlines()
+        prose_lines = [line for line in lines if is_prose_line(line)]
+
+        if len(prose_lines) < MIN_PROSE_LINES:
+            continue
+
+        relative = doc_file.relative_to(docs_root)
+        in_backtick_fence = False
+        in_rst_code_block = False
+        in_rst_skip_block = False
+        in_html_comment = False  # inside <!-- ... --> block
+        in_html_open_tag = False  # inside a multi-line HTML opening tag
+        kept = []
+        for line in lines:
+            stripped = line.strip()
+            # Backtick fences (MyST/Markdown)
+            if stripped.startswith("```"):
+                in_backtick_fence = not in_backtick_fence
+                kept.append(line)
+                continue
+            if in_backtick_fence:
+                kept.append(line)
+                continue
+            # HTML comment block (<!-- ... -->): discard all content until -->
+            if in_html_comment:
+                if "-->" in stripped:
+                    in_html_comment = False
+                continue
+            # RST skip block (e.g. .. raw::): discard all indented content
+            if in_rst_skip_block:
+                if not stripped or line[0] in (" ", "\t"):
+                    continue
+                in_rst_skip_block = False
+            # RST code block: exit when a non-blank, non-indented line appears
+            if in_rst_code_block:
+                if not stripped or line[0] in (" ", "\t"):
+                    kept.append(line)
+                    continue
+                in_rst_code_block = False
+            # RST raw block: enter and discard both the directive and its body
+            if _RST_SKIP_BLOCK_RE.match(stripped):
+                in_rst_skip_block = True
+                continue
+            # RST code block: enter on directive line (directive itself is dropped)
+            if _RST_CODE_BLOCK_RE.match(stripped):
+                in_rst_code_block = True
+                continue
+            # HTML comment open (<!-- ... -->): discard opener and enter state
+            if stripped.startswith("<!--"):
+                if "-->" not in stripped:
+                    in_html_comment = True
+                continue
+            # Multi-line HTML opening tag: skip continuation lines until >
+            if in_html_open_tag:
+                if ">" in stripped:
+                    in_html_open_tag = False
+                continue
+            # Detect HTML opening tags that wrap across lines (no > on this line)
+            if _HTML_TAG_RE.match(stripped) and ">" not in stripped:
+                in_html_open_tag = True
+                continue
+            if not stripped:
+                kept.append(line)
+            elif is_prose_line(line):
+                # Strip trailing HTML close tags (e.g. "See the guide.</p>")
+                cleaned = _TRAILING_HTML_CLOSE_RE.sub("", line).rstrip()
+                cleaned_stripped = cleaned.strip()
+                if not cleaned_stripped:
+                    # Entire line was HTML close tags — keep original (shouldn't
+                    # normally reach here since _is_prose_line filters HTML).
+                    kept.append(line)
+                elif re.search(r"\w", cleaned_stripped):
+                    # Line has real word content after stripping close tags.
+                    kept.append(cleaned)
+                # else: only punctuation remains (e.g. bare ".") — discard.
+        cleaned = "\n".join(kept)
+
+        combined.append(f"\n\n---\n\n# {relative}\n")
+        combined.append(cleaned.strip())
+
+    output_file.write_text(
+        "\n".join(combined) + "\n",
+        encoding="utf-8",
+    )
+
+
+def setup(app):
+    app.add_css_file('css/custom.css')
+    app.connect("build-finished", generate_combined_markdown)
diff --git a/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/128-1024-gpu-tree-design.png b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/128-1024-gpu-tree-design.png
new file mode 100644
index 0000000..015be97
Binary files /dev/null and b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/128-1024-gpu-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/128-1152-gpu-rail-design.png b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/128-1152-gpu-rail-design.png
new file mode 100644
index 0000000..9af54b4
Binary files /dev/null and b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/128-1152-gpu-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/128-gpu-single-sw-design.png b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/128-gpu-single-sw-design.png
new file mode 100644
index 0000000..2090eca
Binary files /dev/null and b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/128-gpu-single-sw-design.png differ
diff --git a/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/129-256-gpu-tree-design.png b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/129-256-gpu-tree-design.png
new file mode 100644
index 0000000..7ede5d9
Binary files /dev/null and b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/129-256-gpu-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/129-288-gpu-rail-design.png b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/129-288-gpu-rail-design.png
new file mode 100644
index 0000000..a815fa4
Binary files /dev/null and b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/129-288-gpu-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/257-512-gpu-tree-design.png b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/257-512-gpu-tree-design.png
new file mode 100644
index 0000000..8da30a6
Binary files /dev/null and b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/257-512-gpu-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/289-576-gpu-rail-design.png b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/289-576-gpu-rail-design.png
new file mode 100644
index 0000000..6a3a93f
Binary files /dev/null and b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/289-576-gpu-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/513-768-gpu-tree-design.png b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/513-768-gpu-tree-design.png
new file mode 100644
index 0000000..746affe
Binary files /dev/null and b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/513-768-gpu-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/577-864-gpu-rail-design.png b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/577-864-gpu-rail-design.png
new file mode 100644
index 0000000..8da4eee
Binary files /dev/null and b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/577-864-gpu-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/8-128-gpu-single-sw-design.png b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/8-128-gpu-single-sw-design.png
new file mode 100644
index 0000000..eaea5bd
Binary files /dev/null and b/reference-architecture/MI3XX/data/1k-gpu-topology-design-examples/8-128-gpu-single-sw-design.png differ
diff --git a/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048-2304GPU-rail-design.png b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048-2304GPU-rail-design.png
new file mode 100644
index 0000000..127383a
Binary files /dev/null and b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048-2304GPU-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048-2304GPU-rail-scalable-unit.png b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048-2304GPU-rail-scalable-unit.png
new file mode 100644
index 0000000..c5ed7ec
Binary files /dev/null and b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048-2304GPU-rail-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048GPU-tree-design.png b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048GPU-tree-design.png
new file mode 100644
index 0000000..8b9adaa
Binary files /dev/null and b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048GPU-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048GPU-tree-scalable-unit.png b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048GPU-tree-scalable-unit.png
new file mode 100644
index 0000000..5c5f09a
Binary files /dev/null and b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2048GPU-tree-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2072GPU-tree-design.png b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2072GPU-tree-design.png
new file mode 100644
index 0000000..923744e
Binary files /dev/null and b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2072GPU-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2072GPU-tree-scalable-unit.png b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2072GPU-tree-scalable-unit.png
new file mode 100644
index 0000000..58b2571
Binary files /dev/null and b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2072GPU-tree-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2080GPU-rail-design.png b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2080GPU-rail-design.png
new file mode 100644
index 0000000..cc12b45
Binary files /dev/null and b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2080GPU-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2080GPU-rail-scalable-unit.png b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2080GPU-rail-scalable-unit.png
new file mode 100644
index 0000000..39d516f
Binary files /dev/null and b/reference-architecture/MI3XX/data/2k-gpu-topology-design-examples/network-diagram-2080GPU-rail-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096-4608GPU-rail-design.png b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096-4608GPU-rail-design.png
new file mode 100644
index 0000000..6dcb144
Binary files /dev/null and b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096-4608GPU-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096-4608GPU-rail-scalable-unit.png b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096-4608GPU-rail-scalable-unit.png
new file mode 100644
index 0000000..684daad
Binary files /dev/null and b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096-4608GPU-rail-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096GPU-tree-design.png b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096GPU-tree-design.png
new file mode 100644
index 0000000..703816c
Binary files /dev/null and b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096GPU-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096GPU-tree-scalable-unit.png b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096GPU-tree-scalable-unit.png
new file mode 100644
index 0000000..26998e2
Binary files /dev/null and b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4096GPU-tree-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4104GPU-rail-design.png b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4104GPU-rail-design.png
new file mode 100644
index 0000000..64807c6
Binary files /dev/null and b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4104GPU-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4104GPU-rail-scalable-unit.png b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4104GPU-rail-scalable-unit.png
new file mode 100644
index 0000000..286c630
Binary files /dev/null and b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4104GPU-rail-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4144GPU-tree-design.png b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4144GPU-tree-design.png
new file mode 100644
index 0000000..caac2e5
Binary files /dev/null and b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4144GPU-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4144GPU-tree-scalable-unit.png b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4144GPU-tree-scalable-unit.png
new file mode 100644
index 0000000..b16de2d
Binary files /dev/null and b/reference-architecture/MI3XX/data/4k-gpu-toplogy-design-examples/network-diagram-4144GPU-tree-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6016GPU-tree-design.png b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6016GPU-tree-design.png
new file mode 100644
index 0000000..bed51a0
Binary files /dev/null and b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6016GPU-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6016GPU-tree-scalable-unit.png b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6016GPU-tree-scalable-unit.png
new file mode 100644
index 0000000..f387991
Binary files /dev/null and b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6016GPU-tree-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6032GPU-rail-design.png b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6032GPU-rail-design.png
new file mode 100644
index 0000000..f793c77
Binary files /dev/null and b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6032GPU-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6032GPU-rail-scalable-unit.png b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6032GPU-rail-scalable-unit.png
new file mode 100644
index 0000000..fe09ea7
Binary files /dev/null and b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6032GPU-rail-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6048GPU-tree-design.png b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6048GPU-tree-design.png
new file mode 100644
index 0000000..4ac8d9e
Binary files /dev/null and b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6048GPU-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6048GPU-tree-scalable-unit.png b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6048GPU-tree-scalable-unit.png
new file mode 100644
index 0000000..a5114e6
Binary files /dev/null and b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6048GPU-tree-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6144-6912GPU-rail-design.png b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6144-6912GPU-rail-design.png
new file mode 100644
index 0000000..3c16e88
Binary files /dev/null and b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6144-6912GPU-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6144-6912GPU-rail-scalable-unit.png b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6144-6912GPU-rail-scalable-unit.png
new file mode 100644
index 0000000..30730e3
Binary files /dev/null and b/reference-architecture/MI3XX/data/6k-gpu-toplogy-design-examples/network-diagram-6144-6912GPU-rail-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192-9216GPU-rail-design.png b/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192-9216GPU-rail-design.png
new file mode 100644
index 0000000..232f14b
Binary files /dev/null and b/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192-9216GPU-rail-design.png differ
diff --git a/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192-9216GPU-rail-scalable-unit.png b/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192-9216GPU-rail-scalable-unit.png
new file mode 100644
index 0000000..bd22e33
Binary files /dev/null and b/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192-9216GPU-rail-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192GPU-tree-design.png b/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192GPU-tree-design.png
new file mode 100644
index 0000000..f14d402
Binary files /dev/null and b/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192GPU-tree-design.png differ
diff --git a/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192GPU-tree-scalable-unit.png b/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192GPU-tree-scalable-unit.png
new file mode 100644
index 0000000..b98e651
Binary files /dev/null and b/reference-architecture/MI3XX/data/8k-gpu-toplogy-design-examples/network-diagram-8192GPU-tree-scalable-unit.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/2-tier-network-backend-scaling.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/2-tier-network-backend-scaling.png
new file mode 100644
index 0000000..14bd7e9
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/2-tier-network-backend-scaling.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/2-tier-rail-network.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/2-tier-rail-network.png
new file mode 100644
index 0000000..146331e
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/2-tier-rail-network.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/2-tier-tree-network.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/2-tier-tree-network.png
new file mode 100644
index 0000000..a599e49
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/2-tier-tree-network.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-fully-scheduled-rail-network.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-fully-scheduled-rail-network.png
new file mode 100644
index 0000000..d26aa1b
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-fully-scheduled-rail-network.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-hybrid-rail-network.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-hybrid-rail-network.png
new file mode 100644
index 0000000..d26aa1b
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-hybrid-rail-network.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-network-backend-deploy-rail.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-network-backend-deploy-rail.png
new file mode 100644
index 0000000..33ef204
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-network-backend-deploy-rail.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-network-backend-scaling.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-network-backend-scaling.png
new file mode 100644
index 0000000..ea89e7b
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-network-backend-scaling.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-rail-optimized-network.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-rail-optimized-network.png
new file mode 100644
index 0000000..38c18f2
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-rail-optimized-network.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-tree-network.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-tree-network.png
new file mode 100644
index 0000000..79686c6
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/3-tier-tree-network.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/cross-rank-traffic-tree.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/cross-rank-traffic-tree.png
new file mode 100644
index 0000000..25ac70e
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/cross-rank-traffic-tree.png differ
diff --git a/reference-architecture/MI3XX/data/basic-network-topology-design-examples/rail-network-traversals.png b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/rail-network-traversals.png
new file mode 100644
index 0000000..c94e208
Binary files /dev/null and b/reference-architecture/MI3XX/data/basic-network-topology-design-examples/rail-network-traversals.png differ
diff --git a/reference-architecture/MI3XX/index.rst b/reference-architecture/MI3XX/index.rst
new file mode 100644
index 0000000..08f5888
--- /dev/null
+++ b/reference-architecture/MI3XX/index.rst
@@ -0,0 +1,5 @@
+:orphan:
+
+.. raw:: html
+
+   <meta http-equiv="refresh" content="0; url=overview.html">
diff --git a/reference-architecture/MI3XX/legal-information.rst b/reference-architecture/MI3XX/legal-information.rst
new file mode 100644
index 0000000..1d2ebc3
--- /dev/null
+++ b/reference-architecture/MI3XX/legal-information.rst
@@ -0,0 +1,40 @@
+Legal information
+========================================================================================================================
+
+DISCLAIMER
+
+The information contained herein is for informational purposes only, and is subject to change without notice.
+While every precaution has been taken in the preparation of this document, it may contain technical inaccuracies,
+omissions and typographical errors, and AMD is under no obligation to update or otherwise correct this information.
+Advanced Micro Devices, Inc. makes no representations or warranties with respect to the accuracy or completeness of the
+contents of this document, and assumes no liability of any kind, including the implied warranties of noninfringement,
+merchantability or fitness for particular purposes, with respect to the operation or use of AMD hardware, software or
+other products described herein.  No license, including implied or arising by estoppel, to any intellectual property
+rights is granted by this document.  Terms and limitations applicable to the purchase or use of AMD's products are as
+set forth in a signed agreement between the parties or in AMD's Standard Terms and Conditions of Sale. GD-18 
+
+COMPLIANCE WITH LAWS
+
+Customer shall adhere to all applicable export laws and regulations including, without limitation, those
+administered by the U.S. Department of Commerce - Bureau of Industry and Security (U.S. Export Administration
+Regulations 15 CFR 730 et seq.) and those administered by the U.S. Department of State in accordance with the U.S.
+International Traffic in Arms Regulations (ITAR) set forth in Subchapter M, Title 22, Code of Federal Regulations, Parts
+120 through 130 (22 CFR 120-130), as the same may be amended from time to time, and shall not export, re-export, resell,
+transfer, or disclose, directly or indirectly, any Products or technical data, or the direct product of any Products or
+technical data, to any proscribed person, entity, or country, or foreign national thereof, unless properly authorized by
+the U.S. government and/or any other applicable or relevant government or regulatory body, including the export
+authorities of all respective countries. For the avoidance of doubt, Customer shall not use Products in, or re-export
+Products to Belarus, Russia and the Donetsk (DNR) or Luhansk (LNR) regions of Ukraine, regardless of the applicable
+export laws and regulations. Customer shall impose upon its customers terms at least as restrictive as those contained
+in this Clause 14 with respect to any sale, distribution or export of Products. 
+
+© 2025 Advanced Micro Devices, Inc. All
+rights reserved. AMD, the AMD Arrow logo, AMD Instinct, AMD together we advance\_, Infinity Fabric, ROCm, and
+combinations thereof are trademarks of Advanced Micro Devices, Inc. Amazon S3, Arista, Arista OSFP, APC, Broadcom,
+Ciena, Cisco, CloudVision, DataDirect Networks, Dell, DriveNets, EOS, FS.com, Hammerspace, Hewlett-Packard Enterprise, IOS,
+Juniper, JUNOS, Lenovo, Linux, MTP, Netshelter, Nokia, Proliant, Pure Storage, Schneider Electric, SONiC, Super Micro
+Computer Inc, Tomahawk, Ubuntu, Vast Data, Weka, and other product names used in this publication are for identification
+purposes only and may be trademarks of their respective owners. Certain AMD technologies may require third-party
+enablement or activation. Supported features may vary by operating system. Please confirm with the system manufacturer
+for specific features. No technology or product can be completely secure.
+
diff --git a/reference-architecture/MI3XX/overview.rst b/reference-architecture/MI3XX/overview.rst
new file mode 100644
index 0000000..9ba35dd
--- /dev/null
+++ b/reference-architecture/MI3XX/overview.rst
@@ -0,0 +1,599 @@
+.. meta::
+   :description: Reference materials for cluster and associated network builds
+   :keywords: network validation, cluster, cluster design, cluster architecture, cluster network
+
+************************************************************************************************************************
+AMD Instinct MI3XX Reference Design
+************************************************************************************************************************
+
+This document provides a common reference for designing GPU cluster networks using AMD Instinct MI300X, MI325X, MI350X,
+and MI355X series accelerators, supporting up to 8192 GPUs. It covers fundamental cluster design principles, network
+topologies, scalable architectures, and bill of materials for large-scale deployments. Also included are practical
+examples, diagrams, and recommendations for both fat tree and rail network designs, as well as guidance on scaling,
+hardware selection, and best practices for high-performance AI/ML workloads. The audience for this content encompasses
+architects, engineers, and IT professionals.
+
+Common cluster design principles
+========================================================================================================================
+
+Fat tree network topologies
+------------------------------------------------------------------------------------------------------------------------
+
+The canonical fat tree topology is a network concept where a switch's connection to upstream peers has at least parity
+bandwidth with the total aggregate bandwidth of its downstream connections. This causes links between switches to become
+"fatter" as they get closer to the core.
+
+The "fat tree" topology for AI/ML clusters instead refers to how a host is connected to its upstream switches; in this
+case all host NICs terminate on the same switch. It can also be considered 1-rail network. The network itself is
+generally a 3-stage or 5-stage folded Clos network due to the fixed radix of network switches.
+
+Rail network topologies
+------------------------------------------------------------------------------------------------------------------------
+
+Rail networks leverage the same folded Clos network as tree networks, but host connections are instead aggregated onto
+switches based on NIC rank. These shared ranks are referred to as rails and allow the network to provide preferential
+latency for connections which share the same rail. The downside to this design is any traffic which needs to cross
+rails/ranks must traverse either the network spine layer, or Infinity Fabric (PXN). 
+
+Comparison between fat tree and rail networks
+------------------------------------------------------------------------------------------------------------------------
+
+Rail networks can provide better latency for traffic within the same rail, enabling larger single hop ring domains.
+However, traffic that needs to cross rails can experience higher latency, which can be a bottleneck in large clusters
+with high cross-rail traffic.
+
+.. image:: ./data/basic-network-topology-design-examples/rail-network-traversals.png
+   :alt: Example of benefits and limitations of rail network traversals 
+
+Fat tree networks handle cross-rank traffic better, but may have higher latency for traffic that could have been
+contained within a single rail in a rail network.
+
+.. image:: ./data/basic-network-topology-design-examples/cross-rank-traffic-tree.png
+   :alt: Example of benefits and limitations of fat tree network traversals
+
+The choice between the two often depends on the specific workload and communication patterns of the applications being
+run on the cluster.
+
+Basic network topologies
+========================================================================================================================
+
+The following sections describe basic layouts for rail, tree, and hybrid network topologies that can be used as building
+blocks for larger cluster designs. These layouts are not exhaustive, but provide a starting point for understanding the
+trade-offs between different network architectures.
+
+2-tier rail network
+------------------------------------------------------------------------------------------------------------------------
+
+The 2-tier rail network design enables large, scalable unit sizes suitable for large jobs or replica sizes, offering
+efficiency for workloads that utilize ring-based collectives, though it also results in higher infrastructure costs due
+to the need for additional networking hardware.
+
+.. image:: ./data/basic-network-topology-design-examples/2-tier-rail-network.png
+   :alt: 2-tier rail network diagram
+
+2-tier tree network
+------------------------------------------------------------------------------------------------------------------------
+
+The 2-tier tree network design is efficient for small workloads or replicas and can easily scale by adding capacity with
+proper planning. It also has the potential to reduce overall infrastructure costs, while its design helps limit the
+blast radius compared to rail networks.
+
+.. image:: ./data/basic-network-topology-design-examples/2-tier-tree-network.png
+   :alt: 2-tier tree network diagram
+
+3-tier rail TH5/J3 network
+------------------------------------------------------------------------------------------------------------------------
+
+In the 3-tier rail TH5/J3 network design, spine switches are replaced with a two-tier Jericho3-AI/Ramon3 fabric to
+enable a larger maximum cluster size, where deeper buffers and scheduled fabric help alleviate congestion in large
+clusters with only a small latency trade-off. 
+
+3-tier tree TH5/J3 network
+------------------------------------------------------------------------------------------------------------------------
+
+The 3-tier tree TH5/J3 network design provides all the same benefits from switching to a scheduled spine fabric as with
+rail, but retains the primary characteristics of tree networks.
+
+3-tier rail optimized network
+------------------------------------------------------------------------------------------------------------------------
+
+The 3-tier rail optimized network design allows for massive scalable unit sizes and delivers the best ring-based
+collective performance at scale, though this comes with the trade-off of weaker any-to-any communication performance.
+
+.. image:: ./data/basic-network-topology-design-examples/3-tier-rail-optimized-network.png
+   :alt: 3-tier rail optimized network diagram
+
+3-tier tree network
+------------------------------------------------------------------------------------------------------------------------
+
+The 3-tier tree network design allows for massive cluster sizes and delivers excellent any-to-any performance at scale,
+making it well-suited for large deployments that need strong, predictable connectivity. This architecture is
+particularly effective for campus-style environments, where broad distribution and high performance are both required.
+
+.. image:: ./data/basic-network-topology-design-examples/3-tier-tree-network.png
+   :alt: 3-tier tree optimized network diagram
+
+3-tier hybrid rail network
+------------------------------------------------------------------------------------------------------------------------
+
+The 3-tier hybrid rail network design allows for massive cluster sizes with large scalable units, favoring ring-based
+collectives while still maintaining solid any-to-any performance for large jobs. These characteristics also make it
+well-suited for campus-style deployments that balance scalability with broad connectivity requirements.
+
+.. image:: ./data/basic-network-topology-design-examples/3-tier-hybrid-rail-network.png
+   :alt: 3-tier hybrid rail network diagram
+
+3-tier fully scheduled rail network
+------------------------------------------------------------------------------------------------------------------------
+
+The 3-Tier fully scheduled rail network designuses medium-sized scalable units and delivers excellent congestion
+performance thanks to deep buffers and scheduled fabric, though technical limitations restrict the recommended cluster
+size to roughly 32,000 GPUs.
+
+.. image:: ./data/basic-network-topology-design-examples/3-tier-fully-scheduled-rail-network.png
+   :alt: 3-tier fully scheduled rail network diagram
+
+Scaling networks
+========================================================================================================================
+
+As cluster size increases, the network must be scaled to accommodate the additional bandwidth and connectivity
+requirements. For a 2-tier tree network, spine switches do not need to be added until a second scalable unit is deployed
+as all rail/rank traffic occurs at the unit-level. In a 2-tier rail network, spine switches are needed at deployment to
+connect rails at any scalable unit number.
+
+.. image:: ./data/basic-network-topology-design-examples/2-tier-network-backend-scaling.png
+    :alt: Example of network backend scaling for 2-tier network design
+
+In a 3-tier network, a tree design does not require a super spine until a super-scalable unit is deployed.
+
+.. image:: ./data/basic-network-topology-design-examples/3-tier-network-backend-scaling.png
+    :alt: Example of network backend scaling for 3-tier network design
+
+This holds true for hybrid rail as well, where the super spine is only needed at super-scalable unit deployments, but a
+fully scheduled rail network requires a super spine from the initial deployment.
+
+.. image:: ./data/basic-network-topology-design-examples/3-tier-network-backend-deploy-rail.png
+    :alt: Example of network backend scaling for 3-tier network design
+
+Network subscription
+========================================================================================================================
+
+Subscription is the relationship between what is provided by the upstream network and what is required by the downstream
+network in demand side. 
+
+It is typically represented as a ratio: 
+
+.. math::
+
+   Downstream Demand : Upstream Capacity
+
+In a 1:1 subscribed network the downstream capacity is equal to the upstream capacity, while in 1:1.16 subscribed
+network there is .16 more upstream capacity.
+
+This can also be represented as a percentage: 
+
+.. math::
+   
+   Subscription Rate =  \frac{Downstream Demand}{Upstream Capacity} 
+
+An 80% subscription ratio could be referred to as "20% undersubscribed", or a 120% subscription ratio could be referred
+to as "20% oversubscribed".
+
+Hardware and software components
+========================================================================================================================
+
+128 to 1024 GPU generic BOM
+------------------------------------------------------------------------------------------------------------------------
+
+The following table provides a generic bill of materials (BOM) for cluster and network designs ranging from 128 to 1024
+GPUs. The actual components and quantities may vary based on specific design choices, vendor selection, and scalability
+requirements.
+
+**Cluster**
+
++------------------+-------------------------+ 
+| Cluster Size     | 128 to 1024 GPU         |
++==================+=========================+
+| Platforms        | Dell XE9680             |
+|                  | Lenovo SR685a V3        |
+|                  | SMCI AS-8125GS          |
++------------------+-------------------------+
+| OS               | Ubuntu 22.04 (or above) |
++------------------+-------------------------+
+| Linux kernel     | 5.15 - 6.80             |
++------------------+-------------------------+
+| ROCm             | 6.33 (Or above)         |
++------------------+-------------------------+
+
++------------------+-------------------------+
+| Storage Type     |                         |
++==================+=========================+
+| Local storage    | 1.6 TB (or greater)     |
++------------------+-------------------------+
+| Utility storage  | Pure, Vast, RYO         |
++------------------+-------------------------+
+| Bulk storage     | Vast, DDN, WekaIO       |
++------------------+-------------------------+
+| Scratch storage  | Vast, DDN, WekaIO,      |
+|                  | Hammerspace             |
++------------------+-------------------------+
+| Archive/object   | S3 compataible          |
+| storage          |                         |
++------------------+-------------------------+
+
+**Network**
+
++---------------------------+-----------------------------------------------------+
+| Backside Network Topology | 2 Tier Rail Optimized / Fat Tree                    |
++===========================+=====================================================+
+| NIC                       | Pollara 400, BCM957608 (Thor2)                      |
++---------------------------+-----------------------------------------------------+
+|| Switch                   || Arista, Dell, Juniper, Cisco, Nokia                |
+||                          || (TH 4/5, Jericho/Ramon)                            |
++---------------------------+-----------------------------------------------------+
+| Network OS                | SONiC, Junos, EOS, IOS                              |
++---------------------------+-----------------------------------------------------+
+| Subscription ratio        | 1:1.16=16% Undersubscribed (AMD recommended)        |
++---------------------------+-----------------------------------------------------+
+| Optics                    | Vendor ACL/HCL transceivers or direct attach copper |
++---------------------------+-----------------------------------------------------+
+| Fabric                    | RoCEv2 Ethernet                                     |
++---------------------------+-----------------------------------------------------+
+
++-----------------------------------------+----------------------------------------------+
+| Frontside Network Segement              | Adapter Recommended                          |
++=========================================+==============================================+
+| All-in one network                      | Ethernet 100 GbE 2-port QSFP28 adapter       |
++-----------------------------------------+----------------------------------------------+
+| Storage network (optional)              | Ethernet 100 GbE 2-port QSFP28 adapter       |
++-----------------------------------------+----------------------------------------------+
+| Virtualization network (optional)       | Ethernet 100 GbE 2-port QSFP28 adapter       |
++-----------------------------------------+----------------------------------------------+
+| Host in-band                            | Ethernet 10/25GbE 4-Port SFP28 adapter       |
++-----------------------------------------+----------------------------------------------+
+| BMC OOB Mgt                             | 1G Copper                                    |
++-----------------------------------------+----------------------------------------------+
+
+
+1024 to 8192 GPU generic BOM
+------------------------------------------------------------------------------------------------------------------------
+
+The following table provides a generic bill of materials for cluster and network designs ranging from 1024 to 8192 GPUs.
+The actual components and quantities may vary based on specific design choices, vendor selection, and scalability
+requirements.
+
+**Cluster**
+
++------------------+-------------------------+ 
+| Cluster Size     | 1024 to 8192 GPU        |
++==================+=========================+
+| Platforms        | Dell XE9680             |
+|                  | Lenovo SR685a V3        |
+|                  | SMCI AS-8125GS          |
++------------------+-------------------------+
+| OS               | Ubuntu 22.04 (or above) |
++------------------+-------------------------+
+| Linux kernel     | 5.15 - 6.80             |
++------------------+-------------------------+
+| ROCm             | 6.33 (Or above)         |
++------------------+-------------------------+
+
++------------------+-------------------------+
+| Storage Type     |                         |
++==================+=========================+
+| Local storage    | 1.6 TB (or greater)     |
++------------------+-------------------------+
+| Utility storage  | Pure, Vast, RYO         |
++------------------+-------------------------+
+| Bulk storage     | Vast, DDN, WekaIO       |
++------------------+-------------------------+
+| Scratch storage  | Vast, DDN, WekaIO,      |
+|                  | Hammerspace             |
++------------------+-------------------------+
+| Archive/object   | S3 compataible          |
+| storage          |                         |
++------------------+-------------------------+
+
+**Network**
+
++---------------------------+----------------------------------------------+
+| Backside Network Topology | 2 Tier Rail Optimized / Fat Tree             |
++===========================+==============================================+
+| NIC                       | Pollara 400, BCM957608 (Thor2)               |
++---------------------------+----------------------------------------------+
+| Switch                    | Arista, Dell, Juniper, Cisco, Nokia          |
+|                           | (TH 4/5, Scheduled Fabrics)                  |
++---------------------------+----------------------------------------------+
+| Network OS                | SONiC, Junos, EOS, IOS, DriveNets            |
++---------------------------+----------------------------------------------+
+| Subscription ratio        | 1:1.16=16% Undersubscribed (AMD recommended) |
++---------------------------+----------------------------------------------+
+| Optics                    | Vendor ACL/HCL                               |
++---------------------------+----------------------------------------------+
+| Fabric                    | RoCEv2 Ethernet                              |
++---------------------------+----------------------------------------------+
+
++-----------------------------------------+----------------------------------------------+
+| Frontside Network Segement              | Adapter Recommended                          |
++=========================================+==============================================+
+| All-in one network                      | Ethernet 100 GbE 2-port QSFP28 adapter       |
++-----------------------------------------+----------------------------------------------+
+| Storage network (optional)              | Ethernet 100 GbE 2-port QSFP28 adapter       |
++-----------------------------------------+----------------------------------------------+
+| Virtualization network (optional)       | Ethernet 100 GbE 2-port QSFP28 adapter       |
++-----------------------------------------+----------------------------------------------+
+| Host in-band                            | Ethernet 10/25GbE 4-Port SFP28 adapter       |
++-----------------------------------------+----------------------------------------------+
+| BMC OOB Mgt                             | 1G Copper                                    |
++-----------------------------------------+----------------------------------------------+
+
+Power requirements
+========================================================================================================================
+
+MI355X 
+------------------------------------------------------------------------------------------------------------------------
+
+These are design assumptions for a 4MW cluster with 2K MI355X GPUs, including options for 51.2T or scheduled fabrics
+switches (Arista, Dell, Juniper, Cisco, Nokia). These assumptions are based on typical power consumption values for the
+specified hardware components, and actual power usage may vary based on specific workloads, configurations, and
+environmental conditions.
+
+.. note::
+   These are estimates only; Please consult with hardware vendor model data sheets for more accurate power 
+   specifications.
+
++------------------------------------------------+----------------------------------------------------+
+| **System design**                                                                                   |
++------------------------------------------------+----------------------------------------------------+
+| Quantity                                       | 256 MI355X DLC - 2K GPUs                           |
++------------------------------------------------+----------------------------------------------------+
+| Average Power per system                       | ≈ 14kW                                             |
++------------------------------------------------+----------------------------------------------------+
+| 256 Systems                                    | ≈ 3.584 Megawatts                                  |
++------------------------------------------------+----------------------------------------------------+
+| **51.2T switch design**                                                                             |
++------------------------------------------------+----------------------------------------------------+
+| Quantity                                       | ≈ 61 switches - 51.2T switch (Dell, Cisco, Arista) |
++------------------------------------------------+----------------------------------------------------+
+| Estimated typical/load power per switch        | 540w/1125w ≈ 32.94kW/68.63kW                       |
++------------------------------------------------+----------------------------------------------------+
+| **Scheduled fabrics design**                                                                        |
++------------------------------------------------+----------------------------------------------------+
+| Quantity                                       | ≈ 10 x 7720R4-128PE & 64 x 7700R4C                 |
++------------------------------------------------+----------------------------------------------------+
+| Estimated typical/load power 7720R4-128PE      | 1032w/3848w ≈ 10.32kW/38.48kW                      |
++------------------------------------------------+----------------------------------------------------+
+| Estimated typical/load power 7720R4C-38PE      | 593w/1840w ≈ 37.96kW/117.76kW                      |
++------------------------------------------------+----------------------------------------------------+
+| Scheduled fabrics estimated power typical/load | ≈ 48.28kW/156.24kW                                 |
++------------------------------------------------+----------------------------------------------------+
+| **Storage/Management network design**                                                               |
++------------------------------------------------+----------------------------------------------------+
+| Please consult storage and OEM vendors for design and power specifications.                         |
++------------------------------------------------+----------------------------------------------------+
+
+Network design examples
+========================================================================================================================
+
+Designs included are based on either Jericho or Ramon switch types (Arista, Ciena, Nokia) or 51.2T switch types (Arista,
+Cisco, Dell, Juniper). Vendors and switch models vary for port count and features; please consult your desired vendor's
+port count directly to confirm.
+
+The diagrams presented in this section are designed around a scalable unit or POD, which can determine overall network
+end to end latency and AI use cases. Certain ML/AI workloads may require a change of scalable unit size. Please consult
+with AMD Architecture as required.
+
+128 GPU topology design examples 51.2T
+------------------------------------------------------------------------------------------------------------------------
+
+**Single switch design - 8-128 GPU (1-16 nodes)**
+
+.. image:: ./data/1k-gpu-topology-design-examples/8-128-gpu-single-sw-design.png
+   :alt: 8-128 GPU single switch design diagram
+
+256 - 864 GPU topology design examples scheduled fabrics
+------------------------------------------------------------------------------------------------------------------------
+
+**Tree design - 129-256 GPU (17-32 nodes)**
+
+.. image:: ./data/1k-gpu-topology-design-examples/129-256-gpu-tree-design.png
+   :alt: 129-256 GPU tree design diagram
+
+**Rail design - 129-288 GPU (17-36 nodes)**
+
+.. image:: ./data/1k-gpu-topology-design-examples/129-288-gpu-rail-design.png
+   :alt: 129-288 GPU rail design diagram
+
+**Tree design - 257-512 GPU (33-64 nodes)**
+
+.. image:: ./data/1k-gpu-topology-design-examples/257-512-gpu-tree-design.png
+   :alt: 257-512 GPU tree design diagram
+
+**Rail design - 289-576 GPU (37-72 nodes)**
+
+.. image:: ./data/1k-gpu-topology-design-examples/289-576-gpu-rail-design.png
+   :alt: 289-576 GPU rail design diagram
+
+**Tree design - 513-768 GPU (65-96 nodes)**
+
+.. image:: ./data/1k-gpu-topology-design-examples/513-768-gpu-tree-design.png
+   :alt: 513-768 GPU tree design diagram
+
+**Rail design - 577-864 GPU (73-108 nodes)**
+
+.. image:: ./data/1k-gpu-topology-design-examples/577-864-gpu-rail-design.png
+   :alt: 577-864 GPU rail design diagram
+
+1K GPU topology design examples scheduled fabrics
+------------------------------------------------------------------------------------------------------------------------
+
+**Tree design - 128-1024 GPU (16-128 nodes)**
+
+.. image:: ./data/1k-gpu-topology-design-examples/128-1024-gpu-tree-design.png
+   :alt: 128-1024 GPU tree design diagram
+
+**Rail design - 128-1152 GPU (16-144 nodes)**
+
+.. image:: ./data/1k-gpu-topology-design-examples/128-1152-gpu-rail-design.png
+   :alt: 128-1152 GPU rail design diagram
+
+2K GPU topology design examples scheduled fabrics
+------------------------------------------------------------------------------------------------------------------------
+
+**Tree design - 2048 GPU (256 Nodes)**
+
+.. image:: ./data/2k-gpu-topology-design-examples/network-diagram-2048GPU-tree-design.png
+   :alt: Network diagram - 2048 GPU (256 Nodes), Tree design
+
+**Tree scalable unit - 2048 GPU (256 Nodes)**
+
+.. image:: ./data/2k-gpu-topology-design-examples/network-diagram-2048GPU-tree-scalable-unit.png
+   :alt: Network diagram - 2048 GPU (256 Nodes), Tree scalable unit
+
+**Rail design - 2048-2304 GPU (256-288 Nodes)**
+
+.. image:: ./data/2k-gpu-topology-design-examples/network-diagram-2048-2304GPU-rail-design.png
+   :alt: Network diagram - 2048-2304 GPU (256-288 Nodes), Rail design
+
+**Rail scalable unit - 2048-2304 GPU (256-288 Nodes)**
+
+.. image:: ./data/2k-gpu-topology-design-examples/network-diagram-2048-2304GPU-rail-scalable-unit.png
+   :alt: Network diagram - 2048-2304 GPU (256-288 Nodes), Rail scalable unit
+
+2K GPU topology design examples 51.2T
+------------------------------------------------------------------------------------------------------------------------
+
+**Tree design - 2072 GPU (259 Nodes)**
+
+.. image:: ./data/2k-gpu-topology-design-examples/network-diagram-2072GPU-tree-design.png
+   :alt: Network diagram - 2072 GPU (259 Nodes), Tree design
+
+**Tree scalable unit - 2072 GPU (259 Nodes)**
+
+.. image:: ./data/2k-gpu-topology-design-examples/network-diagram-2072GPU-tree-scalable-unit.png
+   :alt: Network diagram - 2072 GPU (259 Nodes), Tree scalable unit
+
+**Rail design - 2080 GPU (260 Nodes)**
+
+.. image:: ./data/2k-gpu-topology-design-examples/network-diagram-2080GPU-rail-design.png
+   :alt: Network diagram - 2080 GPU (260 Nodes), Rail design
+
+**Rail scalable unit - 2080 GPU (260 Nodes)**
+
+.. image:: ./data/2k-gpu-topology-design-examples/network-diagram-2080GPU-rail-scalable-unit.png
+   :alt: Network diagram - 2080 GPU (260 Nodes), Rail scalable unit
+
+4K GPU topology design examples scheduled fabrics
+------------------------------------------------------------------------------------------------------------------------
+
+**Tree design - 4096 GPU (512 Nodes)**
+
+.. image:: ./data/4k-gpu-toplogy-design-examples/network-diagram-4096GPU-tree-design.png
+   :alt: Network diagram - 4096 GPU (512 Nodes), Tree design
+
+**Tree scalable unit - 4096 GPU (512 Nodes)**
+
+.. image:: ./data/4k-gpu-toplogy-design-examples/network-diagram-4096GPU-tree-scalable-unit.png
+   :alt: Network diagram - 4096 GPU (512 Nodes), Tree scalable unit
+
+**Rail design - 4096-4608 GPU (512-576 Nodes)**
+
+.. image:: ./data/4k-gpu-toplogy-design-examples/network-diagram-4096-4608GPU-rail-design.png
+   :alt: Network diagram - 4096-4608 GPU (512-576 Nodes), Rail design
+
+**Rail scalable unit - 4096-4608 GPU (512-576 Nodes)**
+
+.. image:: ./data/4k-gpu-toplogy-design-examples/network-diagram-4096-4608GPU-rail-scalable-unit.png
+   :alt: Network diagram - 4096-4608 GPU (512-576 Nodes), Rail scalable unit
+
+4K GPU topology design examples 51.2T
+------------------------------------------------------------------------------------------------------------------------
+
+**Tree design - 4144 GPU (518 Nodes)**
+
+.. image:: ./data/4k-gpu-toplogy-design-examples/network-diagram-4144GPU-tree-design.png
+   :alt: Network diagram - 4144 GPU (518 Nodes), Tree design
+
+**Tree scalable unit - 4144 GPU (518 Nodes)**
+
+.. image:: ./data/4k-gpu-toplogy-design-examples/network-diagram-4144GPU-tree-scalable-unit.png
+   :alt: Network diagram - 4144 GPU (518 Nodes), Tree scalable unit
+
+**Rail design - 4104 GPU (513 Nodes)**
+
+.. image:: ./data/4k-gpu-toplogy-design-examples/network-diagram-4104GPU-rail-design.png
+   :alt: Network diagram - 4104 GPU (518 Nodes), Rail design
+
+**Rail scalable unit - 4104 GPU (513 Nodes)**
+
+.. image:: ./data/4k-gpu-toplogy-design-examples/network-diagram-4104GPU-rail-scalable-unit.png
+   :alt: Network diagram - 4104 GPU (518 Nodes), Rail scalable unit
+
+6K GPU topology design examples scheduled fabrics
+------------------------------------------------------------------------------------------------------------------------
+
+**Tree design - 6016 GPU (752 Nodes)**
+
+.. image:: ./data/6k-gpu-toplogy-design-examples/network-diagram-6016GPU-tree-design.png
+   :alt: Network diagram - 6016 GPU (752 Nodes), Tree design
+
+**Tree scalable unit - 6016 GPU (752 Nodes)**
+
+.. image:: ./data/6k-gpu-toplogy-design-examples/network-diagram-6016GPU-tree-scalable-unit.png
+   :alt: Network diagram - 6016 GPU (752 Nodes), Tree scalable unit
+
+**Rail design - 6144-6912 GPU (768-864 Nodes)**
+
+.. image:: ./data/6k-gpu-toplogy-design-examples/network-diagram-6144-6912GPU-rail-design.png
+   :alt: Network diagram - 6144-6912 GPU (768-864 Nodes), Rail design
+
+**Rail scalable unit - 6144-6912 GPU (768-864 Nodes)**
+
+.. image:: ./data/6k-gpu-toplogy-design-examples/network-diagram-6144-6912GPU-rail-scalable-unit.png
+   :alt: Network diagram - 6144-6912 GPU (768-864 Nodes), Rail scalable unit
+
+6K GPU topology design examples 51.2T
+------------------------------------------------------------------------------------------------------------------------
+
+**Tree design - 6048 GPU (756 Nodes)**
+
+.. image:: ./data/6k-gpu-toplogy-design-examples/network-diagram-6048GPU-tree-design.png
+   :alt: Network diagram - 6048 GPU (756 Nodes), Tree design
+
+**Tree scalable unit - 6048 GPU (756 Nodes)**
+
+.. image:: ./data/6k-gpu-toplogy-design-examples/network-diagram-6048GPU-tree-scalable-unit.png
+   :alt: Network diagram - 6048 GPU (756 Nodes), Tree scalable unit
+
+**Rail design - 6032 GPU (754 Nodes)**
+
+.. image:: ./data/6k-gpu-toplogy-design-examples/network-diagram-6032GPU-rail-design.png
+   :alt: Network diagram - 6032 GPU (754 Nodes), Rail design
+
+**Rail scalable unit - 6032 GPU (754 Nodes)**
+
+.. image:: ./data/6k-gpu-toplogy-design-examples/network-diagram-6032GPU-rail-scalable-unit.png
+   :alt: Network diagram - 6032 GPU (754 Nodes), Rail scalable unit
+
+8K GPU topology design examples scheduled fabrics
+------------------------------------------------------------------------------------------------------------------------
+
+**Tree design - 8192 GPU (1024 Nodes)**
+
+.. image:: ./data/8k-gpu-toplogy-design-examples/network-diagram-8192GPU-tree-design.png
+   :alt: Network diagram - 8192 GPU (1024 Nodes), Tree design
+
+**Tree scalable unit - 8192 GPU (1024 Nodes)**
+
+.. image:: ./data/8k-gpu-toplogy-design-examples/network-diagram-8192GPU-tree-scalable-unit.png
+   :alt: Network diagram - 8192 GPU (1024 Nodes), Tree scalable unit
+
+**Rail design - 8192-9216 GPU (1024-1152 Nodes)**
+
+.. image:: ./data/8k-gpu-toplogy-design-examples/network-diagram-8192-9216GPU-rail-design.png
+   :alt: Network diagram - 8192-9216 GPU (1024-1152 Nodes), Rail design
+
+**Rail scalable unit - 8192-9216 GPU (1024-1152 Nodes)**
+
+.. image:: ./data/8k-gpu-toplogy-design-examples/network-diagram-8192-9216GPU-rail-scalable-unit.png
+   :alt: Network diagram - 8192-9216 GPU (1024-1152 Nodes), Rail scalable unit
\ No newline at end of file
diff --git a/reference-architecture/MI3XX/sphinx/_toc.yml.in b/reference-architecture/MI3XX/sphinx/_toc.yml.in
new file mode 100644
index 0000000..a915ac8
--- /dev/null
+++ b/reference-architecture/MI3XX/sphinx/_toc.yml.in
@@ -0,0 +1,9 @@
+# mi3xx-reference/sphinx/_toc.yml
+defaults:
+  numbered: False
+  maxdepth: 6
+root: index
+entries:
+  - file: overview
+    title: Overview    
+  - file: legal-information 
\ No newline at end of file