NVIDIA Cache Mover

Relocate NVIDIA & DirectX shader and driver-installer caches off your C: drive to any NTFS volume — live, update-proof, and without booting into Safe Mode.

---

Why This Exists

NVIDIA drivers (DXCache, GLCache, ComputeCache), Windows Direct3D (D3DSCache), and the NVIDIA App updater (ota-artifacts) hardcode their storage locations onto C: and routinely consume 10 GB to 40+ GB of primary SSD space.

Traditionally, moving DXCache required:

1. Rebooting into Windows Safe Mode (because explorer.exe, browsers, and overlays memory-map .nvph shader header files as soon as you log in).
2. Manually running mklink /J commands in an elevated prompt.
3. Redoing links whenever an NVIDIA driver update wiped and recreated child cache folders like GLCache.

NVIDIA Cache Mover automates the entire workflow in a single ~225 KB native Windows utility — live on your active desktop, without closing running apps, requiring Administrator privileges, or restarting Windows.

---

Key Features

• Zero-Reboot Live NT Lock Release — Inspects and releases active .nvph / .nvbin locks held by user-session processes (explorer.exe, Chromium/Electron GPU processes, game overlays) in real time without terminating any applications.
• Driver-Update Proof Architecture — Redirects parent directories (%LOCALAPPDATA%\NVIDIA and AppData\LocalLow\NVIDIA) via NTFS Directory Junctions and sets __GL_SHADER_DISK_CACHE_PATH (WM_SETTINGCHANGE broadcast), so future driver updates delete and recreate DXCache / GLCache inside your target drive without breaking the junction on C:.
• 5 Complete Cache Modules — Relocates DirectX/OpenGL shader caches, Windows D3DSCache, sandboxed browser/WebView2 GPU caches, CUDA ComputeCache, and NVIDIA App's multi-gigabyte ota-artifacts installer cache.
• One-Click Installer Purge & Safe Restore — Reclaim ~4 GB of leftover downloaded driver installers anytime, or atomically migrate caches back to C: with a single click.
• Native Windows 11 Liquid Glass UI + Headless CLI — Borderless Windows 11 Desktop Acrylic (DWMSBT_TRANSIENTWINDOW) interface alongside full command-line support for scripts and power users.

---

Managed Cache Locations

Module	Default Path on C:	Target Subfolder	Notes
NVIDIA Shader Cache	%LOCALAPPDATA%\NVIDIA	<Target>\Local_NVIDIA	Covers DXCache & GLCache at parent level + sets __GL_SHADER_DISK_CACHE_PATH
DirectX System Cache	%LOCALAPPDATA%\D3DSCache	<Target>\D3DSCache	Windows Direct3D pipeline state cache
Sandboxed Shader Cache	%USERPROFILE%\AppData\LocalLow\NVIDIA	<Target>\LocalLow_NVIDIA	Low-integrity GPU shader cache used by browsers, WebView2 & overlays
CUDA / Compute Cache	%APPDATA%\NVIDIA\ComputeCache	<Target>\ComputeCache	Compute & AI kernel compilation cache
NVIDIA App OTA Cache	%ProgramData%\NVIDIA Corporation\NVIDIA App\UpdateFramework\ota-artifacts	<Target>\ota-artifacts	Downloaded Game Ready / Studio driver packages (~3.5–4.5 GB per update)

---

How the Live Lock Engine Works

When nvwgf2umx.dll initializes DirectX in a desktop application, it holds three distinct kernel references to .nvph cache files:

1. A memory-mapped view (MapViewOfFile) in the process's virtual address space.
2. A Section object handle (CreateFileMappingW) in the process handle table.
3. A File object handle (CreateFileW) in the process handle table.

Because those processes run under your current Windows user token, LiveHandleUnlocker queries NtQueryInformationFile(FileProcessIdsUsingFileInformation) to pinpoint only the exact PIDs referencing the target cache folder, copies all unlocked shader binaries first, unmaps matching .nvph views (NtUnmapViewOfSection), closes matching File and Section handles (DuplicateHandle with DUPLICATE_CLOSE_SOURCE), copies the remaining header files, swaps the folder for an NTFS Directory Junction, and lets running apps continue uninterrupted.

---

Getting Started

Requirements

• OS: Windows 10 or Windows 11 (x64)
• Target Drive: Formatted as NTFS (required for NTFS Directory Junctions)
• Runtime: .NET 9.0 Desktop Runtime (x64) (not needed if built with --self-contained true)

GUI Usage

1. Launch NvidiaCacheMover.exe.
2. Click one of the detected NTFS volume cards (e.g., D:\ or G:\) or click Browse to choose a custom folder.
3. Toggle the cache modules you want to manage and click Relocate & Link Selected Modules.

CLI Usage

# Inspect current junction targets and cache footprint
.\NvidiaCacheMover.exe --status

# Relocate and link all cache modules to a target NTFS folder
.\NvidiaCacheMover.exe --move "D:\NVIDIA_Cache"

# Purge leftover downloaded driver installers from ota-artifacts
.\NvidiaCacheMover.exe --clean-ota

---

Building from Source

# Framework-dependent single-file (~225 KB, requires .NET 9 Desktop Runtime)
dotnet publish -c Release -r win-x64 --self-contained false -p:PublishSingleFile=true -p:DebugType=None -o dist

# Fully standalone single-file (~65 MB, zero runtime dependencies)
dotnet publish -c Release -r win-x64 --self-contained true -p:PublishSingleFile=true -p:IncludeNativeLibrariesForSelfExtract=true -p:EnableCompressionInSingleFile=true -p:DebugType=None -o dist-standalone

---

Safety & FAQ

• Does this use real space on C: after linking?
  No. File Explorer still displays the linked folder at %LOCALAPPDATA%\NVIDIA (with a shortcut arrow) and reports the target's size in Properties, but 100% of the bytes live physically on your target drive and consume 0 bytes on C:.
• Will updating my NVIDIA drivers undo this?
  No. Because the junction is placed on the parent %LOCALAPPDATA%\NVIDIA folder rather than individual child folders, driver updates that clear/recreate DXCache or GLCache perform those operations directly inside your target folder.
• Can deleting a link delete my target files?
  JunctionHelper strictly checks FileAttributes.ReparsePoint before unlinking and never recursively deletes through reparse points. To undo everything cleanly, use the built-in Restore to C: button.

---

Disclaimer

This is an independent open-source system utility and is not affiliated with, endorsed by, or sponsored by NVIDIA Corporation or Microsoft Corporation. NVIDIA, GeForce, and CUDA are trademarks and/or registered trademarks of NVIDIA Corporation.