A modern Rust GPU library that deliberately sheds legacy baggage. Goldy targets only modern GPU APIs (Vulkan 1.4+, DX12, Metal Tier2+) and can therefore be significantly simpler than libraries that must maintain backward compatibility.
use goldy::{Color, DeviceType, Instance, RenderTarget, TaskGraph, TextureFormat};
fn main() -> anyhow::Result<()> {
let instance = Instance::new()?;
let device = instance.create_device(DeviceType::DiscreteGpu)?;
let ctx = device.create_context()?;
let target = RenderTarget::new(&device, 800, 600, TextureFormat::Rgba8Unorm)?;
let mut graph = TaskGraph::new();
let mut pass = graph.render_pass("clear", &target);
pass.clear(Color::CORNFLOWER_BLUE);
pass.finish_recorded();
graph.dispatch(&ctx)?;
let _pixels = target.read_to_cpu()?;
Ok(())
}| Attribute | Description |
|---|---|
| Rust-native | Idiomatic Rust API, not a wrapper around C |
| Modern-only | Vulkan 1.4+, DX12, Metal Tier 2 |
| Slang shaders | Single shader language for all backends |
| Unified | Graphics and compute in one API |
[dependencies]
goldy = "0.1"Goldy uses Slang for shader compilation. The
Rust build.rs downloads (if needed) and embeds the pinned Slang version at
compile time; at runtime Goldy extracts and loads it automatically. Application
developers do not install Slang separately.
- Set
GOLDY_SLANG_PATHonly to override with a custom Slang build slang/download.shis for maintainers bumping the pinned Slang version inslang/manifest.json, not for normal project setup
Release packaging for Python wheels and FFI redistributions is described in PACKAGING.md. See DEBUGGING.md if shader compilation fails at runtime.
Optional Rust vs Slang struct layout checks at shader compile time: set GOLDY_VALIDATE_LAYOUTS=1 and pass LayoutCheck data from #[derive(LayoutCheckable)] into ShaderModule::from_slang_with_options (see DEBUGGING.md and the gradient / checkerboard examples).
Run the interactive examples:
cargo run --example triangle --release # Basic triangle
cargo run --example digital_clock --release # 7-segment clock
cargo run --example plasma --release # Demoscene plasma
cargo run --example mandelbrot --release # Fractal explorer
cargo run --example starfield --release # 3D starfield
cargo run --example particles --release # Rain/snowBy default, Goldy uses DX12 on Windows and Vulkan on Linux. Override with GOLDY_BACKEND:
# Run with Vulkan backend (on Windows)
GOLDY_BACKEND=vulkan cargo run --example triangle --release
# Run with DX12 backend
GOLDY_BACKEND=dx12 cargo run --example triangle --releaseSee all examples.
Goldy is inspired by Sebastian Aaltonen's "No Graphics API" vision of what's possible with modern GPU hardware. By targeting only modern GPUs (2018+), Goldy can:
- Use dynamic rendering (no render pass objects)
- Use bindless descriptors (no descriptor sets)
- Assume coherent caches (simpler synchronization)
- Provide a dramatically simpler API
Goldy is also inspired by:
- Wayland's compositor architecture
- Ralph Levien's "Requiem for piet-gpu-hal"
- Slang's vision for unified shader language
- WGPU
- This paper on GPU abstractions: https://www.kom.tu-darmstadt.de/papers/KCGS17.pdf
Read more in Design Philosophy.
| Platform | Minimum |
|---|---|
| NVIDIA | RTX 2000 / GTX 1600 (2018+) |
| AMD | RDNA 1 / RX 5000 (2019+) |
| Intel | Xe / Alchemist (2022+) |
| Apple | M1 / A14 (2020+) |
Before submitting a PR, run the CI checks locally:
cargo fmt --all -- --check
cargo clippy -- -D warnings
cargo testGoldy is dual-licensed:
- Open source — GNU Lesser General Public License v2.1 or later (LGPL-2.1-or-later). You may use, modify, and distribute Goldy as a dynamically linked library in any project, including proprietary software, under the terms of the LGPL.
- Commercial — A commercial license is available for use cases where LGPL terms are not suitable (static linking, proprietary modifications, embedded distribution without LGPL obligations, etc.). Contact [koubaa at github](permanent email TBD) for terms.
Mohamed Koubaa