SCALE: Run CUDA on AMD GPUs

Have you ever wished you could run your CUDA code on AMD GPUs without rewriting everything? Well, your wish has come true!

Let me introduce you to SCALE, the GPGPU toolkit.

The CUDA Conundrum

For years, CUDA has been the go-to for GPU programming, but it's been locked to NVIDIA hardware. This left AMD users out in the cold, forced to use alternatives or maintain separate codebases. Not anymore!

Enter SCALE

SCALE is here change things up. It's not just another cross-platform solution; it's a new design in GPGPU programming. Here's why:

1. No Code Changes Required: Your existing CUDA code? It just works. Even that tricky inline PTX assembly!
2. Seamless Integration: SCALE slips into your workflow. It mimics nvcc so perfectly, your build system won't even notice the difference.
3. True CUDA Compatibility: We're talking the real deal here. SCALE aims for 100% CUDA compatibility.

What's Under the Hood?

SCALE isn't just smoke and mirrors. It's a robust toolkit comprising:

• A cutting-edge compiler that speaks fluent CUDA and AMD
• CUDA runtime and driver APIs reimagined for AMD GPUs
• Clever wrapper libraries that bridge CUDA-X APIs to their ROCm counterparts

But Does It Really Work?

You bet! We've put SCALE through its paces with some heavy hitters in the open-source world:

• NVIDIA Thrust? Check.
• Blender Cycles? Absolutely.
• AMGX, llama-cpp, faiss? Yes, yes, and yes!

And that's just the tip of the iceberg. Our nightly tests ensure SCALE keeps crushing it across a range of projects.

Supported Hardware

Right now, SCALE is rocking it on:

• AMD gfx1030 (Navi 21, RDNA 2.0)
• AMD gfx1100 (Navi 31, RDNA 3.0)

We're also making friends with gfx1010 and gfx1101, with more on the horizon. Vega fans, we haven't forgotten you – gfx900 support is in the works!

Why SCALE Stands Out

While others are reinventing the wheel, SCALE is paving a superhighway. We're not just another GPGPU solution; we're the bridge between CUDA's vast ecosystem and AMD's powerful hardware.

How to get started ?

Getting started is a breeze:

1. Check out our tutorial
2. Explore our examples

Remember, with SCALE, your CUDA code isn't just crossing platforms it's making things seemingly easy to setup.