ArchitectAI

7 個月前discuss

[flagged]

7 個月前discuss

I built a lightweight (93KB) CUDA→AMD translation layer using LD_PRELOAD.

It intercepts CUDA API calls at runtime and translates them to HIP/rocBLAS/MIOpen.

No source code needed. No recompilation. Just:

  LD_PRELOAD=./libapex_hip_bridge.so ./your_cuda_app

Currently supports:

- 38 CUDA Runtime functions

- 15+ cuBLAS operations (matrix multiply, etc)

- 8+ cuDNN operations (convolutions, pooling, batch norm)

- PyTorch training and inference

Built in ~10 hours using dlopen/dlsym for dynamic loading. 100% test pass rate.

The goal: break NVIDIA's CUDA vendor lock-in and make AMD GPUs viable for

existing CUDA workloads without months of porting effort.

ArchitectAI

8 個月前discuss

RTX 5080 (sm_120) Throughput Test ============================

Matrix size: 4096x4096 FLOAT32 → 50.90 TFLOPS FLOAT16 → 114.54 TFLOPS BFLOAT16 → 94.76 TFLOPS

Matrix size: 8192x8192 FLOAT32 → 57.98 TFLOPS FLOAT16 → 118.84 TFLOPS BFLOAT16 → 120.16 TFLOPS

=========================== Benchmark completed.

# PyTorch 2.10 for RTX 5080 - Windows 11

## Requirements - Windows 11 - Python 3.10 or 3.11 - NVIDIA GeForce RTX 5080 - Latest NVIDIA drivers (560+)

## Installation

1. Create a virtual environment: ```powershell python -m venv .venv .\.venv\Scripts\Activate.ps1 ```

2. Run the installer: ```powershell .\install.ps1 ```

## Verify Installation ```powershell python -c "import torch; print(torch.cuda.is_available())" ```

## Package Contents - PyTorch 2.10.0a0 (custom build for SM 120 / Blackwell) - CUDA 13.0 runtime libraries - cuDNN support - All required DLL dependencies

## Build Info - Built with CUDA 13.0 - Compute capability: SM 120 (Blackwell) - Build date: [ADD DATE] - Source: PyTorch main branch

## Troubleshooting

If you get DLL errors: 1. Ensure you have the latest NVIDIA drivers 2. Check that Python 3.10 or 3.11 is being used 3. Make sure you're in a clean virtual environment

## License PyTorch is BSD-licensed. See torch/LICENSE for details.

## Verification SHA256: `6202cfa3f4dac89e87bd21b754b3778288849428576e1bfd1dc11de4cfee421d` Verified on: Windows 11 Pro 23H2

ArchitectAI

Submissions

[untitled]

Apex GPU: Run CUDA Apps on AMD GPUs Without Recompilation

I made PyTorch 2.10.0a0 for sm120

I just trained a physics-based earthquake forecasting model on a $1000 GPU

Nvidia GPU Boost: My Stock RTX 5080 Is Consistently Beating Advertised

comments