Shulou(Shulou.com)11/24 Report--

CTOnews.com Sept. 3, Triton is an open source programming language similar to Python, which enables researchers with no CUDA experience to write efficient GPU code (which can be understood as a simplified version of CUDA), and claims that rookies can also write code comparable to professionals, that is, to allow users to achieve the highest hardware performance with relatively little effort, but Triton initially only supports Nvidia GPU.

OpenAI claims that Triton can achieve the same performance as cuBLAS in FP16 matrix multiplication with only 25 lines of code.

As we can see from Github, OpenAI has begun to incorporate AMD ROCm-related branch code in the latest version of Triton, which reveals a lot. In other words, the latest Triton backend has adapted to the AMD platform, which is of great significance.

Officially, they have passed most of the unit tests on test_core.py, but skipped some tests for a variety of reasons.

CTOnews.com noted that OpenAI also announced that a Triton developer conference would be held from 10:00 to 4 p.m. on September 20 at Microsoft's Silicon Valley campus in Mountain View, Calif., and that the schedule included "introducing Triton to AMD GPU" and "Intel XPU for Triton", and that Triton would soon break away from the history of Nvidia CUDA monopoly.

It is worth mentioning that Triton is open source, and other hardware accelerators can be directly integrated into Triton compared to closed-source CUDA, greatly reducing the time it takes to build AI compiler stacks for new hardware.

In the previous release of PyTorch 2.0, TorchInductor introduced OpenAI Triton support to automatically generate fast code for multiple accelerators and backends, and to write the underlying hardware code with Python instead of CUDA programming. In other words, Triton is already a key component of the PyTorch 2.0 back-end compiler.

In fact, previously, AMD ROCm mainly used Hipify tools to achieve CUDA compatibility, but as AMD begins to provide ROCm support for RDNA 3 consumer graphics cards, it is expected that more platforms will choose to adapt AMD hardware.

Related readings:

"AMD will provide ROCm support for RDNA 3 consumer graphics cards, starting with Radeon RX 7900 XTX and Pro W7900"

"AMD officially launches HIP SDK: expanding the ROCm ecosystem to support running Nvidia CUDA applications"

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.