AMD releases ROCm 5.7to enable PyTorch support for RX 7900 XTX and PRO W7900 graphics cards

CTOnews.com, October 22, ROCm is an open source software platform that allows researchers to leverage the potential of the AMD Instinct Accelerator to promote cross-platform high-performance computing and AI innovation.

AMD released the ROCm 5.7.1 driver for Ubuntu Linux this week, bringing support for PyTorch 2.0.1 for RDNA 3-based Radeon Pro W7900 and Radeon RX 7900 XTX graphics cards.

This move enables developers to use GPU, the consumer flagship of AMD, to develop artificial intelligence applications using the PyTorch library, thus avoiding over-reliance on Nvidia.

According to reports, the driver is specifically optimized for PyTorch 2.0.1 and adapts to the functions of AMD RDNA3-based GPU.

However, the ROCm 5.7.1 driver currently supports only two flagship graphics cards: the Radeon RX 7900 XTX 24GB and the Radeon Pro W7900 48GB, because AI applications such as the large language Model (LLM) tend to require large amounts of video memory, and AMD wants to sell as much high-end hardware as possible.

AMD points out that Navi 31 GPU has 192 built-in AI accelerators, which doubles the 96 implementations of Navi 21 and is suitable for artificial intelligence applications. Meanwhile, the AMD Radeon RX 7900 XTX 24 GB is priced at $999 (CTOnews.com notes: the initial price of Bank of China is 7999 yuan, which is generally lower than that of professional AI and HPC GPU).

The AMD Radeon RX 7900 XTX has a 96CU6144 stream processor, GPU frequency 1900-2500MHz, and single-precision performance up to 61TFLOPS. The video memory is 24GB 20Gbps GDDR6, and the bandwidth can reach 960GB / s. Coupled with 96MB unlimited cache, the equivalent bandwidth can reach 3.5TB / s.