AMD RX 7900 XT now supported by way of ROCm 5.7
3 weeks in the past, AMD introduced that it might beef up the primary RDNA3 GPUs via its ROCm platform for PyTorch. Lately, new gaming GPU is being added to this checklist.
AMD is keen to place some cash and energy into supporting system studying building on client {hardware}. AMDs engineers are actively running on including further RDNA3 GPUs to be to be had to boost up PyTorch. This open-source and well-liked framework can be utilized to coach new neural networks. It takes benefit of tensor cores, or on the subject of AMD, their AI Accelerators.
AMD is selling its newest ROCm 5.7 platform, AMD’s personal open-source initiative and device stack that permits GPU computation for those synthetic intelligence workloads on Ubuntu 22.04.3 working gadget. The ROCm supplies helps for various system studying fashions corresponding to HIP, OpenMP or OpenCL, but additionally provides integration of neural studying frameworks like PyTorch or TensorFlow.
ROCm 5.7 replace, Supply: AMD
The newest information is that the {hardware} beef up is extending to Radeon RX 7900 XT which options 20 GB of reminiscence and 168 AI Accelerators. It’s in line with the similar Navi 31 GPU because the RX 7900 XT, however with fewer cores enabled. It’s nonetheless a high-end GPU, however at an much more inexpensive worth of round $749 (associate hyperlink), down from its unique MSRP of $899 for rather a while now.
“We’re interested by this newest addition to our portfolio. Together with ROCm, those high-end GPUs make AI extra obtainable each from a device and {hardware} point of view, so builders can make a choice the answer that most closely fits their wishes”,
— Erik Hultgren, Tool Product Supervisor at AMD.
AMD ROCm 5.7 already helps Radeon RX 7900 XTX and Radeon PRO W7900, provided with 24GB and 48GB reminiscence capability respectively. So far, AMD made no feedback on doable beef up for different GPUs corresponding to Navi 32 primarily based RX 7800 XT 16GB or Navi 31 primarily based RX 7900 GRE 20GB. One would additionally wish to see AMD including RX 6800/6900 collection which function 16GB reminiscence as smartly.
Whilst it’s nice information AMD is including extra GPUs to ROCm, it nonetheless feels underwhelming in comparison to CUDA beef up on NVIDIA playing cards, which is able to almost be taken without any consideration throughout maximum GPUs, notes Phoronix.
Supply: AMD
Allow 48h for review and removal.