AMD releases ROCm and PyTorch supported driver for Radeon RX 7900 XTX

[harlan4096]

Gaming RX 7900 XTX Navi 31 support for PyTorch and ROCm on Linux

AMD has released a new driver for Ubuntu with support for consumer Radeon GPU.



PyTorch, a deep learning library developed by Facebook AI Research, stands as an open-source tool embraced by various companies and enjoys extensive support from both CPU and GPU manufacturers. AMD’s initial response to integrating PyTorch support, particularly for consumer Radeon GPUs and Radeon PRO, was somewhat sluggish. However, there is now a notable shift as AMD introduces ROCm 5.7.1.

The latest version of Radeon Software for Linux, specifically Ubuntu 22.04.03 HEW, ushers in PyTorch 2.0.1 support via ROCm (Radeon Open Compute Platform) for two RDNA3-based GPUs: the PRO W7900 and RX 7900 XT. This marks the first official instance of AMD enabling ROCm support for the RX 7900 XTX.
 

We are excited to offer the AI community new support for machine learning development using PyTorch built on the AMD Radeon RX 7900 XTX and Radeon PRO W7900 GPUs and the ROCm open software platform. This is our first RDNA 3 architecture-based implementation, and we are looking forward to partnering with the community.

— Dan Wood, Corp. Vice President, Radeon Product Management at AMD.

Although it hasn’t been explicitly confirmed whether lower SKUs will also receive support, AMD is focusing on the high memory capacities of 24GB (7900XTX) and 48GB (W7900). Additionally, the Radeon 7900 series, based on the Navi 31 GPU, boasts 192 AI Accelerators, which AMD claims delivers up to 2x higher performance compared to the previous generation (Radeon PRO W6800 with Navi 21).



AMD’s approach deserves commendation, as the company extends support for ML-powered applications in consumer-grade products, making affordable workstation builds a reality. The distinction between gaming and professional GPUs is gradually diminishing, with the emphasis on AI-accelerated applications shifting towards open-source software optimized for cost-effective processors. While AMD Radeon Pro and NVIDIA RTX workstations may offer more memory, their cost can be a significant concern for individuals looking to begin their journey in AI learning.

Radeon™ Software for Linux® version 23.20.00.48 for Ubuntu 22.04.3 HWE with ROCm 5.7
 

Radeon™ Software for Linux® 23.20 Highlights

• Introducing PyTorch 2.0.1 support through ROCm on Radeon™ RX 7900XTX and Radeon™ PRO W7900
  
• Beta support for multiple RX 7900XTX or W7900 GPUs  (x2 current maximum)
  

Known Issues

• Running PyTorch with iGPU enabled + Discrete GPU enabled may cause crashes.
  
• Display may remain off after waking from standby while running ROCm workloads. It is recommended to not allow display to turn off during ML workloads (configurable in Ubuntu settings).
  
• Blender exhibits intermittent hangs when using cycles render in HIP mode and running separate ML workload at same time.
  
• Intermittent gpureset errors may be seen with Automatic 1111 webUI with IOMMU enabled.
  

AMD Radeon™ Product Family Compatibility

• AMD Radeon™ RX 7900 XTX
  
• AMD Radeon™ PRO W7900
  
• AMD Radeon™ PRO W6800
  

...

Continue Reading