AMD ROCm 7.1 vs. RADV Vulkan For Llama.cpp With The Radeon AI PRO R9700
Interestingly when testing a small Granite 3.0 3B model, the Vulkan back-end of Llama.cpp was delivering significantly faster performance on RADV compared to using ROCm 7.1 on this Radeon AI PRO R9700 workstation.
AMD ROCm 7.1 was largely faster than Vulkan with RADV driver for most of the prompt processing benchmarks but RADV+Vulkan remained dominant in text generation performance. There were some exceptions where the RADV Vulkan driver performance remained faster than ROCm for prompt processing. In any case it's exciting to see the good performance overall still of using Llama.cpp with Vulkan as an alternative to vendor-specific AI/compute interfaces.
That's where things stand today with this brief testing while some benchmarks using larger LLMs and the like are underway.
If you enjoyed this article consider joining Phoronix Premium to view this site ad-free, multi-page articles on a single page, and other benefits. PayPal or Stripe tips are also graciously accepted. Thanks for your support.
