TL;DR: AMD has revealed benchmarks that seem to counsel its Radeon RX 7900 XTX graphics card is considerably sooner than Nvidia’s RTX 4090 and RTX 4080 Tremendous relating to DeepSeek R1 benchmarks. Crew Purple additionally launched directions for DIY lovers to run the R1 on Ryzen AI CPUs and Radeon GPUs.
In keeping with an X submit by David McAfee, the VP and GM of Ryzen CPU and Radeon graphics at AMD, the corporate examined the 7900 XTX towards the RTX 4090 and RTX 4080 Tremendous utilizing DeepSeek R1, and the RDNA 3 GPU was between 13 p.c and 34 p.c sooner than Nvidia {hardware} relying on the LLM and variety of parameters.
DeepSeek performing very properly on @AMDRadeon 7900 XTX. Discover ways to run on Radeon GPUs and Ryzen AI APUs right here: https://t.co/FVLDLJ18Ov pic.twitter.com/5OKEkyJjh3
– David McAfee (@McAfeeDavid_AMD) January 29, 2025
With seven billion parameters, the 7900 XTX trumped the RTX 4090 by 13 p.c in Distill Qwen, whereas it was 11 p.c sooner in Distill Llama with eight billion parameters. With 14 billion parameters, the AMD card beat the Ada Lovelace flagship by two p.c in Distill Qwen. Nonetheless, when the variety of parameters went as much as 32 billion, the RTX 4090 had a 4 p.c benefit over the RDNA 3 flagship in Distill Qwen.
AMD additionally in contrast the 7900 XTX to the RTX 4080 Tremendous, exhibiting its card was sooner in each take a look at. The AMD product outperformed the Nvidia GPU by 34 p.c utilizing DeepSeek R1 Distill Qwen with seven billion parameters and 27 p.c utilizing Distill Llama with eight billion parameters. With the variety of parameters elevated to 14 billion, the cardboard from Crew Purple was nonetheless 22 p.c sooner utilizing Distill Qwen.
Along with the benchmarks, AMD additionally revealed detailed directions on find out how to run DeepSeek regionally in your PC. In keeping with the corporate, all its RDNA 3 desktop GPUs are suitable with a number of LLMs utilizing R1. Most of those fashions may run on choose Ryzen CPUs with XDNA NPUs.
AMD additionally launched an inventory of the utmost supported LLM parameters and supplied directions on how LM Studio’s one-click installer might be tuned for its {hardware}. In keeping with the corporate, the Radeon RX 7900 XTX helps as much as 32 billion parameters, whereas the entry-level RX 7600, with its 8GB of VRAM, helps as much as eight billion parameters in Distill Llama.
