Radeon RX 7900 XTX surpasses RTX 4080 and 4090 in versions of DeepSeek R1, according to AMD

One of the advantages of DeepSek’s R1 technology is the ability to use “distilled” versions of the locally. Taking advantage of this popularity, AMD has released a manual on how to run the model on systems with its Radeon GPUS, especially from the RX 7000 series.

To publicize the performance of its plates, the AMD manual has been accompanied by performance results of deepseek R1 versions running in a RX RX 7900 XTX. According to the manufacturer’s tests, its RDNA 3 GPU can overcome the RTX 4080 and even RTX 4090 in various versions of Chinese AI.

In DeepSek R1 Distill Qwen 7B and 14B, as well as in R1 Distill Llama 8B, the Radeon RX 7900 XTX has overcome its Nvidia rivals. However, we can see that performance falls as the model advances. In DeepSek R1 Distill Qwen 32b A RTX 4090 from 24GB surpasses its competitor Radeon.

Users of AMD video cards interested in testing the model will need LM Studio 0.3.8 or more recent, in addition to Radeon Software Adrenalin 25.1.1 Beta or more recent. Step by step is available at this company link.

Related News:

AMD news to DeepSeek bring information about Strix Halo

AMD takes advantage of the manual to also promote its processors from the Strix Halo series, commercially known as Ryzen Ai Max 300. The company says its new more advanced Apuss with “Avantjada” GPU should account for DeepSeek R1 Distill Lamma 70B and even R1 Distill Qwen 32b.

Accidentally or not, the disclosure is accompanied by information that Strix Halo processors support 32GB to 128GB LPCAMM2 memory settings. That is, we will not see 16GB notebooks with these more expensive processors.

Finally, AMD points out that Ryzen AI HX 370 and 365 processors can run DeepSeek R1 Distll Qwen 14B, while Ryzen 8040 and Ryzen 7040 can hold R1 Distill Llama 14b.

Via: TechPowerUp