In general, Nvidia has always been very ahead of AMD as far as computing for artificial intelligence is concerned, but he says AMDthe shots have changed when it comes to the new fashion, Deepseek: Lisa’s those claim that their current flagship graphics card, the Radeon RX 7900 XTXsurpasses the RTX 4090 of Nvidia in the first benchmarks.
If artificial intelligence has been in everyone’s mouth for a few months, Depseek is the hot topic of the moment, and proof of this is that product manufacturers are getting most to measure how good or bad their products with it work. And in AMD they have had a pleasant surprise when they verify that their range of the range exceeds the rap of Nvidia range … of the previous generation, because these data have arrived late and the RTX 4090 has ceased to be the graphics more powerful than green.
AMD measures its performance with Depseek
AMD has published some comparative tests of its flagship flake, the RX 7900 XTX, compared to the NVIDIA RTX 4090 and 4080 Super of previous generation using the Deepseek AI model. As published by David Mcaffee in X, the RDNA 3 -based GPU exceeded RTX 4090 up to 13%, while RTX 4080 Super stays a little further back as is normal, and the difference is up to 34% .
Deepseek Performing Vry Well on @amdradeon 7900 XTX. Learn How to Run On Radeon Gpus and Ryzen Ai Apus Here: https://t.co/fvldlj18Ov https://t.co/5Okekyjjh3
January 29, 2025 • 16:02
AMD has tested the three graphs with several LLM and various parameters using the Depseek R1 algorithm. The RX 7900 XTX won its greatest victory against the RTX 4090 using Deepseek R1 Distill Qwen 7B, where it exceeded the NVIDIA GPU in 13%. It was also 11% faster using distill calls 8b and 2% better in distill Qwen 14b. It is not that the difference is a barbarity, but it is a small victory for AMD, which has not stopped seeing that Nvidia was distanced in all this of AI without being able to do anything.
Of course, all these data should be taken with tweezers because we cannot be sure of how they configured the NVIDIA GPUs to make the comparison, and of course they were used by AMD so you will know (the same now Nvidia responds with their own tests and denies all this data). In addition, not all artificial intelligence workloads take advantage of the computational performance of a GPU.
On the other hand, it is not usual (in fact, we believe that it is the first time they do it in a “official” way) that an AMD graph is used for tests with AI, although what these data do demonstrate is that, of course, of course , they are valid for this. The RDNA 3 architecture on which RX 7900 XTX is based is capable of performing matrix operations and admits BF16 and Int8 without problems. Of course, now AMD has officially added terminology «Ai Accelerator» A RDNA 3 to denote your skill in the processing of workloads for AI: RX 7900 XTX without going any further incorporates 192 hardware centers for it.
Recently, AMD published a tutorial on how you can get Depseek R1 to run in consumer hardware (processors and commercial graphs, which anyone can buy) of AMD, including the graph that concerns us in this article. Deepseek R1 is a new AI model that offers a performance comparable to that of the Avaguardia AI, but with a fraction of its computational cost, which is why it has become famous immediately.
Deepseek R1 uses a series of hardware -based optimizations so that its AI model works up to 11 times faster than competition, including the use of NVIDIA PTX programming language. So be careful, as we have said before, maybe Nvidia is not late to go out to the arena with her own tests to deny all this that AMD is saying.
