Nvidia and AMD trade blows over who is faster on DeepSeek AI benchmarks, so which team is telling the truth?

• AMD’s benchmarks show RX 7900 XTX beating RTX 4090, 4080 Super in AI tests
• Nvidia hit back, claiming RTX 5090 is 2.2x faster than AMD’s GPU
• Benchmarks differ, but AMD’s RX 7900 XTX is far cheaper than Nvidia’s cards

DeepSeek is the new AI darling – for now at least – and Nvidia and AMD have been arguing over which of them runs it fastest.

AMD began the spat by releasing benchmarks showing its RX 7900 XTX high-end graphics card outperforming Nvidia’s RTX 4090 and RTX 4080 Super in DeepSeek R1 tests. Nvidia – inevitably - responded with its own results which show the opposite. Because of course they do.

According to David McAfee on X, AMD’s RDNA3-based GPU was up to 13% faster than the RTX 4090 and 34% ahead of the RTX 4080 Super. The RX 7900 XTX performed best against the RTX 4090 using DeepSeek R1 Distill Qwen 7B, where it led by 13%. AMD also tested Distill Llama 8B and Distill Qwen 14B, with its GPU being 11% and 2% faster, respectively. The RTX 4090 had one advantage, it was 4% ahead in Distill Qwen 32B.

Against the RTX 4080 Super, AMD's GPU showed larger leads. The RX 7900 XTX was 34% faster using DeepSeek R1 Distill Qwen 7B, 27% ahead with Distill Llama 8B, and 22% faster using Distill Qwen 14B. Obviously, we can’t be sure quite how Nvidia’s GPUs were configured for the tests and, as AMD ran them, it’s fair to say that Team Red may not have gone out of its way to optimize things for Team Green's hardware.

Nvidia fights back

As Tom’s Hardware points out, the RX 7900 XTX is not widely used for AI, but its RDNA3 architecture includes AI processing capabilities and AMD has marketed this aspect under the “AI Accelerator” label.

Shortly after AMD’s benchmarks went live, Nvidia countered in a blog post with its own results, also reported by Tom's Hardware, claiming the RTX 5090 is up to 2.2 times faster than the RX 7900 XTX. Using Qwen 32B, Nvidia reported a 124% advantage, while the RTX 4090 was 47% ahead. With Llama 8B, the RTX 5090 was 106% faster, and the RTX 4090 led by 47%.

The back-and-forth inevitably highlights the importance of treating manufacturer-released benchmarks with caution. Different optimizations, driver versions, and testing conditions can result in different outcomes.

It’s also important to note that even if Nvidia’s benchmarks are true, AMD still comes out on top in one very important aspect – price. Its GPU is significantly cheaper than both of Nvidia’s powerhouse offerings.

You might also like

• Check out our list of the best AI tools around today
• A look at the Nvidia GPU that powers DeepSeek's AI global ambition
• Could hardware hack be behind DeepSeek's groundbreaking AI efficiency?