The recently very popular DeepSeek model—many people will surely want to deploy it locally, and those considering upgrading or buying a graphics card will definitely want to know whether the AI performance gap between NVIDIA and AMD is really that large. Recently, NVIDIA officially shared DeepSeek R1 performance benchmark data for the RTX 5090, RTX 4090, and RX 7900 XTX. Even though the RX 7900 XTX is AMD’s most powerful last-generation card, the results show it trails the RTX 4090 by a wide margin, and the RTX 5090 goes without saying—the gap is enormous.
DeepSeek AI performance testing shows the RTX 5090 and RTX 4090 crushing the RX 7900 XTX.
A few days ago, NVIDIA shared an article about An article on DeepSeek running on AI PCs with RTX 50 series graphics cards.Among them, they shared test data for the RTX 5090 and RTX 4090.
NVIDIA stated that in this new type of reasoning model, inference speed is the key, and the NVIDIA GeForce RTX 50 series GPUs feature fifth-generation Tensor cores based on the same Blackwell GPU architecture as data center AI solutions, ensuring DeepSeek achieves maximum inference performance on PC.
Below is NVIDIA’s test data:
- DeepSeek R1 Distill Qwen 7b: Compared to Radeon 7900 XTX, RTX 4090 leads by 46%, RTX 5090 leads by 103%
- DeepSeek R1 Distill Llama 8b: Compared to the Radeon 7900 XTX, the RTX 4090 leads by 47%, and the RTX 5090 leads by 106%.
- DeepSeek R1 Distill Qwen 32b: Compared to the Radeon 7900 XTX, the RTX 4090 is 47% ahead, and the RTX 5090 is 124% ahead.
Overall, when running the DeepSeek-R1 model, the RTX 4090 beats the 7900 XTX by at least 46%, and the RTX 5090 takes the lead by more than 103%. NVIDIA is clearly much faster than AMD, so there’s no doubt about it—NVIDIA is definitely the graphics card to go with.
NVIDIA also mentioned that the RTX AI platform provides the broadest range of AI tools, SDKs, and models, fully leveraging the capabilities of the DeepSeek-R1 model. Users can easily experience it on Llama.cpp, Ollama, LM Studio, AnythingLLM, Jan.AI, GPT4All, and OpenWebUI.
DeepSeek R1 is a brand-new AI model that caused a stir across the AI market the moment it launched, because its performance is on par with the most powerful AI models currently available, yet its computational requirements are 11 times lower. For those interested in local deployment but unsure where to start, you can read our DeepSeek R1 Tutorial。
Data source:NVIDIA
Source: KOCPC Chinese