The massive image: Nvidia has fired again at AMD with new benchmark outcomes showcasing the superior efficiency of its newest GPUs working DeepSeek’s AI fashions. This comes after AMD’s latest publication of benchmarks that positioned its Radeon RX 7900 XTX forward of Nvidia’s choices.
Nvidia’s counterattack claims that its new GeForce RTX 5090 GPU outperforms AMD’s flagship by a staggering margin. In line with Group Inexperienced, the RTX 5090 is as much as 2.2 occasions quicker than the RX 7900 XTX when working DeepSeek R1 AI fashions.
The tech big performed intensive benchmarks utilizing three variations of the DeepSeek R1 AI mannequin: Distill Qwen 7b, Llama 8b, and Qwen 32b. When utilizing the Qwen LLM with 32b parameters, Nvidia reviews that the RTX 5090 was 124 p.c quicker than AMD’s contender, whereas the previous-generation RTX 4090 nonetheless managed a 47 p.c lead.
Comparable patterns emerged throughout different checks. With Llama 8b, the RTX 5090 reportedly outpaced the RX 7900 XTX by 106 p.c, whereas the RTX 4090 maintained a 47 p.c benefit. Even within the Qwen 7b take a look at, Nvidia’s newest providing was 103 p.c faster, with the RTX 4090 exhibiting a 46 p.c efficiency edge.
These outcomes starkly distinction with AMD’s earlier benchmarks, which had proven the RX 7900 XTX outperforming NVIDIA’s RTX 4090 and 4080 in most situations, with leads of as much as 113 p.c and 134 p.c, respectively.
Nvidia additionally claimed that its GeForce RTX 50 Sequence GPUs, powered by as much as 3,352 trillion operations per second of AI processing functionality, are uniquely positioned to run DeepSeek’s household of distilled fashions quicker than every other choice within the PC market. It is because DeepSeek’s R1 mannequin household, which Nvidia described as a part of a brand new class of ‘reasoning fashions.
These LLMs are designed to imitate human problem-solving processes by allocating extra computational sources to ‘considering’ and ‘reflecting’ on complicated points. This strategy, referred to as test-time scaling, permits the mannequin to dynamically allocate computing sources throughout inference to motive via issues extra successfully.
Nvidia additionally famous that its RTX 50 Sequence GPUs, that includes devoted fifth-generation Tensor Cores, are constructed on the identical Blackwell GPU structure that drives AI improvements in knowledge facilities. This structure permits RTX to completely speed up DeepSeek fashions, delivering peak inference effectivity on private computer systems.
The corporate additionally touted its RTX AI platform, an ecosystem that opens up DeepSeek-R1 capabilities to over 100 million Nvidia RTX AI PCs worldwide, together with these geared up with the newest GeForce RTX 50 Sequence GPUs.
Nvidia argued that high-performance RTX GPUs guarantee AI capabilities stay accessible, even with out an web connection. This not solely provides low latency but additionally enhances privateness, as customers can keep away from importing delicate supplies or exposing their queries to on-line providers.
