NVIDIA’s GeForce RTX 5090 Outperforms in DeepSeek’s R1 AI Model Inference, Vastly Exceeding AMD’s RX 7900 XTX

NVIDIA’s latest GeForce RTX 5090 is making waves with its impressive performance on the DeepSeek R1, significantly outpacing AMD’s RX 7900 XTX. This leap is largely thanks to NVIDIA’s innovative fifth-generation Tensor Cores, which have set a new standard in graphical processing capabilities.

Effortless Access to DeepSeek’s Cutting-Edge Models with NVIDIA’s RTX Series

It turns out that consumer-grade GPUs might just be the key to effectively running high-end large language models (LLMs) right on your personal computer. Both NVIDIA and AMD are vying to provide the optimal environment for such advanced computations. Not long ago, AMD showcased the might of its RDNA 3 flagship GPU with the DeepSeek R1 LLM model. In response, NVIDIA, often referred to as Team Green, has stepped up, revealing inference benchmarks on their latest RTX Blackwell GPUs. The outcome? The GeForce RTX 5090 clearly takes the lead.

When tested across different DeepSeek R1 models, the RTX 5090 consistently outperformed AMD’s Radeon RX 7900 XTX and even NVIDIA’s previous generation GPUs. It managed to process up to 200 tokens per second in both Distill Qwen 7b and Distill Llama 8b models—almost double what AMD’s RX 7900 XTX could handle. These metrics clearly indicate that NVIDIA’s GPUs are setting the bar high for AI performance, promising more frequent integration of cutting-edge AI on consumer PCs with comprehensive "RTX on AI" support.

For those interested in running DeepSeek R1 on NVIDIA’s RTX GPUs, the company has provided a straightforward blog post explaining the process. Surprisingly, using DeepSeek R1 is as simple as utilizing any online chatbot. Here’s a quick guide:

NVIDIA has taken steps to enable developers to securely explore and build tailored agents with the powerful 671-billion-parameter DeepSeek-R1 model. This is now available as a preview microservice on build.nvidia.com, offering up to 3,872 tokens per second on a single NVIDIA HGX H200 system. Developers can experiment with the API, which will soon be downloadable as part of NVIDIA AI Enterprise’s NIM microservice.

The DeepSeek-R1 NIM microservice streamlines deployment with compatibility with industry-standard APIs, allowing enterprises to optimize data security and privacy by running the service on their chosen accelerated computing framework.

With the help of NVIDIA’s NIM, both developers and tech enthusiasts can effortlessly test the AI model locally, ensuring not only data protection but also potentially enhanced performance, given your hardware is up to par. This latest development from NVIDIA underscores their commitment to pushing the boundaries of what’s possible in AI and consumer computing.