NVIDIA’s GeForce RTX 5090 has demonstrated significantly superior inference performance compared to AMD’s Radeon RX 7900 XTX on DeepSeek’s R1 AI models. This dominance is attributed to the RTX 5090’s new fifth-generation Tensor Cores, highlighting NVIDIA’s strength in the AI acceleration space.
While AMD recently showcased the capabilities of its RDNA 3 flagship GPU on the DeepSeek R1 LLM, NVIDIA has responded with benchmarks showcasing its RTX Blackwell GPUs, revealing a clear performance advantage for the RTX 5090. Across various DeepSeek R1 models, the RTX 5090 outperforms both the RX 7900 XTX and its previous-generation NVIDIA counterparts. The RTX 5090 achieved up to 200 tokens per second in Distill Qwen 7b and Distill Llama 8b, nearly double the performance of AMD’s RX 7900 XTX. This stark difference underscores NVIDIA’s dominance in AI performance and suggests a future where edge AI on consumer PCs becomes increasingly prevalent, especially with the growing “RTX on AI” ecosystem.
NVIDIA has simplified access to DeepSeek R1 on RTX GPUs. A dedicated blog post provides guidance, making the process as straightforward as using an online chatbot. Furthermore, NVIDIA’s NIM microservice allows developers to experiment with the 671-billion-parameter DeepSeek-R1 model. Available as a preview on build.nvidia.com, the DeepSeek-R1 NIM microservice can deliver up to 3,872 tokens per second on a single NVIDIA HGX H200 system. An API is also planned for release as a downloadable NIM microservice within the NVIDIA AI Enterprise software platform.
NVIDIA’s NIM simplifies deployments with industry-standard API support. Enterprises can prioritize security and data privacy by running the NIM microservice on their preferred accelerated computing infrastructure. This accessibility empowers developers and enthusiasts to experiment with the AI model locally, enhancing data security and potentially improving performance based on local hardware capabilities. This local execution capability is a significant advantage, allowing for faster iteration and more secure handling of sensitive data. The combination of superior hardware performance and simplified deployment through NIM positions NVIDIA as a leader in enabling local AI processing.