AMD’s Radeon RX 7900 XTX Outperforms NVIDIA’s GeForce RTX 4090 in DeepSeek’s AI Inference Tests: A Guide to Running R1 Locally on AMD Systems

AMD’s Radeon RX 7900 XTX is outshining NVIDIA’s GeForce RTX 4090 when it comes to running the DeepSeek R1 AI model, as demonstrated by impressive inference benchmarks.

### AMD’s Swift Rollout for DeepSeek’s R1 Model Support Delivers Stunning Performance

DeepSeek’s latest AI model has stirred up considerable excitement in the tech world. While curiosity about the kind of computing power needed to train such a model is rampant, the good news is that everyday users can get top-notch performance with AMD’s Radeon RX 7900 XTX GPU, featuring the “RDNA 3” architecture. AMD recently shared comparison benchmarks, illustrating that their flagship RX 7000 series GPU outpaces NVIDIA’s in performance across various models.

— January 29, 2025

Consumer GPUs have long been appealing for AI tasks, mainly because they offer significant performance per dollar when compared to standard AI accelerators. Running AI models locally not only provides excellent performance but also ensures privacy—a significant concern related to DeepSeek’s AI models. AMD has thoughtfully released a comprehensive guide for setting up DeepSeek R1 distillations on their GPUs. Here’s a quick rundown of the steps:

1. Ensure your system is running the 25.1.1 Optional or a later version of the Adrenalin driver.
2. Download LM Studio 0.3.8 or later from lmstudio.ai/ryzenai.
3. Install LM Studio and skip the onboarding screen.
4. Access the ‘discover’ tab.
5. Select your preferred DeepSeek R1 Distill model. Smaller ones like Qwen 1.5B offer incredibly fast performance and make a great starting point, though larger models deliver enhanced reasoning capabilities.
6. On the right, ensure “Q4 K M” quantization is chosen, then hit “Download”.
7. Return to the chat tab, select the DeepSeek R1 distill model from the drop-down menu, and check “manually select parameters”.
8. Move the GPU offload layers slider to its maximum setting.
9. Click on “model load”.
10. Interact with a reasoning model running entirely on your local AMD hardware!

If you’re finding these steps challenging, AMD has also produced a detailed tutorial video on YouTube, which breaks down each step. By setting up DeepSeek’s LLMs on your local AMD system, you’ll have peace of mind that your data remains secure. As NVIDIA and AMD continue to develop their next-generation GPUs, we can anticipate a significant boost in inferencing power, thanks to the integration of dedicated AI engines that will enhance such tasks.