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

AMD’s Radeon RX 7900 XTX is impressively outperforming NVIDIA’s GeForce RTX 4090 in inference benchmarks, specifically with the DeepSeek R1 AI model.

—

AMD has swiftly taken the lead in supporting DeepSeek’s R1 LLM Models, showcasing remarkable performance enhancements. DeepSeek’s latest AI model has truly shaken up the industry. For many people, there’s been curiosity about the computing power used for training this model. Interestingly, folks with a keen interest in tech can now harness ample performance using AMD’s “RDNA 3” Radeon RX 7900 XTX GPU. AMD, often referred to as Team Red, has unveiled inference benchmarks of DeepSeek’s R1, highlighting its flagship RX 7000 series GPU outpacing NVIDIA’s equivalent across multiple models.

For everyday users, leveraging consumer GPUs for AI tasks has proven to be quite beneficial—mainly due to their cost-efficiency compared to mainstream AI accelerators. Running models locally also gives users peace of mind, addressing privacy concerns associated with DeepSeek’s AI models. Fortunately, AMD has rolled out a comprehensive guide for running DeepSeek R1 distillations on their GPUs, and here’s how you can get started:

1. First, ensure your system is running on the 25.1.1 Optional or a later version of the Adrenalin driver.
2. Download LM Studio version 0.3.8 or higher from lmstudio.ai/ryzenai.
3. Install LM Studio, and bypass the onboarding screen when prompted.
4. Navigate to the discover tab.
5. Select your DeepSeek R1 Distill. Starting with smaller distills like the Qwen 1.5B is advised due to their impressive speed, although larger distills can provide better reasoning capabilities. Each option is highly capable in its own right.
6. On the right panel, ensure the “Q4 K M” quantization option is chosen, then hit “Download.”
7. After downloading, return to the chat tab, select the DeepSeek R1 distill from the drop-down, and make sure “manually select parameters” is ticked.
8. Slide the GPU offload layers all the way to the max.
9. Click on model load to get started.
10. You can now engage with a reasoning model that’s fully operational on your local AMD setup!

If these steps don’t work for you, don’t worry. AMD has also uploaded a step-by-step tutorial on YouTube, which explains each stage in detail. Check it out to confidently run DeepSeek’s LLMs on your AMD rig while keeping your data secure. As new GPUs from NVIDIA and AMD hit the market, we anticipate a considerable boost in inference power, given the presence of dedicated AI engines designed for such tasks.