AMD’s Radeon RX 7900 XTX has set a new benchmark for performance with the DeepSeek R1 AI model, surpassing even NVIDIA’s GeForce RTX 4090 in inference tests.
AMD Ramps Up Support for DeepSeek’s R1 LLM Models, Delivering Impressive Results
Over the past few months, DeepSeek’s latest AI model has taken the tech world by surprise. While many of us have questioned the computing power required to train this model, AMD’s flagship "RDNA 3" Radeon RX 7900 XTX GPU makes it possible for the average user to achieve optimal performance. AMD has released data from DeepSeek’s R1 inference benchmarks, showcasing the RX 7900 XTX outperforming NVIDIA’s offerings across various models.
A recent tweet by David McAfee highlighted the Radeon RX 7900 XTX’s impressive performance with DeepSeek. You can catch the details and learn how to get started with AMD GPUs and Ryzen AI APUs through the link provided in the tweet.
For many users, consumer GPUs have proven to be a great alternative for AI workloads, offering a solid balance of performance and cost compared to mainstream AI accelerators. Moreover, by keeping the models operating locally, users can maintain a level of privacy—an increasingly important consideration given concerns around DeepSeek’s AI models. To aid this, AMD has issued a comprehensive guide detailing how to run DeepSeek R1 on their GPUs. Here’s a quick rundown:
- Check Your Driver: Ensure your system is running the 25.1.1 Optional or higher Adrenalin driver.
- Download LM Studio: Grab version 0.3.8 or above from lmstudio.ai/ryzenai.
- Install the Software: Skip the onboarding screen once installed.
- Find Your Model: Navigate to the ‘discover’ tab.
- Select Your Distill: Choose a DeepSeek R1 distill. For beginners, the Qwen 1.5B is recommended for its speed, though larger distills will provide deeper reasoning abilities.
- Quantization Selection: Ensure "Q4 K M" quantization is selected and initiate the download.
- Set Up for Use: Back on the chat tab, select the downloaded distill and enable the "manually select parameters" option.
- Maximize Offloading: Turn the GPU offload layers slider to maximum.
- Load and Go: Press ‘model load’ and start interacting with the reasoning model on your local hardware.
If you hit any snags following these steps, AMD has a step-by-step tutorial available on YouTube. This will guide you through the process, ensuring you can run DeepSeek’s LLMs on your own AMD devices securely. With new GPUs on the horizon from both NVIDIA and AMD, we’re anticipating a tremendous boost in inference capabilities as these chips come equipped with dedicated AI engines designed to handle such demanding tasks.
