Best local AI models for AMD R9 M385X

The largest models that fit fully

The 30 largest of the 81 models that fit; every smaller model in the catalog fits too. Best quant means the highest quality compression whose weights and 4k context both sit inside the memory.

Close, but only with CPU offload

These need more than the card holds at their smallest practical quant, so part of the model runs from system memory (figures assume 32 GB of it). They work, several times slower.

How to read this

The AMD Radeon R9 M385X graphics card features 4 GB of GDDR5 memory. This dedicated video memory determines the size of the artificial intelligence models you can run directly on the hardware. To run a model entirely on the graphics processor, the model files and the active memory space must fit within this 4 GB limit. If a model exceeds this capacity, it cannot run solely on the video card.

Quantization is a method that compresses model files to save space. The quant column indicates the specific compression level used to fit these models into your hardware. For example, a Q4_K_M quant uses a four bit quantization level, while a Q8_0 quant uses an eight bit level. Higher quantization levels like Q8_0 preserve more model accuracy but require more memory. Lower quantization levels like Q4_K_M or Q5_K_M allow larger models to fit into the 4 GB limit.

With 4 GB of video memory, you can run several models entirely on your graphics card. The largest fitting models include Lumina-Next or Lumina-Image 2.0 at 5B using a Q4_K_M quant which uses 3.7 GB. DeepSeek-VL2 at 4.5B fits using a Q5_K_M quant and uses 3.8 GB. Qwen3 4B, Gemma 3 4B, Gemma 4 E4B, MiniCPM 3 4B, Danube 3 4B, and Fish Speech 1.5 or OpenAudio S1 at 4B all fit using a Q6_K quant which uses 3.9 GB. You can also run Stable Diffusion 3.5 Medium at 2.5B using a Q8_0 quant which uses 3.2 GB.

When a model is too large for the 4 GB video memory, you can use CPU offload. This process splits the workload between your graphics card and your system memory. We assume your computer has 32 GB of system RAM for these setups. Offloading allows you to run larger models, but it costs performance because system RAM is much slower than video memory. This transfer speed bottleneck will result in slower generation times.

Several models can run using CPU offload on this system. Mistral 7B needs 5.7 GB at Q4_K_M and requires 7.7 GB of system RAM. Qwen2.5 7B, OLMo 2 7B, Falcon 3 7B, Command R7B, and OpenHermes 2.5 at 7B all need 5.1 GB at Q4_K_M and require 7.1 GB of system RAM. Stable Diffusion XL at 3.417B needs 4.1 GB at FP8 or optimized settings and requires 6.1 GB of system RAM.

You must also consider the context window when running these models. The memory numbers listed here are calculated using a standard 4k context window. If you increase the context length to process longer documents or longer chat histories, the memory usage will increase. Running larger context windows might require you to use lower quantization levels or force more CPU offload.