Best local AI models for AMD R9 M270X

The largest models that fit fully

The 30 largest of the 56 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 R9 M270X is an older graphics card equipped with 2 GB of GDDR5 memory. This onboard memory size is the strict limit for running local AI models entirely on your hardware. To fit models inside this small space, you must use quantized versions. Quantization reduces the precision of model weights to save memory. The quant column shows the best option that balances performance and size for each model.

For models that fit completely within the 2 GB limit, you can run them entirely on the graphics card. The largest models that fit this way include Allegro at 2.8B using the Q4_K_M quant with 2 GB used. Open-Sora Plan at 2.7B also fits at Q4_K_M with 2 GB used. You can also run LFM2 2.6B and Playground v2.5 at Q4_K_M, which both use 1.9 GB of memory.

Other options that fit on the card include Stable Diffusion 3.5 Medium and Canary 2.5B at Q4_K_M, using 1.8 GB of memory. SeamlessM4T v2 at 2.3B, Parler-TTS at 2.2B, and Kimi K3 DSpark at 2.2B fit using the Q5_K_M quant. You can run SmolVLM 2B, Stable Diffusion 3 Medium, Pyramid Flow, and Wav2Vec2 at Q6_K, which use exactly 2 GB of memory. Moondream 2 at 1.9B fits at Q6_K with 1.9 GB used.

Smaller models can run at higher precision. Qwen3 1.7B and SmolLM2 1.7B fit at Q6_K using 1.7 GB of memory. StableLM 2 1.6B, Sana 1.6B, Zonos 0.1, and Dia 1.6B can run at Q8_0 using 2 GB of memory. Whisper Large v3 at 1.55B also runs at Q8_0 using 2 GB of memory. ControlNet, Hunyuan-DiT, Stable Video Diffusion, Whisper Large v2, AudioGen, and AudioLDM 2 all run at Q8_0 using 1.9 GB of memory. Tango 2 at 1.4B runs at Q8_0 using 1.8 GB of memory. TinyLlama 1.1B and SantaCoder 1.1B run at Q8_0 using 1.4 GB of memory.

If you want to run larger models, you must use CPU offloading. This process spills extra data into your system RAM. Offloading allows you to run models like SmolLM3 3B, Replit Code v1.5 3B, Kandinsky 3.1, Voxtral Mini, Orpheus TTS, and Higgs Audio v2. These models need 2.2 GB at Q4_K_M and require 4.2 GB of system RAM. MusicGen needs 2.4 GB at Q4_K_M and requires 4.4 GB of system RAM.

Larger image generation models also require offloading. Stable Diffusion XL at 3.417B needs 4.1 GB at FP8 and requires 6.1 GB of system RAM. SDXL Turbo at 3.5B and SDXL Lightning at 3.5B both need 2.6 GB at Q4_K_M and require 4.6 GB of system RAM. Offloading makes these models possible but it reduces your generation speed significantly.

You must also consider the context window when running local text models. The memory figures listed here assume a standard context size. If you increase the context window to 4k tokens, the model will require more memory. This extra memory usage can easily exceed the 2 GB limit of your card and force the system to slow down.