Best local AI models for AMD HD 7970M

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 HD 7970M is a mobile graphics card equipped with 2 GB of GDDR5 video memory. This hardware limit determines which artificial intelligence models can run directly on your graphics processor. When running local models, the entire active weight set should ideally fit inside this video memory to ensure acceptable generation speeds.

The quantization column indicates the compression level applied to each model. For example, the Allegro 2.8B model fits in your 2 GB memory limit when compressed to the Q4_K_M quantization level. Smaller models like SmolLM2 1.7B can run at a higher quality Q6_K quantization while using 1.7 GB of video memory. Ultra small models like TinyLlama 1.1B can run at Q8_0 quantization using only 1.4 GB of video memory.

If a model exceeds your video memory, you can offload parts of it to your system RAM. This page assumes you have 32 GB of system RAM available for offloading. For instance, the SmolLM3 3B model requires 2.2 GB of video memory at Q4_K_M quantization and offloads the remaining data to use 4.2 GB of system RAM. This allows you to run larger options like SDXL Turbo or SDXL Lightning which need 2.6 GB of video memory and 4.6 GB of system RAM.

Offloading comes with a significant performance cost. Sending data between your AMD HD 7970M and your system RAM over the system bus is much slower than reading directly from GDDR5 memory. While offloading allows you to run larger models like MusicGen or Kandinsky 3.1, your generation speeds will drop noticeably compared to models that fit entirely within your 2 GB video memory.

You must also consider the context window size when calculating memory usage. Running a model with a standard 4k context window requires additional video memory to store the active conversation history. If you use the maximum context length, you may need to select a smaller model size or a lower quantization level to prevent your graphics card from running out of memory.