Best local AI models for NVIDIA 910M

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 NVIDIA 910M is an entry level mobile graphics card equipped with 2 GB of DDR3 video memory. This memory size is the absolute limit for loading artificial intelligence models directly onto the hardware. To run a model entirely on this graphics card the total size of the model files must not exceed this physical limit. Because of this tight boundary you must select highly optimized versions of your chosen models.

The quantization column shows the best format to use for each model. Quantization is a compression method that reduces the precision of model weights to save space. For example a Q4_K_M quantization uses a four bit format to compress larger models like Allegro 2.8B or Open-Sora Plan 2.7B down to 2 GB of used video memory. Models with smaller parameter counts like SmolLM2 1.7B can run at a higher quality Q6_K quantization while using 1.7 GB of memory. The smallest models like TinyLlama 1.1B can run at the highest quality Q8_0 quantization while using only 1.4 GB of memory.

When a model exceeds the 2 GB limit you must use CPU offload. This technique splits the workload between your graphics card and your system memory. For this setup we assume your computer has 32 GB of system RAM. Offloading allows you to run larger models like SmolLM3 3B or Kandinsky 3.1 which need 2.2 GB of video memory and 4.2 GB of system RAM. You can even run Stable Diffusion XL which requires 4.1 GB of video memory and 6.1 GB of system RAM by offloading the extra weight.

CPU offload comes with a significant performance cost. System RAM is much slower than video memory especially when paired with DDR3 technology. While offloading makes it possible to run models like SDXL Turbo or MusicGen on your system the processing speed will drop. The transfer of data between the system RAM and the graphics card creates a bottleneck that slows down generation times.

You must also consider the context window when running text models on this hardware. The memory figures listed here are calculated using a basic 4k context window. If you increase the context length to process longer documents the memory usage will rise quickly. Keeping your context window at or below 4k is necessary to prevent your system from running out of memory during generation.