Best local AI models for NVIDIA 930MX

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 930MX is an entry level laptop graphics card equipped with 2 GB of DDR3 video memory. This limited VRAM capacity dictates which artificial intelligence models you can run locally. To load a model entirely on this GPU the model files and runtime data must fit within this 2 GB boundary. Running out of video memory causes execution to fail or slow down significantly.

Quantization is a compression method that reduces model size so it fits into smaller memory spaces. The quant column shows the best format for each model on this hardware. For example the 2.8B Allegro model fits in 2 GB of VRAM when compressed to the Q4_K_M quantization level. Smaller models like the 1.7B SmolLM2 can run at the higher quality Q6_K quantization level while using 1.7 GB of VRAM.

When a model exceeds the 2 GB VRAM limit you must use CPU offload. This technique splits the model layers between your GPU and your system RAM. We assume your system has 32 GB of system RAM for these scenarios. For instance running the 3B SmolLM3 requires 2.2 GB of VRAM at the Q4_K_M quantization level plus an additional 4.2 GB of system RAM.

Offloading models to system RAM comes with a performance cost. DDR3 video memory is already slow but system RAM is much slower. Models like Stable Diffusion XL which requires 4.1 GB of VRAM at FP8 and 6.1 GB of system RAM will experience slow generation times. The same speed penalty applies to other offloaded models such as the 3.5B SDXL Turbo.

Context window size also affects your memory usage. Running a text model with a standard 4k context window requires extra VRAM to store the active conversation history. If you use the maximum context length you might need to select a smaller model or a lower quantization level to prevent memory overflow on your 2 GB card.