Best local AI models for NVIDIA GTX 770M

The largest models that fit fully

The 30 largest of the 76 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 GTX 770M is a mobile graphics card equipped with 3 GB of GDDR5 video memory. This VRAM capacity determines which artificial intelligence models can run entirely on your graphics hardware. To load a model successfully, its active weights must fit within this 3 GB limit. Running models locally on your GPU ensures faster generation speeds compared to relying on your system processor.

The quantization column indicates the compression level applied to each model. Quantization reduces the precision of model weights to save memory. For example, Q4_K_M represents a four bit quantization format that balances model size and output quality. A higher quantization level like Q6_K or Q8_0 offers better accuracy but requires more memory. Models like Qwen3 4B and Gemma 3 4B use Q4_K_M to fit within 2.9 GB of VRAM.

When a model exceeds the 3 GB VRAM limit, you must use CPU offload. This technique splits the workload between your graphics card and your system RAM. For instance, running Phi-3 Mini at Q4_K_M requires 4.4 GB of total memory, which utilizes your 3 GB of VRAM and 6.4 GB of system RAM. CPU offload allows you to run larger models like Mistral 7B or CogVideoX 2B / 5B, but it significantly reduces processing speed.

Be aware of the memory cost associated with context length. The listed VRAM requirements assume a standard 4k context window for text generation. As you input longer prompts or generate longer responses, the memory footprint increases. If you push past the 4k context limit, the model may run out of memory and fail to respond on your 3 GB card.

This page helps you select the best configuration for your hardware. You can choose small, highly quantized models like SmolLM3 3B at Q6_K to run entirely in VRAM. Alternatively, you can leverage CPU offload with your 32 GB of system RAM to run larger architectures like Phi-4-multimodal or Magicoder-S-DS 6.7B at the cost of generation speed.