Best local AI models for NVIDIA GTX 850M

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 GTX 850M is an older mobile graphics card equipped with 2 GB of DDR3 memory. This hardware configuration places strict limits on the size of local AI models you can run entirely on the GPU. To load a model successfully, its active weights must fit within this 2 GB limit. If a model exceeds this capacity, your system will experience severe slowdowns or fail to run the model at all.

To fit larger models into this limited space, you must use quantized files. The quantization column shows the optimal format for each model. Quantization compresses the model weights to use fewer bits. For example, a Q4_K_M quant uses a 4 bit format to save space. A Q8_0 quant uses 8 bits, which preserves more quality but requires more memory. Choosing the right quant is necessary to balance performance and accuracy on this hardware.

Several small models can run entirely within your 2 GB memory limit. The largest fully compatible model is Allegro 2.8B using the Q4_K_M quant, which uses exactly 2 GB. Other options include Open-Sora Plan 2.7B at Q4_K_M using 2 GB, and LFM2 2.6B at Q4_K_M using 1.9 GB. For audio tasks, you can run Whisper Large v3 1.55B at Q8_0 using 2 GB, or Parler-TTS 2.2B at Q5_K_M using 1.9 GB.

If you want to run larger models, you must use CPU offload. This process splits the model between your GPU and your system RAM. We assume your system has 32 GB of system RAM for these setups. For example, SmolLM3 3B needs 2.2 GB of GPU memory at Q4_K_M and requires 4.2 GB of system RAM. MusicGen small/medium/large 3.3B needs 2.4 GB of GPU memory at Q4_K_M and 4.4 GB of system RAM. Stable Diffusion XL 3.417B needs 4.1 GB of GPU memory at FP8 / optimized and 6.1 GB of system RAM.

CPU offload comes with a major performance cost. Transferring data between the DDR3 GPU memory and system RAM is very slow. This transfer bottleneck will greatly reduce your generation speed. Additionally, running text models with a standard 4k context window increases memory usage during generation. You must monitor your memory closely because long conversations can quickly exceed the 2 GB limit.