Best local AI models for NVIDIA 810M

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 810M is an entry level mobile graphics card equipped with 2 GB of DDR3 memory. This hardware configuration places strict limits on the size of the artificial intelligence models you can run locally. Because the onboard memory is capped at 2 GB, any model you load entirely onto the graphics card must fit within this small footprint. The memory size determines whether a model can run directly on your hardware or if it will fail to load due to out of memory errors.

To fit models onto this hardware, you must use quantized versions. Quantization reduces the precision of model weights to save space. The best quant column shows the optimal balance of size and quality for each model. For example, the Allegro 2.8B and Open-Sora Plan 2.7B models require a Q4_K_M quantization to fit exactly into the 2 GB limit. Smaller models like SmolVLM 2B or Stable Diffusion 3 Medium can run at a higher quality Q6_K quantization while still using exactly 2 GB of memory.

Very small models can run at even higher precision levels. The StableLM 2 1.6B and Sana 1.6B models run at Q8_0 quantization using 2 GB of memory. You can also run Whisper Large v3 at Q8_0 using 2 GB of memory. Extremely compact options like TinyLlama 1.1B and SantaCoder 1.1B at Q8_0 quantization require only 1.4 GB of memory. This leaves a small safety margin of free memory on your graphics card.

When a model is too large for the 2 GB graphics memory, you must use CPU offloading. This process splits the model between your graphics card and your system RAM. We assume your computer has 32 GB of system RAM for these setups. For instance, SmolLM3 3B and Replit Code v1.5 3B need 2.2 GB of memory at Q4_K_M quantization. This configuration requires 4.2 GB of system RAM to handle the overflow. Similarly, Stable Diffusion XL needs 4.1 GB at FP8 or optimized settings, which requires 6.1 GB of system RAM.

CPU offloading allows you to run larger tools but it comes with a severe performance cost. System RAM is much slower than graphics memory, especially with DDR3 technology. Models like MusicGen at 3.3B or SDXL Turbo at 3.5B will experience slow processing speeds during offloading. You will get much slower generation times compared to models that fit entirely within the 2 GB graphics memory.

You must also consider the context window when running text models. Running models at their maximum context limit, such as 4k context, requires extra memory for the active session data. This extra data can easily push your memory usage past the 2 GB limit. If you experience crashes during long conversations, you must lower the active context limit to keep the total memory footprint within your hardware limits.