Best local AI models for NVIDIA 820M

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 820M is an entry level mobile graphics card equipped with 2 GB of DDR3 memory. This dedicated video memory is the primary constraint when running artificial intelligence models locally. To run a model entirely on this hardware, the model files must fit within this 2 GB limit. If a model exceeds this capacity, the system must offload data to your system memory, which slows down processing speeds.

The quantization column indicates the compression level used to shrink these models. Quantization reduces the precision of model weights to save space. For example, the Allegro 2.8B model fits in 2 GB of memory when compressed to the Q4_K_M quantization level. Smaller models like SmolLM2 1.7B can run at the higher quality Q6_K quantization level while using 1.7 GB of memory. TinyLlama 1.1B can run at the Q8_0 quantization level using only 1.4 GB of memory.

When a model size exceeds the 2 GB limit of the graphics card, you can use CPU offloading. This process splits the workload between your graphics card and your system RAM. For instance, running the SmolLM3 3B model at Q4_K_M requires 2.2 GB of video memory and 4.2 GB of system RAM. Similarly, the Stable Diffusion XL model requires 4.1 GB of video memory and 6.1 GB of system RAM when using FP8 or optimized settings.

CPU offloading allows you to run larger architectures like the 3.5B parameter SDXL Turbo or SDXL Lightning. These models require 2.6 GB of video memory and 4.6 GB of system RAM at the Q4_K_M quantization level. While offloading enables these models to function on your system, the transfer of data over the system bus reduces generation speeds compared to running entirely within the dedicated video memory.

You must also consider the memory cost of context length when running text models. The memory figures listed cover the base model weights only. Running a model with a standard 4k context window requires additional memory to store the active conversation history. On a card with 2 GB of DDR3 memory, allocating space for this context window may require you to choose smaller models or lower quantization levels to prevent system slowdowns.