Best local AI models for NVIDIA 830A

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 830A is an entry level graphics processor with 2 GB of DDR3 memory. This hardware configuration limits the size of the artificial intelligence models you can run locally. To load a model entirely on this GPU, the total memory used by the model must remain under the 2 GB physical limit.

The quant column shows the quantization level recommended for each model. Quantization reduces the precision of model weights to save space. For example, the 2.8B Allegro and 2.7B Open-Sora Plan models fit in 2 GB of memory using the Q4_K_M quantization. Smaller models like the 2B Stable Diffusion 3 Medium can run at a higher Q6_K quantization while using 2 GB of memory. Models like the 1.6B Dia and 1.55B Whisper Large v3 can run at Q8_0 quantization because they require less space.

If a model is too large for the 2 GB GPU memory, you can use CPU offload. This method splits the model between your GPU and your system RAM. We assume your computer has 32 GB of system RAM for these calculations. Offloading allows you to run larger models like the 3B SmolLM3 or the 3.5B SDXL Turbo. The downside of CPU offload is a slower processing speed because data must travel between the system RAM and the GPU.

When using CPU offload, you must track both GPU memory and system RAM usage. The 3B Kandinsky 3.1 needs 2.2 GB of GPU memory at Q4_K_M quantization and 4.2 GB of system RAM. The 3.417B Stable Diffusion XL needs 4.1 GB of GPU memory at FP8 or optimized settings and 6.1 GB of system RAM. This configuration exceeds the physical 2 GB limit of your GPU, so offloading is required.

You must also consider the 4k context caveat when running local text models. As your conversation or document length increases, the model requires more memory to track the context. A model that fits perfectly at start up might run out of memory as the conversation grows. You should monitor your memory usage closely during long sessions to avoid crashes.