Best local AI models for NVIDIA 845M

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 845M is an entry level mobile graphics card equipped with 2 GB GDDR5 video memory. This hardware configuration dictates the maximum size of the artificial intelligence models you can run locally. To load a model entirely on this graphics processor, the total memory footprint must remain under the 2 GB physical limit. This constraint requires careful selection of model sizes and quantization levels.

Quantization is a compression method that reduces the precision of model weights. The quant column indicates the specific compression format used to fit these models into your hardware. For example, the Allegro 2.8B model fits into 2 GB of video memory when compressed to the Q4_K_M quantization level. Using a higher compression level like Q4_K_M or Q5_K_M allows larger models to run, while smaller models like TinyLlama 1.1B can run at the higher quality Q8_0 quantization level using 1.4 GB of video memory.

Running models at their maximum capacity leaves very little room for context. When you run a model near the 2 GB limit, the active memory context is severely restricted. Running a 2B model like Moondream 2 at Q6_K uses 1.9 GB of video memory, which leaves almost no space for processing long text prompts or generating extended responses. For practical tasks requiring a standard 4k context window, you must choose smaller models like the SantaCoder 1.1B at Q8_0 to avoid running out of memory.

If you want to run larger models, you must use CPU offloading. This technique splits the model weights between your 2 GB video memory and your system RAM. We assume your computer has 32 GB of system RAM for these scenarios. Offloading allows you to run models like SmolLM3 3B, which requires 2.2 GB of video memory at Q4_K_M and an additional 4.2 GB of system RAM. You can also run image generators like Stable Diffusion XL, which requires 4.1 GB of video memory at FP8 or optimized settings along with 6.1 GB of system RAM.

CPU offloading comes with a significant performance cost. Transferring data between the system RAM and your NVIDIA 845M processor over the system bus is much slower than reading directly from the onboard GDDR5 memory. While offloading makes it possible to run larger tools like MusicGen or SDXL Turbo, the generation speed will be much slower than running a fully contained model like the Qwen3 1.7B at Q6_K directly on the graphics card.