Best local AI models for NVIDIA 945M

The largest models that fit fully

The 28 largest of the 28 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 945M graphics card features 1 GB of DDR3 video memory. This limited memory capacity dictates which local artificial intelligence models you can run directly on the hardware. To fit within this 1 GB limit, models must be small or use quantization to reduce their size. The memory size listed for each model shows the exact amount of video memory required to load and execute the network.

Quantization is a compression method that lowers the precision of model weights to save space. The best quant column shows the optimal balance between model accuracy and memory usage. For example, Tango 2 at 1.4B parameters fits in 1 GB of video memory using the Q4_K_M quantization level. Smaller models like Qwen3 0.6B can use the higher quality Q8_0 quantization level while only requiring 0.8 GB of video memory.

Very small models can run without quantization at their native precision. Models such as VALL-E X (unofficial) at 0.4B parameters and ChatTTS at 0.3B parameters run at FP16 precision while using 1 GB and 0.7 GB of video memory respectively. Other lightweight models like F5-TTS and E2-TTS also run at FP16 precision and use 0.7 GB of video memory. StyleTTS 2 is the smallest option at 0.15B parameters using 0.4 GB of video memory at FP16.

When a model exceeds the 1 GB video memory limit of the NVIDIA 945M, you must use CPU offload. This process splits the model workload between your graphics card and your system RAM. We assume your computer has 32 GB of system RAM for these scenarios. CPU offload allows you to run larger models, but it significantly reduces processing speed because system RAM is much slower than video memory.

Several popular models require CPU offload to function. Stable Diffusion 1.5 has 1.07B parameters and needs 1.3 GB of memory at FP8 or optimized settings, which requires 3.3 GB of system RAM. Larger 1.5B parameter models like Hunyuan-DiT, Stable Video Diffusion, and Whisper Large v2 or turbo require 1.1 GB of video memory at Q4_K_M quantization and 3.1 GB of system RAM. StableLM 2 1.6B requires 1.2 GB of video memory at Q4_K_M quantization and 3.2 GB of system RAM.

You must consider the 4k context caveat when running text models on this hardware. Generating longer responses or processing large prompts increases memory consumption during runtime. A model that fits perfectly at startup might run out of memory as the conversation history grows. You should monitor your memory usage closely if you plan to use these models for extended tasks.