Best local AI models for NVIDIA GTX 750 Ti

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 GTX 750 Ti is an entry level graphics card with 2 GB GDDR5 memory. This memory size is the main limit for running local AI models. The memory size column shows how much video memory the model needs to run. To fit inside this limit, models must use quantization. The quant column shows the best format that fits your hardware. A format like Q4_K_M or Q8_0 reduces the model size so it can run on your card.

For models that fit entirely on the card, Allegro 2.8B is the largest option using a Q4_K_M quant and 2 GB used. Other large options include Open-Sora Plan 2.7B at Q4_K_M with 2 GB used, and LFM2 2.6B at Q4_K_M with 1.9 GB used. You can also run Playground v2.5 2.6B at Q4_K_M with 1.9 GB used, or Stable Diffusion 3.5 Medium 2.5B at Q4_K_M with 1.8 GB used. Canary 2.5B uses 1.8 GB with a Q4_K_M quant.

Medium sized options include SeamlessM4T v2 2.3B using 2 GB with a Q5_K_M quant. Parler-TTS 2.2B and Kimi K3 DSpark 2.2B run well using Q5_K_M quants. You can run SmolVLM 2B, Stable Diffusion 3 Medium 2B, Pyramid Flow 2B, and Wav2Vec2 2B using Q6_K quants which use exactly 2 GB. Moondream 2 1.9B fits well using Q6_K with 1.9 GB used. Qwen3 1.7B and SmolLM2 1.7B both use 1.7 GB with Q6_K quants.

Smaller models can use higher quality quants like Q8_0. StableLM 2 1.6B, Sana 1.6B, Zonos 0.1 1.6B, and Dia 1.6B all use Q8_0 and require 2 GB. Whisper Large v3 1.55B also uses Q8_0 with 2 GB used. ControlNet 1.5B, Hunyuan-DiT 1.5B, Stable Video Diffusion 1.5B, Whisper Large v2 1.5B, AudioGen 1.5B, and AudioLDM 2 1.5B all run at Q8_0 using 1.9 GB. Tango 2 1.4B uses 1.8 GB at Q8_0. TinyLlama 1.1B and SantaCoder 1.1B use 1.4 GB at Q8_0.

You can run larger models by using CPU offload if you have 32 GB system RAM. Offload costs speed because system RAM is much slower than video memory. SmolLM3 3B, Replit Code v1.5 3B, Kandinsky 3.1 3B, Voxtral Mini 3B, Orpheus TTS 3B, and Higgs Audio v2 3B need 2.2 GB on the card at Q4_K_M and 4.2 GB system RAM. MusicGen 3.3B needs 2.4 GB on the card and 4.4 GB system RAM. SDXL Turbo 3.5B and SDXL Lightning 3.5B need 2.6 GB on the card and 4.6 GB system RAM. Stable Diffusion XL 3.417B needs 4.1 GB at FP8 and 6.1 GB system RAM.

There is an important caveat regarding the 4k context window. Running text models with a 4k context window increases memory usage. This extra memory demand can exceed the 2 GB limit of your card. If you experience out of memory errors, you must reduce the context length or use CPU offload.