Best local AI models for NVIDIA GTX 980 Ti
6 GB GDDR5. At a 4k context, 114 of the 233 models in our catalog with verified parameter counts fit fully, up to Granite 3.3 2B / 8B at 8B parameters.
Check your own machine against every model →The largest models that fit fully
The 30 largest of the 114 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.
| Model | Parameters | Best quant that fits | Memory used at 4k |
|---|---|---|---|
| Granite 3.3 2B / 8B | 8B | Q4_K_M | 5.9 GB |
| Ministral 3B / 8B | 8B | Q4_K_M | 5.9 GB |
| InternLM 3 8B | 8B | Q4_K_M | 5.9 GB |
| OpenCoder 1.5B / 8B | 8B | Q4_K_M | 5.9 GB |
| Seed-Coder 8B | 8B | Q4_K_M | 5.9 GB |
| MiniCPM-V 2.6 / MiniCPM-o 2.6 | 8B | Q4_K_M | 5.9 GB |
| Idefics 3 8B | 8B | Q4_K_M | 5.9 GB |
| Fuyu-8B | 8B | Q4_K_M | 5.9 GB |
| Emu3 | 8B | Q4_K_M | 5.9 GB |
| Stable Diffusion 3.5 Large / Turbo | 8B | Q4_K_M | 5.9 GB |
| EXAONE 3.5 2.4B / 7.8B | 7.8B | Q4_K_M | 5.7 GB |
| Mistral 7B | 7B | Q4_K_M | 5.7 GB |
| Qwen2.5 0.5B / 1.5B / 3B / 7B | 7B | Q5_K_M | 6 GB |
| OLMo 2 1B / 7B | 7B | Q5_K_M | 6 GB |
| Falcon 3 1B / 3B / 7B | 7B | Q5_K_M | 6 GB |
| Command R7B | 7B | Q5_K_M | 6 GB |
| OpenHermes 2.5 | 7B | Q5_K_M | 6 GB |
| Zephyr 7B Beta | 7B | Q5_K_M | 6 GB |
| OpenChat 3.5 | 7B | Q5_K_M | 6 GB |
| Starling LM 7B | 7B | Q5_K_M | 6 GB |
| Codestral Mamba 7B | 7B | Q5_K_M | 6 GB |
| CodeGemma 2B / 7B | 7B | Q5_K_M | 6 GB |
| aiXcoder-7B | 7B | Q5_K_M | 6 GB |
| Nxcode / CodeQwen 1.5 7B | 7B | Q5_K_M | 6 GB |
| Janus-Pro 1B / 7B | 7B | Q5_K_M | 6 GB |
| Ruyi-Mini-7B | 7B | Q5_K_M | 6 GB |
| Qwen2-Audio 7B | 7B | Q5_K_M | 6 GB |
| Qwen2.5-Omni 3B / 7B | 7B | Q5_K_M | 6 GB |
| YuE | 7B | Q5_K_M | 6 GB |
| Magicoder-S-DS 6.7B | 6.7B | Q5_K_M | 5.7 GB |
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.
| Model | Parameters | Memory at Q4_K_M | System RAM at 4k |
|---|---|---|---|
| Llama 3.1 8B | 8B | 6.4 GB needed | 8.4 GB |
| Chroma | 8.9B | 6.5 GB needed | 8.5 GB |
| Gemma 2 9B | 9B | 8 GB needed | 10 GB |
| Nemotron Nano 4B / 9B | 9B | 6.6 GB needed | 8.6 GB |
| GLM-4 9B / GLM-4.5-Air | 9B | 6.6 GB needed | 8.6 GB |
| Yi-Coder 1.5B / 9B | 9B | 6.6 GB needed | 8.6 GB |
| GLM-4-9B-Chat / CodeGeeX4 | 9B | 6.6 GB needed | 8.6 GB |
| GLM-4V-9B / GLM-4.1V-Thinking | 9B | 6.6 GB needed | 8.6 GB |
| Mochi 1 | 10B | 7.3 GB needed | 9.3 GB |
| Open-Sora 2.0 | 11B | 8.1 GB needed | 10.1 GB |
How to read this
The NVIDIA GTX 980 Ti features 6 GB of GDDR5 onboard memory. This memory size determines which local AI models you can run entirely on your graphics hardware. To run a model smoothly without system slowdowns, the model files and the active context must fit completely within this 6 GB limit.
The quantization column indicates the compression level applied to each model. For example, the Q4_K_M and Q5_K_M formats compress model weights to four or five bits per parameter. This compression allows larger models to fit into your VRAM. A Q4_K_M quant uses less memory than a Q5_K_M quant but sacrifices a small amount of output precision.
Many modern 7B and 8B models fit directly into your 6 GB VRAM. Granite 3.3 8B, Ministral 8B, InternLM 3 8B, OpenCoder 8B, Seed-Coder 8B, MiniCPM-V 2.6, MiniCPM-o 2.6, Idefics 3 8B, Fuyu-8B, Emu3, and Stable Diffusion 3.5 Large use 5.9 GB of VRAM at Q4_K_M. EXAONE 3.5 7.8B and Mistral 7B require 5.7 GB of VRAM at Q4_K_M. Magicoder-S-DS 6.7B also fits well at Q5_K_M using 5.7 GB of VRAM.
Several 7B models utilize exactly 6 GB of VRAM at the Q5_K_M quantization level. These models include Qwen2.5 7B, OLMo 2 7B, Falcon 3 7B, Command R7B, OpenHermes 2.5, Zephyr 7B Beta, OpenChat 3.5, Starling LM 7B, Codestral Mamba 7B, CodeGemma 7B, aiXcoder-7B, Nxcode, CodeQwen 1.5 7B, Janus-Pro 7B, Ruyi-Mini-7B, Qwen2-Audio 7B, Qwen2.5-Omni 7B, and YuE. Running these models at 6 GB leaves no headroom for extended context windows.
When a model exceeds 6 GB you must offload layers to your system RAM. If you have 32 GB of system RAM you can run larger models with a speed penalty. Llama 3.1 8B needs 6.4 GB at Q4_K_M and requires 8.4 GB of system RAM. Chroma needs 6.5 GB at Q4_K_M and requires 8.5 GB of system RAM. Gemma 2 9B needs 8 GB at Q4_K_M and requires 10 GB of system RAM. Mochi 1 needs 7.3 GB at Q4_K_M and requires 9.3 GB of system RAM. Open-Sora 2.0 needs 8.1 GB at Q4_K_M and requires 10.1 GB of system RAM.
Other offload options include Nemotron Nano 9B, GLM-4 9B, Yi-Coder 9B, GLM-4-9B-Chat, CodeGeeX4, GLM-4V-9B, and GLM-4.1V-Thinking. These models all need 6.6 GB at Q4_K_M and require 8.6 GB of system RAM. Offloading layers to system RAM prevents out of memory errors but reduces generation speeds significantly.
VRAM calculations assume a standard 4k context window. If you increase the context length during your session the memory usage will rise. This extra memory demand can cause your system to exceed the 6 GB VRAM limit of the GTX 980 Ti and trigger slow CPU offloading.