NVIDIA GeForce RTX 3080 Ti — Local LLM Performance & Compatibility
作者： Jakub Rusinowski · 最后更新： 2026年9月19日
12 GB、912 GB/s——相对于该容量而言带宽异常高，适合能装进12 GB且受带宽限制的模型。仅二手市场可得。
Technical Specifications
| VRAM | 12 GB |
| Memory Bandwidth | 912 GB/s |
| TDP | 350 W |
| Architecture | Ampere GA102 |
| Release Year | 2021 |
| MSRP at Launch | $1,199 |
| Inference Speed (Llama 3.1 8B Q4_K_M) | 59–112 tok/s (estimated) |
| Inference Speed (Llama 3.3 70B Q4_K_M) | Does not fit — needs ~44 GB of 12 GB usable |
或在 Vast.ai 比较，低至 $0.35/小时 (typical low · varies)
作为亚马逊联盟成员，我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金，您无需额外付费。
LLMs Compatible with 12 GB VRAM
All models below run comfortably in 12 GB VRAM with Q4_K_M quantization.
| StarCoder 2 | StarCoder 2 15B · 10 GB VRAM · Q4_K_M · ollama run starcoder2:15b |
| Qwen 3 | Qwen 3 14B · 10 GB VRAM · Q4_K_M · ollama run qwen3:14b |
| DeepSeek R1 | DeepSeek R1 Distill Qwen 14B · 9 GB VRAM · Q4_K_M · ollama run deepseek-r1:14b |
| Phi-4 Family | Phi-4 (14B) · 9 GB VRAM · Q4_K_M · ollama run phi4 |
| Qwen 2.5 Family | Qwen 2.5 14B Instruct · 9 GB VRAM · Q4_K_M · ollama run qwen2.5:14b |
| Cogito v1 | Cogito v1 14B · 9 GB VRAM · Q4_K_M · ollama run cogito:14b |
| Ministral 3 | Ministral 3 14B · 9 GB VRAM · Q4_K_M · ollama run ministral-3:14b |
| OLMo 2 | OLMo 2 13B Instruct · 9 GB VRAM · Q4_K_M · ollama run olmo2:13b |
36 more families also fit 12 GB — browse the full model library.
Best Use Cases
- used 12 GB
- value inference
- high bandwidth per pound
FAQ
Can the NVIDIA GeForce RTX 3080 Ti run local LLMs?
Yes — the NVIDIA GeForce RTX 3080 Ti has 12 GB VRAM and runs 12 GB、912 GB/s——相对于该容量而言带宽异常高，适合能装进12 GB且受带宽限制的模型。仅二手市场可得。
How fast is the NVIDIA GeForce RTX 3080 Ti for AI inference?
The NVIDIA GeForce RTX 3080 Ti is estimated to run Llama 3.1 8B at 59–112 tok/s with Q4_K_M quantization. Llama 3.3 70B does not fit: it needs about 44 GB against 12 GB usable. These are modelled estimates, not measurements — see /en/methodology.
What LLMs can I run on 12 GB VRAM?
With 12 GB you can run: StarCoder 2, Qwen 3, DeepSeek R1, Phi-4 Family, Qwen 2.5 Family. Use Ollama for the easiest setup: ollama run llama3.1:8b.
Compare Similar GPUs
- NVIDIA GeForce RTX 3080 12GB (12 GB, 0 t/s)
- NVIDIA GeForce RTX 4060 Ti 8GB (8 GB, 0 t/s)
- AMD Radeon RX 7700 XT (12 GB, 0 t/s)
- NVIDIA GeForce RTX 4090 Laptop GPU (16 GB, 0 t/s)
VRAM Tier
Buying Guide
← All GPU Reviews | All Hardware | Check Your Hardware | Full Benchmarks | Can I Run It?