NVIDIA RTX 3080 Ti

The RTX 3080 Ti is a high-end Ampere GPU with 12 GB of GDDR6X memory and impressive 912 GB/s bandwidth - more bandwidth than the RTX 4070 Ti Super despite being two generations older. In 2026, used prices around $500-600 make it a fast option for running 7-13B models, but the 12 GB VRAM limit is felt with larger models.

Quick verdict

The 3080 Ti has excellent memory bandwidth (912 GB/s) that still competes with modern cards, but 12 GB VRAM limits it to 7-13B models at most. For $500-600 used, it's a decent gaming card that can also run local AI small models fast.

Spec breakdown

• VRAM: 12 GB GDDR6X
• Memory bandwidth: 912 GB/s (19 Gbps)
• TDP: 350 W (recommend 750W+ PSU)
• PCIe: 4.0 ×16
• Architecture: Ampere GA102-225
• CUDA cores: 10,240
• Tensor cores: 320 (3rd gen)

Real-world AI inference

Model	Tokens/sec	Source
Llama 3.1 8B Q4_K_M	~52 tok/s	Community
Qwen3-8B Q4_K_M	~50 tok/s	Community
Mistral Small 3 Q4_0	~22 tok/s	Tight fit
Qwen3-30B Q3_K_M	~8 tok/s	Offload needed

Honest alternatives

• Used RTX 3090 (~$700): 24 GB - double the VRAM for $200 more
• RTX 4070 Super (~$550): Similar VRAM, newer arch, lower power
• RTX 3080 12GB (~$450): Virtually identical, slightly cheaper