NVIDIA GeForce RTX 3080 Ti
Ampere flagship-minus-one. 12 GB GDDR6X at 912 GB/s — closer to the 3090 in raw bandwidth than to the 3080. Fits 13B Q4 with full context, 32B Q4 with offload. ~155-205 tok/s on 7B Q4. Used $440-520 in 2026; the 'budget 3090' for operators willing to give up 12 GB of VRAM headroom for a meaningful price cut.
Affiliate disclosure: as an Amazon Associate and partner of other retailers, we earn from qualifying purchases. The verdict on this page is our editorial opinion; affiliate links never influence what we recommend.
Sub-scores sum to 652 / 1000. Headline = 652 × 0.70 (Estimated-confidence discount) = 456. This is an algorithmic performance-tier score — distinct from, and often lower than, the editorial “Our verdict” below, which weighs value and real-world fit (especially for hardware we haven’t measured yet). How scoring works →
Extrapolated from 912 GB/s bandwidth — 109.4 tok/s estimated. No measured benchmarks yet.
Plain-English: Comfortable at 14B and below — snappy enough for a coding agent; vision models supported.
Verdicts extrapolated from catalog VRAM + bandwidth + ecosystem flags. Hover any chip for the rationale. Want measured numbers? Submit your own run with runlocalai-bench --submit.
WHO THIS CARD IS FOR
The operator who wants 3090-like inference speed on 7B-13B models but can work within a 12 GB VRAM budget and prefers to spend under $500. This is the card for running 13B Q4 with full context or 30B Q4 with partial offload, not for 70B or large multimodal models.
WHAT IT RUNS WELL
7B Q4 models run at 155-205 tok/s, making this one of the fastest cards for small-to-medium local inference. 13B Q4 fits entirely in VRAM at 80-110 tok/s. 30B Q4 (with ~20 GB weights) requires offloading to system RAM, but the high bandwidth keeps the CPU-GPU transfer bottleneck tolerable for interactive use.
WHAT BREAKS
12 GB VRAM is the hard ceiling. 70B Q4 is out of reach without aggressive quantization and heavy offload. 32B Q4 with full context may spill over, causing performance drops. The card runs hot and draws 350W under load; operators need adequate cooling and a 750W+ PSU.
WHEN TO PASS
If the workload regularly exceeds 12 GB (e.g., 30B+ models with long context, or running multiple models simultaneously), step up to a used 3090 or 4090. For pure 7B inference at lower cost, a used 3080 10 GB or 3070 may suffice.
PRICE / VALUE NOTE
At ~$480 used, this is the best price-to-bandwidth ratio for 7B-13B inference, effectively a 'budget 3090' for operators who can live with 12 GB.
›Why this rating
The RTX 3080 Ti delivers exceptional inference speed for its price, with bandwidth rivaling the 3090. The 12 GB VRAM is the main limiter, but for the target workload of 7B-13B models, it's a top value. Loses points for high power draw and lack of future-proofing for larger models.
Overview
Ampere flagship-minus-one. 12 GB GDDR6X at 912 GB/s — closer to the 3090 in raw bandwidth than to the 3080. Fits 13B Q4 with full context, 32B Q4 with offload. ~155-205 tok/s on 7B Q4. Used $440-520 in 2026; the 'budget 3090' for operators willing to give up 12 GB of VRAM headroom for a meaningful price cut.
Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.
Specs
| VRAM | 12 GB |
| Power draw (peak) | 350 W |
| Released | 2021 |
| MSRP | $1199 |
| Backends | CUDA Vulkan |
Models that fit
Open-weight models small enough to run on NVIDIA GeForce RTX 3080 Ti with usable context.
Frequently asked
What models can NVIDIA GeForce RTX 3080 Ti run?
Does NVIDIA GeForce RTX 3080 Ti support CUDA?
How much does NVIDIA GeForce RTX 3080 Ti cost?
Where next?
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify hardware specifications.