Skip to content
RfRigFit

What fits in 12 GB of VRAM

Data updated . Last reviewed the same day, when this catalog snapshot was written.

Usable memory here is 11.4 GB, which is 95% of 12 GB. 21 of 49 catalog models fit at Q4, Q8, or FP16. 6 catalog GPUs have this much VRAM. 6 of the fits do not fit the 8 GB tier.

Models

Sorted by the highest quant that fits, then by headroom. The tok/s estimate uses GeForce RTX 5070 bandwidth only.

Models that fit in 12 GB VRAM
ModelBest quantEstimateHeadroomEst. tok/s
Llama 3.2 1B InstructFP163.0 GB8.4 GB272 tok/s
Llama 3.2 3B InstructFP167.7 GB3.7 GB105 tok/s
Phi-3.5 miniFP169.2 GB2.2 GB87.9 tok/s
Phi-4 miniFP169.2 GB2.2 GB87.6 tok/s
Gemma 3 4BFP1610.3 GB1.1 GB78.1 tok/s
DeepSeek V4 FlashQ84.5 GB6.9 GB3950 tok/s
Gemma 4 E4BQ88.3 GB3.1 GB82.0 tok/s
Qwen2.5 7B InstructQ88.6 GB2.8 GB83.0 tok/s
Mistral 7B Instruct v0.3Q88.8 GB2.6 GB87.2 tok/s
DeepSeek R1 Distill Llama 8BQ89.6 GB1.8 GB78.7 tok/s
Llama 3.1 8B InstructQ89.6 GB1.8 GB83.7 tok/s
Qwen3 8BQ89.9 GB1.5 GB77.2 tok/s
Ministral 3 8BQ810.2 GB1.2 GB74.4 tok/s
Gemma 2 9BQ810.7 GB0.7 GB62.9 tok/s
Gemma 3 12BQ47.3 GB4.1 GB92.0 tok/s
Gemma 4 12BQ47.7 GB3.7 GB94.4 tok/s
Mistral Nemo 12BQ48.8 GB2.6 GB89.9 tok/s
Ministral 3 14BQ49.6 GB1.8 GB81.6 tok/s
Qwen3 14BQ410.3 GB1.1 GB74.7 tok/s
Phi-4Q410.6 GB0.8 GB75.6 tok/s
Qwen2.5 14B InstructQ410.6 GB0.8 GB74.8 tok/s

Next misses at Q4: gpt-oss-20b (11.8 GB), Mistral Small 3.2 24B (15.7 GB), Gemma 3 27B (16.5 GB), Gemma 2 27B (16.6 GB), Qwen3.8 27B (17.0 GB).

GPUs at 12 GB

Questions

How many catalog models fit in 12 GB of VRAM?
21 of 49 fit at Q4 or better inside 11.4 GB usable (95% of 12 GB).
Which catalog GPUs have 12 GB of VRAM?
GeForce RTX 3060 12GB, Radeon RX 7700 XT, GeForce RTX 5070, GeForce RTX 4070, GeForce RTX 4070 Super, GeForce RTX 4070 Ti.
Is the tokens/sec column a benchmark?
No. It divides the GeForce RTX 5070's 672 GB/s by active weight size. Other 12 GB cards are slower or faster than that.