GPUFit AIGPUFit AI

30 models · 20 GPUs · updated 2026-10-10

LLM VRAM Calculator

Estimate how much VRAM a local model needs at each quantization, then see which of 20 GPUs can hold it.

31.27B

30 models.

Quantization 4.5 bits

Q4_K_M is the usual starting point for 8 to 24 GB cards.

Optional. Pick your card to get a verdict and see it highlighted in the list.

Estimated VRAM

Smallest card that fits: RTX 3090, with 5.52 GB to spare

Fits at Q4_K_M with 8,192 tokens of context, with room left for your desktop and browser.

18.48 GB

Gemma 4 31B · Q4_K_M · 8,192 tokens

Weights

16.38 GB

params × bits ÷ 8

KV cache

1.09 GB

2 × layers × KV heads × head dim × tokens × 2 B

Overhead

1.00 GB

fixed buffers and context

Enough room to step up to Q5_K_MUses 22.12 GB, 1.88 GB to spare

Which GPUs fit

Cards that fit first 8 of 20 shown

Estimate, not a measurement. Usage varies with runtime, batch size, and drivers. How we calculate.

Popular models

All 30 models

Total VRAM in GB at 8K context. Static table, readable without JavaScript.

Totals are weights, cache, and 1 GB of overhead at 8,192 tokens, or at the model's maximum when that maximum is lower. FP16 weights alone are the parameter count times 2 bytes.
Model Parameters Q4_K_M Q8_0 FP16
Gemma 4Parameters 31.27B · Q8_0 33.04 31.27B 18.48 33.04 60.34
gpt-oss-120bParameters 116.8B · Q8_0 116.9 116.8B 62.49 116.9 218.9
gpt-oss-20bParameters 20.91B · Q8_0 21.89 20.91B 12.15 21.89 40.15
Kimi K2Parameters 1026.4B · Q8_0 1017.2 1026.4B 539.2 1017.2 1913.4
DeepSeek R1Parameters 684.5B · Q8_0 678.9 684.5B 360.1 678.9 1276.5
Llama 3.3 70BParameters 70.55B · Q8_0 73.32 70.55B 40.46 73.32 134.9
Llama 3.1 8BParameters 8.03B · Q8_0 9.95 8.03B 6.21 9.95 16.96
Llama 3.1 70BParameters 70.55B · Q8_0 73.32 70.55B 40.46 73.32 134.9
DeepSeek V3Parameters 684.5B · Q8_0 678.9 684.5B 360.1 678.9 1276.6
Llama 3 8BParameters 8.03B · Q8_0 9.95 8.03B 6.21 9.95 16.96