VRAM & Model Sizer
Estimate GPU memory for LLM weights. Uses only published parameter counts and VRAM specs from our database — no invented figures.
Formula
Weights memory (GB) = parameters (billions) × bytes per parameter
Bytes per parameter: FP16 / BF16 = 2 · Q8 / INT8 = 1 · Q4 / INT4 = 0.5
Weight format
8B params × 2 bytes/param (FP16 / BF16)
Estimated weight memory only. Real deployments need extra memory for the KV cache and runtime overhead, which grow with context length and vary by serving stack. Leave headroom above this figure.
Fits the weights (before overhead)
RTX 5080
16 GB VRAM+0.0 GB headroom
RTX 5070 Ti
16 GB VRAM+0.0 GB headroom
RTX 4080
16 GB VRAM+0.0 GB headroom
T4
16 GB VRAM+0.0 GB headroom
V100 16GB
16 GB VRAM+0.0 GB headroom
RTX 5060 Ti
16 GB VRAM+0.0 GB headroom
RTX 2000 Ada
16 GB VRAM+0.0 GB headroom
Tesla P100 16GB
16 GB VRAM+0.0 GB headroom
RTX A4000 16GB
16 GB VRAM+0.0 GB headroom
Quadro P5000 16GB
16 GB VRAM+0.0 GB headroom
Quadro RTX 5000 16GB
16 GB VRAM+0.0 GB headroom
RTX 4060 Ti 16GB
16 GB VRAM+0.0 GB headroom
RTX 4000 Ada
20 GB VRAM+4.0 GB headroom
RTX 4000 SFF Ada
20 GB VRAM+4.0 GB headroom
RTX A4500 20GB
20 GB VRAM+4.0 GB headroom
L4
24 GB VRAM+8.0 GB headroom
A30
24 GB VRAM+8.0 GB headroom
A10G
24 GB VRAM+8.0 GB headroom
RTX 4090
24 GB VRAM+8.0 GB headroom
RTX 3090
24 GB VRAM+8.0 GB headroom
RTX 4500 Ada
24 GB VRAM+8.0 GB headroom
RTX A5000
24 GB VRAM+8.0 GB headroom
A10 24GB
24 GB VRAM+8.0 GB headroom
RTX A5500 24GB
24 GB VRAM+8.0 GB headroom
Quadro P6000 24GB
24 GB VRAM+8.0 GB headroom
Quadro RTX 6000 24GB
24 GB VRAM+8.0 GB headroom
RTX 5090
32 GB VRAM+16.0 GB headroom
V100 32GB
32 GB VRAM+16.0 GB headroom
RTX 5000 Ada
32 GB VRAM+16.0 GB headroom
H100 40GB
40 GB VRAM+24.0 GB headroom
A100 40GB
40 GB VRAM+24.0 GB headroom
A100 PCIe 40GB
40 GB VRAM+24.0 GB headroom
A100 SXM4 40GB
40 GB VRAM+24.0 GB headroom
L40S
48 GB VRAM+32.0 GB headroom
L40
48 GB VRAM+32.0 GB headroom
A40
48 GB VRAM+32.0 GB headroom
RTX 6000 Ada
48 GB VRAM+32.0 GB headroom
RTX A6000
48 GB VRAM+32.0 GB headroom
RTX 5880 Ada
48 GB VRAM+32.0 GB headroom
Quadro RTX 8000 48GB
48 GB VRAM+32.0 GB headroom
A16
64 GB VRAM+48.0 GB headroom
H100 80GB
80 GB VRAM+64.0 GB headroom
A100 80GB
80 GB VRAM+64.0 GB headroom
H100 SXM5 80GB
80 GB VRAM+64.0 GB headroom
H100 PCIe 80GB
80 GB VRAM+64.0 GB headroom
A100 SXM4 80GB
80 GB VRAM+64.0 GB headroom
A100 PCIe 80GB
80 GB VRAM+64.0 GB headroom
H100 NVL 94GB
94 GB VRAM+78.0 GB headroom
RTX PRO 6000 Blackwell
96 GB VRAM+80.0 GB headroom
GH200 Grace Hopper 96GB
96 GB VRAM+80.0 GB headroom
Gaudi 2 96GB
96 GB VRAM+80.0 GB headroom
MI250X 128GB
128 GB VRAM+112.0 GB headroom
H200 141GB
141 GB VRAM+125.0 GB headroom
H200 SXM 141GB
141 GB VRAM+125.0 GB headroom
H200 NVL 141GB
141 GB VRAM+125.0 GB headroom
B200 SXM 180GB
180 GB VRAM+164.0 GB headroom
B200 SXM 180GB
180 GB VRAM+164.0 GB headroom
MI300X 192GB
192 GB VRAM+176.0 GB headroom
GB200 NVL72
192 GB VRAM+176.0 GB headroom
B200 NVL 192GB
192 GB VRAM+176.0 GB headroom
MI300X 192GB
192 GB VRAM+176.0 GB headroom
MI325X 256GB
256 GB VRAM+240.0 GB headroom
MI325X 256GB
256 GB VRAM+240.0 GB headroom
GB300 NVL72
288 GB VRAM+272.0 GB headroom
B300 SXM6
288 GB VRAM+272.0 GB headroom
GB300 SXM6
288 GB VRAM+272.0 GB headroom
MI355X
288 GB VRAM+272.0 GB headroom
B300 SXM 288GB
288 GB VRAM+272.0 GB headroom
GB300 SXM 288GB
288 GB VRAM+272.0 GB headroom
MI355X 288GB
288 GB VRAM+272.0 GB headroom
This tool uses only published parameter counts from our LLM model database and VRAM figures from our GPU specs database. Results are estimates, not measurements.