Skip to content
14 GPUs · Updated May 2026

Will My Model Fit?

Select any model and precision — instantly see which of 14 data center GPUs can run it, how many you need, and how much VRAM is consumed. No throughput targets, no cloud setup. Just a fast VRAM check.

Best quality · 2 bytes/param · includes ~10% KV cache overhead

VRAM Required
154GB
Single-GPU Fit
6/ 14 GPUs
Cheapest Single GPU
B200 SXM (192GB)
NVIDIA
T4
Turing · 16GB VRAM
154GB needed16GB max
Needs 10+ GPUs
NVIDIA
A10
Ampere · 24GB VRAM
154GB needed24GB max
Needs 7× GPUs
NVIDIA
A100 40GB
Ampere · 40GB VRAM
154GB needed40GB max
Needs 4× GPUs
NVIDIA
L40S
Ada · 48GB VRAM
154GB needed48GB max
Needs 4× GPUs
NVIDIA
A100 80GB
Ampere · 80GB VRAM
154GB needed80GB max
Needs 2× GPUs
NVIDIA
H100 SXM5
Hopper · 80GB VRAM
154GB needed80GB max
Needs 2× GPUs
GOOGLE
TPU v5p
v5p · 96GB VRAM
154GB needed96GB max
Needs 2× GPUs
NVIDIA
H200 SXM
Hopper · 141GB VRAM
154GB needed141GB max
Needs 2× GPUs
NVIDIA
B200 SXM
Blackwell · 192GB VRAM
154GB needed192GB max
Fits — 1 GPU
AMD
MI300X
CDNA 3 · 192GB VRAM
154GB needed192GB max
Fits — 1 GPU
GOOGLE
TPU v7
Ironwood · 192GB VRAM
154GB needed192GB max
Fits — 1 GPU
AMD
MI325X
CDNA 3.5 · 256GB VRAM
154GB needed256GB max
Fits — 1 GPU
NVIDIA
B300 Ultra
Blackwell · 288GB VRAM
154GB needed288GB max
Fits — 1 GPU
AMD
MI355X
CDNA 4 · 288GB VRAM
154GB needed288GB max
Fits — 1 GPU

Multi-GPU Configurations for Llama 3.1 70B

Single GPU
6GPU types
+6 unlocked at 1×
2× GPUs
10GPU types
+4 unlocked at 2×
4× GPUs
12GPU types
+2 unlocked at 4×
8× GPUs
13GPU types

How VRAM Is Calculated

FP16

2 bytes per parameter — full quality, highest VRAM

FP8

1 byte per parameter — near-identical quality to FP16

INT8

1 byte per parameter — minor quality loss, widely supported

INT4

0.5 bytes per parameter — 4-bit GGUF / AWQ / GPTQ quantization

Formula: VRAM = (Parameters × bytes/param) × 1.1 — the 10% overhead accounts for KV cache during typical inference. Training requires 3–4× this amount due to optimizer states and gradients.

Common Model VRAM Reference

ModelParamsFP16FP8 / INT8INT4
Mistral 7B7B15GB8GB4GB
Llama 3.1 8B8B18GB9GB5GB
Gemma 3 27B27B59GB30GB15GB
Llama 3.1 70B70B154GB77GB39GB
Qwen 2.5 72B72B158GB79GB40GB
Llama 4 Scout 109B109B240GB120GB60GB
Llama 3.1 405B405B891GB446GB223GB
DeepSeek R1 671B671B1.4TB738GB369GB