NVIDIA GB200

Data center GPU · Blackwell

Summary

The GB200 system combines NVIDIA's "Grace" CPU with B200 Blackwell GPUs. Deployed in NVL72 configurations (72 Blackwell GPUs; 36 Grace CPUs). Like the GB300, some cloud providers make these available in multiples of 4 GPUs.

Launched
Q1 2025
VRAM
186 GB
Mem. bandwidth
8,000 GB/s
On-demand from
Compare prices from 2 providers →

Tech specs

VRAM 186 GB HBM3e
Memory bandwidth 8,000 GB/s
CUDA cores 18,944
Tensor cores 592 (Gen 5)
TDP 1200 W
Supported data types
FP64FP32TF32FP16BF16FP8FP6FP4INT8

Technical specifications per GPU; for whole system specs see manufacturer documentation.

Cloud rental prices

2 providers · available in 1 country · lowest price per GPU, per hour

Prices updated

On-demand
$10.50 – $16.00 /GPU/h
Provider On-demand Reserved Spot
CoreWeave $10.50
Oracle $16.00
Compare all NVIDIA GB200 cloud providers & configurations →

Models that fit in VRAM

Open-weights models that fit in VRAM. Estimated using 🤗 accelerate, plus approximation for up to 8K context.

Media

Peak theoretical performance

FP4 Tensor Core 10,000 TFLOPS
FP6 Tensor Core 5,000 TFLOPS
FP8 Tensor Core 5,000 TFLOPS
INT8 Tensor Core 5,000 TOPS
BF16 Tensor Core 2,500 TFLOPS
FP16 Tensor Core 2,500 TFLOPS
TF32 Tensor Core 1,250 TFLOPS
FP32 80 TFLOPS
FP64 40 TFLOPS

Performance figures assume no sparsity; in cases where only sparse performance figures are published by the manufacturer, these are halved to give approximate dense performance.

Resources

Detailed documentation from the manufacturer.

Similar GPUs

Other accelerators you might compare.