NVIDIA GB300

Data center GPU · Blackwell · Grace

Summary

The GB300 is currently the most powerful NVIDIA system available, and combines NVIDIA's "Grace" CPU with B300 Blackwell Ultra GPUs.

Deployed in NVL72 configurations (72 Blackwell GPUs; 36 Grace CPUs). Like the GB200, some cloud providers make these available in multiples of 4 GPUs. The amount of memory per GPU in B300 systems is configurable, and can vary between providers.

Launched
Q3 2025
VRAM
288 GB
Mem. bandwidth
8,000 GB/s
On-demand from
$8.62 /GPU/h
Compare prices from 2 providers →

Tech specs

VRAM 288 GB HBM3e
Memory bandwidth 8,000 GB/s
CUDA cores 20,480
Tensor cores 640 (Gen 5)
TDP 1400 W
Supported data types
FP64FP32TF32FP16BF16FP8FP6FP4INT8

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

Cloud rental prices

2 providers · lowest price per GPU, per hour

Prices updated

On-demand
$8.62 – $18.00 /GPU/h
Reserved
$7.93 – $8.45 /GPU/h
Spot
$3.02 /GPU/h
Provider On-demand Reserved Spot
Oracle $18.00 — —
Verda $8.62 $7.93 $3.02
Compare all NVIDIA GB300 cloud providers & configurations →

Additional providers who may offer NVIDIA GB300 without public pricing: Vultr, Nebius

Models that fit in VRAM

Total VRAM From 16-bit inference 8-bit inference 4-bit inference
1× GB300 288 GB $7.93/h gpt-oss-120bNVIDIA-Nemotron-3-Super-120B-A12B-BF16GLM-4.5-Air MiniMax-M2.7DeepSeek-V4-Flashgpt-oss-120b MiniMax-M2.7DeepSeek-V4-Flashgpt-oss-120b
2× GB300 576 GB $15.86/h MiniMax-M2.7DeepSeek-V4-Flashgpt-oss-120b MiniMax-M2.7DeepSeek-V4-Flashgpt-oss-120b DeepSeek-V4-ProMiniMax-M2.7gpt-oss-120b
4× GB300 1,152 GB $31.72/h MiniMax-M2.7DeepSeek-V4-Flashgpt-oss-120b Kimi-K2-Instruct-0905DeepSeek-V4-ProMiniMax-M2.7 Kimi-K2-Instruct-0905DeepSeek-V4-ProMiniMax-M2.7
8× GB300 2,304 GB $63.44/h DeepSeek-V4-ProMiniMax-M2.7gpt-oss-120b Kimi-K2-Instruct-0905DeepSeek-V4-ProMiniMax-M2.7 Kimi-K2-Instruct-0905DeepSeek-V4-ProMiniMax-M2.7

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

Media

World's first at-scale GB300 NVL72 production cluster for OpenAI Microsoft Azure & OpenAI · NVIDIA Blog CoreWeave GB300 NVL72 instances deliver 6.5x DeepSeek-R1 inference CoreWeave · CoreWeave GB300 NVL72 deployed at scale for agentic coding assistants Oracle · NVIDIA Blog

Peak theoretical performance

FP4 Tensor Core 15,000 TFLOPS
FP6 Tensor Core 5,000 TFLOPS
FP8 Tensor Core 5,000 TFLOPS
INT8 Tensor Core 165 TOPS
BF16 Tensor Core 2,500 TFLOPS
FP16 Tensor Core 2,500 TFLOPS
TF32 Tensor Core 1,250 TFLOPS
FP32 80 TFLOPS
FP64 1.3 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.

Datasheet ↗ Official spec sheet Product page ↗ Manufacturer overview

Similar GPUs

Other accelerators you might compare.

NVIDIA B300 View specs & prices → NVIDIA GB200 View specs & prices →