NVIDIA B300

Data center GPU · Blackwell · Ultra

Summary

The NVIDIA B300 is an evolution of the B200 GPU, and comes with significantly higher FP4 performance, making it a powerhouse for inference on quantised models. The amount of VRAM is configurable, and can vary between providers.

The B300 is also available in its GB300 "Grace Blackwell" configuration, pairing B300 GPUs with NVIDIA's Grace CPU.

Launched
Q3 2025
VRAM
288 GB
Mem. bandwidth
7,700 GB/s
On-demand from
$7.50 /GPU/h
Compare prices from 4 providers →

Tech specs

VRAM 288 GB HBM3e
Memory bandwidth 7,700 GB/s
Interface SXM
CUDA cores 20,480
Tensor cores 640 (Gen 5)
TDP 1100 W
Supported data types
FP64FP32FP16BF16FP8FP6FP4INT8

Cloud rental prices

4 providers · available in 2 countries · lowest price per GPU, per hour

Prices updated

On-demand
$7.50 – $17.80 /GPU/h
Reserved
$6.90 – $7.35 /GPU/h
Spot
$2.63 /GPU/h
Provider On-demand Reserved Spot
AWS $17.80 — —
Oracle $15.00 — —
Scaleway $8.58 — —
Verda $7.50 $6.90 $2.63
Compare all NVIDIA B300 cloud providers & configurations →

Models that fit in VRAM

Total VRAM From 16-bit inference 8-bit inference 4-bit inference
1× B300 288 GB $6.90/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× B300 576 GB $13.80/h MiniMax-M2.7DeepSeek-V4-Flashgpt-oss-120b MiniMax-M2.7DeepSeek-V4-Flashgpt-oss-120b DeepSeek-V4-ProMiniMax-M2.7gpt-oss-120b
4× B300 1,152 GB $27.60/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× B300 2,304 GB $55.20/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

CoreWeave Advances AI-Native Cloud Platform for the Next Phase of Production-Scale AI (NVIDIA HGX B300 now GA) CoreWeave · 16 Mar 2026 Lilly and NVIDIA unveil $1B co-innovation lab Eli Lilly · Drug Discovery & Development · 14 Jan 2026

Peak theoretical performance

FP4 Tensor Core 1,400 TFLOPS
FP6 Tensor Core 4,500 TFLOPS
FP8 Tensor Core 4,500 TFLOPS
INT8 Tensor Core 153 TOPS
BF16 Tensor Core 2,250 TFLOPS
FP16 Tensor Core 2,250 TFLOPS
TF32 Tensor Core 1,100 TFLOPS
FP32 75 TFLOPS
FP64 1.2 TFLOPS
FP64 Tensor Core 1.2 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 GB300 View specs & prices → NVIDIA B200 View specs & prices → AMD MI355X View specs & prices →