How much does the GB300 cost per hour?

Runcrate GB300 pricing is $4.00-$4.50/hr (avg $4.25/hr) for reserved capacity. AWS lists $7.20, Azure $7.50, GCP $7.95 — Runcrate offers the lowest published GB300 rate. Reservation-only at launch.

What is the NVIDIA GB300?

GB300 is NVIDIA's flagship Blackwell datacenter GPU (also marketed as Blackwell Ultra) with 256GB HBM3e, 10 TB/s memory bandwidth, and 1.5 PFLOPS FP4 compute. It's the largest single-GPU memory footprint shipping today.

Is the GB300 worth it over GB200?

GB300 has 33% more VRAM (256 vs 192GB) and 25% more bandwidth — meaningful for frontier-model pretraining and trillion-parameter inference. For models under 192GB, GB200 is cheaper and more available.

When can I deploy a GB300 instance?

GB300 capacity rolls out through 2025. Reservation slots are open now; contact us for committed-term pricing and capacity guarantee.

runcrate

Contact Sales Console

NVIDIA · Blackwell · 2025

NVIDIA GB300.

Name: NVIDIA GB300 GPU Cloud Instance
Brand: NVIDIA
Price: 4.25 USD
Availability: PreOrder

256GB HBM3e and 10 TB/s — the most memory and bandwidth shipping in any single GPU.

256 GB

HBM3e

10.0 TB/s

Bandwidth

240

TFLOPS FP16

8× GPU

NVLink

Reserved capacity

Deploy in 60 seconds Reserved quote

Cloud rental

$4.25/hr

On-demand · per-second billing

$4.00–$4.50/hr range

Or $2.97/hr reserved (30% off)

Pricing across clouds

NVIDIA GB300 cloud rental price comparison.

Same GPU, cheapest-first. Prices reflect publicly listed hourly rates for NVIDIA GB300 on each provider. Runcrate is the lowest published rate.

RuncrateCheapest

$4.25/hr

AWS

$7.20/hr

Azure

$7.50/hr

GCP

$7.95/hr

Workloads

What you'll actually use this for.

Real workloads sized for the NVIDIA GB300, with concrete performance numbers. Click to deploy preconfigured.

1T Parameter Training

Single-GPU fits trillion-param shards

Deploy

Reasoning Model Training

Long-context RL at full bf16

Deploy

Frontier Multi-Modal

Image + video + text + audio combined

Deploy

Llama 405B Inference

Single GB300 fits with room for KV cache

Deploy

One command. NVIDIA GB300 in 60 seconds.

Skip the dashboard if you don't need it. SDK, Python, or cURL — copy the snippet, paste your API key, ship.

import Runcrate from "@runcrate/sdk";

const rc = new Runcrate({ apiKey: "rc_live_••••••••••••••••" });

const instance = await rc.instances.create({
  gpu: "gb300",
  region: "auto",
  image: "runcrate/vllm:latest",
});

console.log(`SSH: ssh root@${instance.host}`);