NVIDIA · Blackwell · 2024
Grace + Blackwell unified-memory superchip — frontier-model training at the largest scale.
Cloud rental
On-demand · per-second billing
$3.50–$4.00/hr range
Or $2.63/hr reserved (30% off)
Pricing across clouds
Same GPU, cheapest-first. Prices reflect publicly listed hourly rates for NVIDIA GB200 on each provider. Runcrate is the lowest published rate.
Workloads
Real workloads sized for the NVIDIA GB200, with concrete performance numbers. Click to deploy preconfigured.
Deploy
Skip the dashboard if you don't need it. SDK, Python, or cURL — copy the snippet, paste your API key, ship.
import Runcrate from "@runcrate/sdk";
const rc = new Runcrate({ apiKey: "rc_live_••••••••••••••••" });
const instance = await rc.instances.create({
gpu: "gb200",
region: "auto",
image: "runcrate/vllm:latest",
});
console.log(`SSH: ssh root@${instance.host}`);Benchmarks
Side-by-side specs vs the closest alternatives. Bars are normalized to the highest value in each metric.
FP16
VRAM
Bandwidth
Alternatives
Same Runcrate platform, different price/performance. Hover any GPU to switch.
Blackwell · 192GB HBM3e · 180 TFLOPS FP16
Blackwell · 192GB HBM3e · 180 TFLOPS FP16
Blackwell · 256GB HBM3e · 240 TFLOPS FP16
Hopper · 141GB HBM3e · 134 TFLOPS FP16
Hopper · 80GB HBM3 · 120 TFLOPS FP16
Full specs
Memory
Compute
Power & form factor
Cluster
FAQ
Ready when you are
Reserved capacity available · per-second billing · contact us for an immediate quote.
Other GPUs
NVIDIA Blackwell
NVIDIA Ampere
NVIDIA Ada Lovelace