Blackwell Architecture
NVIDIA GB200
NVIDIA GB200 Grace Blackwell Superchip — Rent GPU servers for AI training, inference, and HPC.
Reserved Clusters Only · Up to 45% Savings
Memory
192 GB
HBM3e
FP32
90
TFLOPS
Bandwidth
8000
GB/s
Avg. Price
$3.75
per hour
Hardware · Specifications
Full specs.
ManufacturerNVIDIA
ArchitectureBlackwell
Release Year2024
VRAM192 GB HBM3e
Memory Bandwidth8000 GB/s
FP32 Performance90 TFLOPS
FP16 Performance180 TFLOPS
INT8 Performance3600 TOPS
Tensor Cores18432
CUDA/Stream ProcessorsN/A
TDP1000W
Form FactorSuperchip
NVLink SupportYes
Max GPUs per Node8
Compute · Performance
Performance overview.
FP32 Single Precision
90
TFLOPS
FP16 Half Precision
180
TFLOPS
INT8 Integer Ops
3600
TOPS
Workloads · Use Cases
Ideal for.
Large Language Models
AI Supercomputing
Enterprise AI
Multi-Modal Training
Cost · Pricing
Transparent pricing.
Hourly$3.75/hr
Daily$90.00/day
Monthly$2737.50/mo
Platform · Why Runcrate
Why reserve NVIDIA GB200.
Guaranteed Capacity
Reserve NVIDIA GB200 clusters from 8-512 GPUs with guaranteed availability.
Up to 45% Savings
Reserved clusters offer significant savings. Lock in lower rates for your infrastructure.
Enterprise Ready
99.9% uptime guarantee. 24/7 support. Production-grade infrastructure.
Support · FAQ