Hopper In stock MIG capable

NVIDIA H100 SXM5

The workhorse for multi-node LLM training. NVLink across all 8 GPUs.

Best suited to

LLM pre-training
Multi-node distributed jobs
Large batch fine-tuning

Available regions

Region	Location	Typical latency
us-east-1	US East (Ashburn, VA)	12 ms
us-west-2	US West (Hillsboro, OR)	38 ms
eu-west-1	EU West (Dublin, IE)	84 ms

Hourly rate

$2.49/hr

Billed per second. $59.76 for a full day, $1,818 if left running a month.

Deploy this GPU All pricing

Full specification

ArchitectureHopper
VRAM80 GB
Memory typeHBM3
Memory bandwidth3.35 TB/s
FP16 dense1,979 TFLOPS
FP8 dense3,958 TFLOPS
CUDA cores16,896
Tensor cores528 (4th gen)
Board power700W
InterconnectNVLink 900 GB/s
Host vCPU26
Host RAM200 GB
NVMe scratch1000 GB
Network3200 Gbps InfiniBand
MIG partitioningYes

Other Hopper cards

H200

141 GB HBM3e

H100 PCIe

80 GB HBM2e