8 × HGX H100 · 80 GB per GPU · 128 vCPU · 2,048 GB RAM · 61.44 TB local storage. North America node price. The page’s $6.16 single-GPU inference price is restricted to inference-platform customers.
Source ↗Node and network
Eight GPUs do not make every node equivalent.
Keep the number of GPUs, memory and between-node network in the same quote.
8-GPU node
8-GPU node
Illustration of node count only. Link width does not represent speed.
Documented node scope
CoreWeave
On-demand node
$49.24 / node-hour
Together AI
On-demand cluster
$3.99 / GPU-hour
H100 · 80 GB per GPU · listed cluster scale starts at 8 GPUs. Do not treat as a single GPU rental. Live availability varies.
Source ↗AWS
P5 family
Up to 8 H100 GPUs · 640 GB total GPU memory
EC2 P5, not P5e/P5en H200 variants. Exact instance, region and price not captured.
Source ↗Google Cloud
A3 High 8GPU
8 × H100 · 640 GB total GPU memory
a3-highgpu-8g · 208 vCPU · 1,872 GB RAM · 6,000 GiB local SSD. Up to 1000 Gbps network. Region/price/stock unknown.
Source ↗Between-node network
AWS
P5 network
Up to 3,200 Gbps EFA
P5/P5e/P5en. Vendor maximum; not measured throughput or one GPU network guarantee.
Source ↗Crusoe
Cluster networking
InfiniBand and Ethernet; zero egress fees claimed
Cloud networking statement. Not a network benchmark or an SLA.
Source ↗