GlobeGPU Cloud GPU price index

← GPU database

NVIDIA · Data centre

H100

Rented by 4 providers. 4 exact variants are priced separately, because they are different silicon.

Best self-service rate
$2.69
per GPU / hour
published price; capacity may vary

Every published rate

cheapest first · USD per GPU per hour
ProviderWhat they call itTheir own price, and the arithmeticPer GPU-hour
RunPod
SXM · 80 GB
H100 SXM
NVIDIA H100 80GB HBM3
$2.69 per hour = $2.6900 per GPU-hour
region Secure Cloud · 1 offer low stock
$2.6900
$2.72 per PFLOP-hr
Vultr
80 GB
H100
vbm-112c-2048gb-8-h100-gpu
$23.92 per hour ÷ 8 GPUs = $2.9900 per GPU-hour
regions not published in this feed · 1 offer availability unknown
$2.9900
$3.02 per PFLOP-hr
Scaleway
PCIe · 80 GB
H100 PCIe
H100-2-80G
0.09555 EUR per 1 minute → 5.7330 EUR/hr × 1.1645 EUR/USD ÷ 2 GPUs = $3.3380 per GPU-hour
region fr-par-2 · 2 offers availability unknown
$3.3380
$4.41 per PFLOP-hr
Scaleway
SXM · 80 GB
H100 SXM
H100-SXM-8-80G
0.42218 EUR per 1 minute → 25.3308 EUR/hr × 1.1645 EUR/USD ÷ 8 GPUs = $3.6872 per GPU-hour
region fr-par-2 · 3 offers availability unknown
$3.6872
$2.72 per PFLOP-hr
Scaleway
form factor and memory not stated
H100
EM-T620E-H100
9.589 EUR per 1 hour → 9.5890 EUR/hr × 1.1645 EUR/USD ÷ 2 GPUs = $5.5832 per GPU-hour
region fr-par-2 · 1 offer availability unknown
$5.5832
Azure
form factor and memory not stated
H100
Standard_NC40ads_H100_v5
$6.98 per hour = $6.9800 per GPU-hour
region US East · 450 offers availability unknown spot $1.29
$6.9800

The silicon

vendor-published
VariantArchitectureMemoryFP16 denseBoard power
H100 Hopper 2022 80 GB HBM3 989.4 TFLOPS 700 W
H100 PCIe Hopper 2022 80 GB HBM2e 756.5 TFLOPS 350 W
H100 SXM Hopper 2022 80 GB HBM3 989.4 TFLOPS 700 W

Vendor-published figures. FP16 throughput is dense tensor throughput with no sparsity assumed; vendors often headline the sparse figure, which is twice as large. Vendor page ↗
1 of this chip's 4 variants carry no vendor specification here — they are priced all the same, and simply say nothing where a figure would go.

What they call it

Providers write the same hardware down differently. Every name below is one this index has seen for this chip; nothing was renamed to make the table tidier.

H100H100 PCIeH100 SXM