H100 SXM — specs & cloud pricing
The H100 SXM is a datacenter accelerator from NVIDIA. Live specs and modeled cloud $/hr from the Mintok silicon catalog.
Cloud rate
$5.50per chip-hour (modeled)
Specifications
- Vendor
- NVIDIA
- Class
- Datacenter accelerator
- Architecture
- Hopper
- BF16 throughput
- 1,979 TFLOPS
- On-package memory
- 80 GB
- Memory bandwidth
- 3,350 GB/s
- TDP
- 700 W
More NVIDIA silicon
- Vera Rubin35,000 TFLOPS · 288GB
- Rubin GPU8,000 TFLOPS · 288GB
- GB300 NVL727,000 TFLOPS · 288GB
- Rubin Ultra VR3005,000 TFLOPS · 1,024GB
- B200 SXM4,500 TFLOPS · 192GB
- Rubin VR2002,500 TFLOPS · 288GB
FAQ
How much does the H100 SXM cost per hour in the cloud?
The H100 SXM rents for about $5.50 per chip-hour on cloud providers (modeled from public rates). Own-vs-rent economics — capex, lease and TCO — are in the Mintok sizing report and rack teardowns.
How much memory does the H100 SXM have?
The H100 SXM has 80 GB of on-package memory at 3,350 GB/s bandwidth.
What is the H100 SXM's BF16 compute throughput?
The H100 SXM delivers 1,979 BF16 TFLOPS at a 700 W TDP.
Is the H100 SXM a datacenter or edge accelerator?
The H100 SXM is a datacenter accelerator (700 W TDP).
Cost this chip in a fleet
Size a deployment on the H100 SXM, or see the full priced rack BOM it sits in.
Specs and modeled rates maintained by Mintok's catalog watchers from public sources. Not investment advice. Methodology.