The first entry is always the pin's source. Overall confidence is a weighted average of how firmly each reference supports the start and end times used above; a reference counts half as much for every 180 days older than the newest.
Dated NVIDIA[2] press release (GTC, March 22, 2022).
Something missing or wrong? Say it in your own words: a link that backs this pin up, a different start or end date and why, or a fact it lacks or gets wrong. The AI checks it against this pin's sources, searches for better ones, and adds any page that backs you up. The pin's own sources still count most. A picture that shows something else, or shows it badly, is looked at too, and moved down or replaced.
| Up to 256 H100 GPUs at 9x the previous generation's bandwidth |
| Maker claim[1][2] |
| DGX H100 | 8 H100 GPUs, 32 petaflops of AI performance | Maker claim[1][2] |
|---|
| Availability | Expected worldwide later in 2022 | Maker plan[1][2] |
|---|
NVIDIA says H100 is in full production, with partner products planned from October[1][2]
Dated milestones from opened primary pages; cloud "first" claims are the provider's own.
192 GB HBM3 and 5.3 TB/s peak memory bandwidth on an OAM module with 750 W typical board power, per AMD specifications.
SourceEach TPU v5p pod composes 8,960 chips over a 4,800 Gbps-per-chip 3D torus interconnect, per Google Cloud.
SourceA Trn2 instance has 16 Trainium2 chips, 1.5 TB of HBM3 and up to 20.8 FP8 petaflops, per AWS.
Source
