MXFP8 per Trainium3 chip; 144 GB HBM3E at 4.9 TB/s; first 3nm AWS accelerator
2.52 PFLOPSobserved
| Value kind | observed — Reported measurements, counts, and specifications keep the precision and scope stated by their source; an exact specification is not treated as a range. |
|---|---|
| Scope | AWS-published peak per-chip specification, rounded to 2.52 PFLOPS; Neuron architecture documentation gives 2,517 MXFP8/MXFP4 TFLOPS. AWS's up-to-4x performance-per-watt comparison is at Trn3 UltraServer versus Trn2 UltraServer level, not per chip, and is an AWS claim rather than a measured workload gain. |
| As of | 2025-12 |
| Source | AWS Trainium3 launch announcement (2025-12-02) |
| Review | checking…review by 2027-01-31 · standard cadence |
| Recorded changes | last 2026-10-15 · 3 revisions tracked |
| Claim id | fp8-per-trainium3-chip-144-gb-hbm3e-4-9-tb-s |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.