Trainium2 specs
AWS · Shipping now · first available Dec 2024 (estimated). Dense compute, memory, bandwidth and power from the vendor's documents; each precision stays separate.
All accelerators · Compare with Trainium3
Specifications
| Dense BF16 · FP32 accumulate | 667 TFLOPS |
|---|---|
| Dense FP8 · FP32 accumulate | 1,299 TFLOPS |
| Dense FP4 · FP32 accumulate | Not published |
| Memory | 103.08 GB · HBM96 GiB as published |
| Memory bandwidth | 2.9 TB/s |
| Maximum chip power | 500 West — estimated / secondary-source figure |
| Interconnect | NeuronLink v3 · 1.28 TB/s |
| First availability | Dec 2024est — estimated / secondary-source figure |
| Availability | Shipping |
AWS Neuron publishes dense BF16/FP16/TF32 and FP8 separately from sparse throughput. Board power remains an earlier estimate.
Dense tensor rates with FP32 accumulation, never sparse or FP16-accumulate headline figures. Memory is in decimal GB. The est marker flags a value its vendor-document review did not confirm.
Rent it
$2.24/chip-hrCapacity Blocks · AWS (trn2) · observed
No on-demand quote is tracked; this price is a different billing model and is not compared with on-demand rates.
Trainium2 rental prices →Value ratios
| $ per dense BF16 PFLOP-hour | No on-demand quote |
|---|---|
| Dense BF16 TFLOPS per watt | 1.33 |
| Bandwidth bytes per 1,000 BF16 FLOP | 4.35 |
| $ per dense FP8 PFLOP-hour | No on-demand quote |
| Dense FP8 TFLOPS per watt | 2.60 |
| Bandwidth bytes per 1,000 FP8 FLOP | 2.23 |
| $ per GB of memory per hour | No on-demand quote |
Prices divide the cheapest on-demand list price; theoretical ratios do not measure application speed.
Sources
Specifications reviewed . Dense compute and per-chip memory/interconnect; the 500 W board power remains a secondary estimate.
Rental price: AWS (trn2) published pricing ↗ (trn2.48xlarge; 16 × Trainium2 chips).