DEPLOYDatabase

Chip comparison

AWS Trainium 2 vs NVIDIA GB200 Grace Blackwell Superchip

Side-by-side specs

Straight from each vendor's datasheet. Winner in bold: higher throughput, memory, or transistor count; lower power draw or price. Sparse (2:4) throughput roughly doubles the FP8/INT8 numbers shown.

FieldAWS Trainium 2NVIDIA GB200 Grace Blackwell Superchip
Process nodeTSMC N5TSMC 4NP
TDP500 W2700 W
Memory96 GB HBM3384 GB HBM3e
Memory bandwidth2,900 GB/s16,000 GB/s
BF16 (dense)667 TFLOPS
FP8 (dense)1,300 TFLOPS9,000 TFLOPS
INT8 (dense)1,300 TOPS
Form factorAWS Neuron 2 (Trn2 UltraServer with 64 chips per unit)Superchip (2xB200 + 1xGrace)
Announced2024-12-032024-03-18
Released2025-01-012024-12-01

Robots running AWS Trainium 2

No robots publicly running it yet.

Robots running NVIDIA GB200 Grace Blackwell Superchip

No robots publicly running it yet.

Common questions

AWS Trainium 2 vs NVIDIA GB200 Grace Blackwell Superchip: which is faster for training?

NVIDIA GB200 Grace Blackwell Superchip has 6.92x the dense FP8 throughput of the other (AWS Trainium 2: 1,300 TFLOPS; NVIDIA GB200 Grace Blackwell Superchip: 9,000 TFLOPS). Real-world training speed also depends on memory bandwidth, interconnect, and software stack: see the spec diff for those inputs.

AWS Trainium 2 vs NVIDIA GB200 Grace Blackwell Superchip: which has more memory?

NVIDIA GB200 Grace Blackwell Superchip carries more HBM (AWS Trainium 2: 96 GB HBM3; NVIDIA GB200 Grace Blackwell Superchip: 384 GB HBM3e). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.

AWS Trainium 2 vs NVIDIA GB200 Grace Blackwell Superchip: which is more power-efficient?

Dense FP8 TFLOPS per watt: AWS Trainium 2 2.60 (1,300 ÷ 500 W); NVIDIA GB200 Grace Blackwell Superchip 3.33 (9,000 ÷ 2700 W). NVIDIA GB200 Grace Blackwell Superchip wins at the die level; system-level efficiency also depends on cooling and interconnect.

AWS Trainium 2 vs NVIDIA GB200 Grace Blackwell Superchip: which is more widely used?

AWS Trainium 2: 0 robots and 3 data centers publicly running it. NVIDIA GB200 Grace Blackwell Superchip: 0 robots and 11 data centers. Empty here means no operator has said publicly, not zero adoption in the market.

See also: every chip comparison · AWS Trainium 2 full page · NVIDIA GB200 Grace Blackwell Superchip full page.