Side-by-side, with the numbers that matter.

Two rigs, one table. Higher numbers win in cobalt; TDP and price flip.

Rig A

AWS · cloud-model

Change

Rig B

AMD · accelerator

Change

0 of 43 fields comparable

Spec	AWS Trainium2	AMD Instinct MI355X
Bf16 Matrix Pflops Dense	—	2.5
Bf16 Matrix Pflops Sparse	—	5
Bf16 Tflops PER Chip	667	—
Chips PER Trn2 Instance	16	—
Chips PER Ultraserver	64	—
Compute Units	—	256
Cooling	—	direct liquid cooling (air-cooled option available)
Form factor	—	OAM
Fp16 Matrix Pflops Dense	—	2.5
Fp16 Matrix Pflops Sparse	—	5
Fp32 Vector Tflops	—	157.3
Fp64 Matrix Tflops	—	78.6
FP8 Dense Petaflops PER Chip	1.3	—
FP8 Matrix Pflops Dense	—	5
FP8 Matrix Pflops Sparse	—	10.1
FP8 Sparse Petaflops PER Chip	5.2	—
HBM (GB)	—	288
HBM Bandwidth TB S PER Chip	2.9	—
HBM Capacity GB PER Chip	96	—
HBM Type	HBM3	HBM3E
Helios Gpus PER Rack	—	72
Memory bandwidth (TB/s)	—	8
Mxfp4 Matrix Pflops	—	10.1
Mxfp6 Matrix Pflops	—	10.1
Neuroncores PER Chip	8	—
Pcie GEN	—	5
Process Node NM	—	3
Successor Cdna Version	—	5
Successor Family	—	AMD Instinct MI400 Series (MI455X)
Successor Launch	—	2026-07-23
Successor Mi455x Bandwidth TB S	—	23.3
Successor Mi455x Hbm4 GB	—	432
TBP Watts	—	1,400
Trainium3 GA	2025-12-02	—
Trainium3 HBM Bandwidth TB S	4.9	—
Trainium3 Hbm3e GB PER Chip	144	—
Trn2 Instance FP8 Petaflops	20.8	—
Trn2 Instance HBM TB	1.5	—
Trn3 Ultraserver FP8 Pflops	362	—
Trn3 Ultraserver MAX Chips	144	—
Ultraserver FP8 Petaflops	83.2	—
Ultraserver Total HBM Bandwidth TB S	185	—
Ultraserver Total HBM TB	6	—