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

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

Rig A

NVIDIA · workstation

Change

Rig B

AWS · cloud-model

Change

0 of 39 fields comparable

Spec	NVIDIA DGX Spark	AWS Trainium2
Bf16 Tflops PER Chip	—	667
Chips PER Trn2 Instance	—	16
Chips PER Ultraserver	—	64
CPU Breakdown	10x Arm Cortex-X925 + 10x Arm Cortex-A725	—
CPU Cores Total	20	—
Dimensions MM	150 x 150 x 50.5	—
Ethernet	1x 10 GbE RJ-45	—
FP4 Pflops Sparse	1	—
FP8 Dense Petaflops PER Chip	—	1.3
FP8 Sparse Petaflops PER Chip	—	5.2
Gb10 TDP Watts	140	—
GPU Arch	Blackwell, 5th-gen Tensor Cores, 4th-gen RT Cores	—
HBM Bandwidth TB S PER Chip	—	2.9
HBM Capacity GB PER Chip	—	96
HBM Type	—	HBM3
Launch Price USD	3,999	—
Memory bandwidth (GB/s)	273	—
Memory Interface BIT	256	—
Memory type	LPDDR5X	—
MSRP (USD)↓ better	4,699	—
Neuroncores PER Chip	—	8
NIC	ConnectX-7 @ 200 Gbps	—
Price Change Effective	2026-02-23	—
PSU Watts	240	—
Storage	4TB NVMe M.2 self-encrypting	—
Superchip	NVIDIA GB10 Grace Blackwell	—
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
Unified Memory GB	128	—
Weight KG	1.2	—
Wireless	Wi-Fi 7, Bluetooth 5.4	—