Compare XPUs

Select up to 5 XPUs to compare side-by-side

Select XPUs to Compare

Clear all (5)Jump to results

Filter by Vendor

Showing 72 XPUs • 5 selected

Alibaba

Hanguang 800

AMD

MI100

23.1 TFLOPs

AMD

MI210

181 TFLOPs

AMD

MI250X

383 TFLOPs

AMD

MI300A

980.6 TFLOPs

AMD

MI300X

1,307 TFLOPs

AMD

MI325X

1,400 TFLOPs

AMD

MI350X

2,100 TFLOPs

AMD

Radeon PRO W7900

122 TFLOPs

AMD

Radeon RX 7900 XT

104 TFLOPs

AMD

Radeon RX 7900 XTX

122 TFLOPs

AWS

Inferentia2

190 TFLOPs

AWS

Trainium

190 TFLOPs

AWS

Trainium2

680 TFLOPs

Baidu

Kunlun II

Biren Technology

BR100

Cambricon

MLU370

256 TFLOPs

Cerebras

WSE-3

Enflame Technology

CloudBlazer T20

Etched

Sohu

10,000 TFLOPs

FuriosaAI

Warboy

Google

TPU v4

275 TFLOPs

Google

TPU v5e

197 TFLOPs

Google

TPU v5p

459 TFLOPs

Google

TPU v6e (Trillium)

918 TFLOPs

Graphcore

Bow IPU

Graphcore

IPU-M2000

Groq

LPU Inference Engine

Huawei

Ascend 910B

Iluvatar CoreX

BI-V150

300 TFLOPs

Intel

Data Center GPU Max 1100

177 TFLOPs

Intel

Data Center GPU Max 1550

419 TFLOPs

Intel Habana

Gaudi 2

432 TFLOPs

Intel Habana

Gaudi 3

1,835 TFLOPs

Multi-Metric Comparison

Relative performance across 5 key metrics (normalized to 100 = best in comparison)

Compute Performance (BF16)

Memory Capacity

Power Consumption

Power Efficiency

Specifications

Specification	NVIDIA H200 SXM	AMD MI210	AWS Trainium2	Rebellions ATOM	NVIDIA GB200 Superchip
Architecture	Hopper	CDNA 2	Trainium Gen2	Rebellions NPU	Blackwell
Form Factor	SXM	PCIe	—	—	SXM
VRAM	141 GB	64 GB	—	—	384 GB
Memory Bandwidth	4,800 GB/s	1,638 GB/s	—	—	8,000 GB/s
TFLOPs (FP32)	67	45.3	—	—	90
TFLOPs (FP16)	1,979	181	—	—	5,000
TFLOPs	1,979	181	680	—	5,000
TFLOPs (FP8)	3,958	—	—	—	—
TDP	700 W	300 W	300 W	200 W	2700 W
Launch Date	Nov 2023	Jan 2022	Nov 2023	Jun 2023	Mar 2024

Efficiency Metrics

Metric	H200 SXM	MI210	Trainium2	ATOM	GB200 Superchip
TFLOPs per Watt (FP32-eq)	1.41	0.30	1.13	—	0.93
Memory Bandwidth per GB	34.0 GB/s	25.6 GB/s	—	—	20.8 GB/s

Performance Equivalence

How many units of each GPU are needed to match the performance of the others?

To match 1x NVIDIA H200 SXM

AMD MI210

Compute (FP32-eq)

10.93x

Need 10.93x MI210

FP32 Compute

1.48x

Need 1.48x MI210

VRAM

2.20x

Need 2.20x MI210

Memory Bandwidth

2.93x

Need 2.93x MI210

AWS Trainium2

Compute (FP32-eq)

2.91x

Need 2.91x Trainium2

NVIDIA GB200 Superchip

Compute (FP32-eq)

0.40x

GB200 Superchip is 2.53x faster

FP32 Compute

0.74x

GB200 Superchip is 1.34x faster

VRAM

0.37x

GB200 Superchip has 2.72x more

Memory Bandwidth

0.60x

GB200 Superchip has 1.67x more

To match 1x AMD MI210

NVIDIA H200 SXM

Compute (FP32-eq)

0.09x

H200 SXM is 10.93x faster

FP32 Compute

0.68x

H200 SXM is 1.48x faster

VRAM

0.45x

H200 SXM has 2.20x more

Memory Bandwidth

0.34x

H200 SXM has 2.93x more

AWS Trainium2

Compute (FP32-eq)

0.27x

Trainium2 is 3.76x faster

NVIDIA GB200 Superchip

Compute (FP32-eq)

0.04x

GB200 Superchip is 27.62x faster

FP32 Compute

0.50x

GB200 Superchip is 1.99x faster

VRAM

0.17x

GB200 Superchip has 6.00x more

Memory Bandwidth

0.20x

GB200 Superchip has 4.88x more

To match 1x AWS Trainium2

NVIDIA H200 SXM

Compute (FP32-eq)

0.34x

H200 SXM is 2.91x faster

AMD MI210

Compute (FP32-eq)

3.76x

Need 3.76x MI210

NVIDIA GB200 Superchip

Compute (FP32-eq)

0.14x

GB200 Superchip is 7.35x faster

To match 1x NVIDIA GB200 Superchip

NVIDIA H200 SXM

Compute (FP32-eq)

2.53x

Need 2.53x H200 SXM

FP32 Compute

1.34x

Need 1.34x H200 SXM

VRAM

2.72x

Need 2.72x H200 SXM

Memory Bandwidth

1.67x

Need 1.67x H200 SXM

AMD MI210

Compute (FP32-eq)

27.62x

Need 27.62x MI210

FP32 Compute

1.99x

Need 1.99x MI210

VRAM

6.00x

Need 6.00x MI210

Memory Bandwidth

4.88x

Need 4.88x MI210

AWS Trainium2

Compute (FP32-eq)

7.35x

Need 7.35x Trainium2

Pricing

Price Type	H200 SXM	MI210	Trainium2	ATOM	GB200 Superchip
CAPEX (Street Price)	$40,000	$6,000	—	—	—
OPEX (per hour)	—	—	$1.34/hr	—	—
Price per TFLOPs (FP32-eq)	$40	$66	—	—	—