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	AMD MI250X	AMD MI210	Google TPU v4	Alibaba Hanguang 800	AWS Trainium
Architecture	CDNA 2	CDNA 2	TPU v4	XuanTie	Inferentia/Trainium
Form Factor	OAM	PCIe	Mezzanine	—	—
VRAM	128 GB	64 GB	—	—	32 GB
Memory Bandwidth	3,277 GB/s	1,638 GB/s	—	—	—
TFLOPs (FP32)	47.9	45.3	—	—	—
TFLOPs (FP16)	383	181	—	—	—
TFLOPs	383	181	275	—	190
TFLOPs (FP8)	—	—	—	—	—
TDP	560 W	300 W	300 W	150 W	200 W
Launch Date	Nov 2021	Jan 2022	May 2021	Sep 2019	Nov 2021

Efficiency Metrics

Metric	MI250X	MI210	TPU v4	Hanguang 800	Trainium
TFLOPs per Watt (FP32-eq)	0.34	0.30	0.46	—	0.47
Memory Bandwidth per GB	25.6 GB/s	25.6 GB/s	—	—	—

Performance Equivalence

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

To match 1x AMD MI250X

AMD MI210

Compute (FP32-eq)

2.12x

Need 2.12x MI210

FP32 Compute

1.06x

Need 1.06x MI210

VRAM

2.00x

Need 2.00x MI210

Memory Bandwidth

2.00x

Need 2.00x MI210

Google TPU v4

Compute (FP32-eq)

1.39x

Need 1.39x TPU v4

AWS Trainium

Compute (FP32-eq)

2.02x

Need 2.02x Trainium

VRAM

4.00x

Need 4.00x Trainium

To match 1x AMD MI210

AMD MI250X

Compute (FP32-eq)

0.47x

MI250X is 2.12x faster

FP32 Compute

0.95x

MI250X is 1.06x faster

VRAM

0.50x

MI250X has 2.00x more

Memory Bandwidth

0.50x

MI250X has 2.00x more

Google TPU v4

Compute (FP32-eq)

0.66x

TPU v4 is 1.52x faster

AWS Trainium

Compute (FP32-eq)

0.95x

Trainium is 1.05x faster

VRAM

2.00x

Need 2.00x Trainium

To match 1x Google TPU v4

AMD MI250X

Compute (FP32-eq)

0.72x

MI250X is 1.39x faster

AMD MI210

Compute (FP32-eq)

1.52x

Need 1.52x MI210

AWS Trainium

Compute (FP32-eq)

1.45x

Need 1.45x Trainium

To match 1x AWS Trainium

AMD MI250X

Compute (FP32-eq)

0.50x

MI250X is 2.02x faster

VRAM

0.25x

MI250X has 4.00x more

AMD MI210

Compute (FP32-eq)

1.05x

Need 1.05x MI210

VRAM

0.50x

MI210 has 2.00x more

Google TPU v4

Compute (FP32-eq)

0.69x

TPU v4 is 1.45x faster

Pricing

Price Type	MI250X	MI210	TPU v4	Hanguang 800	Trainium
CAPEX (Street Price)	$12,000	$6,000	—	—	—
OPEX (per hour)	$2.00/hr	—	$3.00/hr	—	—
Price per TFLOPs (FP32-eq)	$63	$66	—	—	—