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 L4	AMD MI250X	AWS Trainium2	Rebellions ATOM	NVIDIA H100 SXM
Architecture	Ada Lovelace	CDNA 2	Trainium Gen2	Rebellions NPU	Hopper
Form Factor	PCIe	OAM	—	—	SXM
VRAM	24 GB	128 GB	—	—	80 GB
Memory Bandwidth	300 GB/s	3,277 GB/s	—	—	3,350 GB/s
TFLOPs (FP32)	30.3	47.9	—	—	67
TFLOPs (FP16)	242	383	—	—	1,979
TFLOPs	121	383	680	—	1,979
TFLOPs (FP8)	485	—	—	—	3,958
TDP	72 W	560 W	300 W	200 W	700 W
Launch Date	Mar 2023	Nov 2021	Nov 2023	Jun 2023	Sep 2022

Efficiency Metrics

Metric	L4	MI250X	Trainium2	ATOM	H100 SXM
TFLOPs per Watt (FP32-eq)	0.84	0.34	1.13	—	1.41
Memory Bandwidth per GB	12.5 GB/s	25.6 GB/s	—	—	41.9 GB/s

Performance Equivalence

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

To match 1x NVIDIA L4

AMD MI250X

Compute (FP32-eq)

0.32x

MI250X is 3.17x faster

FP32 Compute

0.63x

MI250X is 1.58x faster

VRAM

0.19x

MI250X has 5.33x more

Memory Bandwidth

0.09x

MI250X has 10.92x more

AWS Trainium2

Compute (FP32-eq)

0.18x

Trainium2 is 5.62x faster

NVIDIA H100 SXM

Compute (FP32-eq)

0.06x

H100 SXM is 16.36x faster

FP32 Compute

0.45x

H100 SXM is 2.21x faster

VRAM

0.30x

H100 SXM has 3.33x more

Memory Bandwidth

0.09x

H100 SXM has 11.17x more

To match 1x AMD MI250X

NVIDIA L4

Compute (FP32-eq)

3.17x

Need 3.17x L4

FP32 Compute

1.58x

Need 1.58x L4

VRAM

5.33x

Need 5.33x L4

Memory Bandwidth

10.92x

Need 10.92x L4

AWS Trainium2

Compute (FP32-eq)

0.56x

Trainium2 is 1.78x faster

NVIDIA H100 SXM

Compute (FP32-eq)

0.19x

H100 SXM is 5.17x faster

FP32 Compute

0.71x

H100 SXM is 1.40x faster

VRAM

1.60x

Need 1.60x H100 SXM

Memory Bandwidth

0.98x

H100 SXM has 1.02x more

To match 1x AWS Trainium2

NVIDIA L4

Compute (FP32-eq)

5.62x

Need 5.62x L4

AMD MI250X

Compute (FP32-eq)

1.78x

Need 1.78x MI250X

NVIDIA H100 SXM

Compute (FP32-eq)

0.34x

H100 SXM is 2.91x faster

To match 1x NVIDIA H100 SXM

NVIDIA L4

Compute (FP32-eq)

16.36x

Need 16.36x L4

FP32 Compute

2.21x

Need 2.21x L4

VRAM

3.33x

Need 3.33x L4

Memory Bandwidth

11.17x

Need 11.17x L4

AMD MI250X

Compute (FP32-eq)

5.17x

Need 5.17x MI250X

FP32 Compute

1.40x

Need 1.40x MI250X

VRAM

0.63x

MI250X has 1.60x more

Memory Bandwidth

1.02x

Need 1.02x MI250X

AWS Trainium2

Compute (FP32-eq)

2.91x

Need 2.91x Trainium2

Pricing

Price Type	L4	MI250X	Trainium2	ATOM	H100 SXM
CAPEX (Street Price)	$4,000	$12,000	—	—	$30,000
OPEX (per hour)	$0.80/hr	$2.00/hr	$1.34/hr	—	$3.50/hr
Price per TFLOPs (FP32-eq)	$66	$63	—	—	$30