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	Google TPU v5e	AWS Trainium	AMD MI300A	Baidu Kunlun II
Architecture	Ada Lovelace	TPU v5	Inferentia/Trainium	CDNA 3	Kunlun Core
Form Factor	PCIe	Mezzanine	—	OAM	—
VRAM	24 GB	—	32 GB	128 GB	32 GB
Memory Bandwidth	300 GB/s	—	—	5,300 GB/s	—
TFLOPs (FP32)	30.3	—	—	122.6	—
TFLOPs (FP16)	242	—	—	980.6	256
TFLOPs	121	197	190	980.6	—
TFLOPs (FP8)	485	—	—	—	—
TDP	72 W	250 W	200 W	760 W	200 W
Launch Date	Mar 2023	Aug 2023	Nov 2021	Dec 2023	Aug 2021

Efficiency Metrics

Metric	L4	TPU v5e	Trainium	MI300A	Kunlun II
TFLOPs per Watt (FP32-eq)	0.84	0.39	0.47	0.65	—
Memory Bandwidth per GB	12.5 GB/s	—	—	41.4 GB/s	—

Performance Equivalence

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

To match 1x NVIDIA L4

Google TPU v5e

Compute (FP32-eq)

0.61x

TPU v5e is 1.63x faster

AWS Trainium

Compute (FP32-eq)

0.64x

Trainium is 1.57x faster

VRAM

0.75x

Trainium has 1.33x more

AMD MI300A

Compute (FP32-eq)

0.12x

MI300A is 8.10x faster

FP32 Compute

0.25x

MI300A is 4.05x faster

VRAM

0.19x

MI300A has 5.33x more

Memory Bandwidth

0.06x

MI300A has 17.67x more

Baidu Kunlun II

VRAM

0.75x

Kunlun II has 1.33x more

To match 1x Google TPU v5e

NVIDIA L4

Compute (FP32-eq)

1.63x

Need 1.63x L4

AWS Trainium

Compute (FP32-eq)

1.04x

Need 1.04x Trainium

AMD MI300A

Compute (FP32-eq)

0.20x

MI300A is 4.98x faster

To match 1x AWS Trainium

NVIDIA L4

Compute (FP32-eq)

1.57x

Need 1.57x L4

VRAM

1.33x

Need 1.33x L4

Google TPU v5e

Compute (FP32-eq)

0.96x

TPU v5e is 1.04x faster

AMD MI300A

Compute (FP32-eq)

0.19x

MI300A is 5.16x faster

VRAM

0.25x

MI300A has 4.00x more

Baidu Kunlun II

VRAM

1.00x

Kunlun II has 1.00x more

To match 1x AMD MI300A

NVIDIA L4

Compute (FP32-eq)

8.10x

Need 8.10x L4

FP32 Compute

4.05x

Need 4.05x L4

VRAM

5.33x

Need 5.33x L4

Memory Bandwidth

17.67x

Need 17.67x L4

Google TPU v5e

Compute (FP32-eq)

4.98x

Need 4.98x TPU v5e

AWS Trainium

Compute (FP32-eq)

5.16x

Need 5.16x Trainium

VRAM

4.00x

Need 4.00x Trainium

Baidu Kunlun II

VRAM

4.00x

Need 4.00x Kunlun II

To match 1x Baidu Kunlun II

NVIDIA L4

VRAM

1.33x

Need 1.33x L4

AWS Trainium

VRAM

1.00x

Trainium has 1.00x more

AMD MI300A

VRAM

0.25x

MI300A has 4.00x more

Pricing

Price Type	L4	TPU v5e	Trainium	MI300A	Kunlun II
CAPEX (Street Price)	$4,000	—	—	—	—
OPEX (per hour)	$0.80/hr	$1.80/hr	—	—	—
Price per TFLOPs (FP32-eq)	$66	—	—	—	—