Compare XPUs

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

Select XPUs to Compare

Clear all (4)Jump to results

Filter by Vendor

Showing 72 XPUs • 4 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 A100 SXM	NVIDIA A40	Cerebras WSE-3	Google TPU v6e (Trillium)
Architecture	Ampere	Ampere	Wafer Scale Engine	Trillium
Form Factor	SXM	PCIe	—	Custom
VRAM	80 GB	48 GB	44 GB	32 GB
Memory Bandwidth	2,039 GB/s	696 GB/s	—	1,600 GB/s
TFLOPs (FP32)	19.5	37.4	—	459
TFLOPs (FP16)	312	150	—	918
TFLOPs	312	150	—	918
TFLOPs (FP8)	—	—	—	—
TDP	400 W	300 W	23000 W	200 W
Launch Date	May 2020	Oct 2020	Mar 2024	May 2024

Efficiency Metrics

Metric	A100 SXM	A40	WSE-3	TPU v6e (Trillium)
TFLOPs per Watt (FP32-eq)	0.39	0.25	—	2.29
Memory Bandwidth per GB	25.5 GB/s	14.5 GB/s	—	50.0 GB/s

Performance Equivalence

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

To match 1x NVIDIA A100 SXM

NVIDIA A40

Compute (FP32-eq)

2.08x

Need 2.08x A40

FP32 Compute

0.52x

A40 is 1.92x faster

VRAM

1.67x

Need 1.67x A40

Memory Bandwidth

2.93x

Need 2.93x A40

Cerebras WSE-3

VRAM

1.82x

Need 1.82x WSE-3

Google TPU v6e (Trillium)

Compute (FP32-eq)

0.34x

TPU v6e (Trillium) is 2.94x faster

FP32 Compute

0.04x

TPU v6e (Trillium) is 23.54x faster

VRAM

2.50x

Need 2.50x TPU v6e (Trillium)

Memory Bandwidth

1.27x

Need 1.27x TPU v6e (Trillium)

To match 1x NVIDIA A40

NVIDIA A100 SXM

Compute (FP32-eq)

0.48x

A100 SXM is 2.08x faster

FP32 Compute

1.92x

Need 1.92x A100 SXM

VRAM

0.60x

A100 SXM has 1.67x more

Memory Bandwidth

0.34x

A100 SXM has 2.93x more

Cerebras WSE-3

VRAM

1.09x

Need 1.09x WSE-3

Google TPU v6e (Trillium)

Compute (FP32-eq)

0.16x

TPU v6e (Trillium) is 6.12x faster

FP32 Compute

0.08x

TPU v6e (Trillium) is 12.27x faster

VRAM

1.50x

Need 1.50x TPU v6e (Trillium)

Memory Bandwidth

0.43x

TPU v6e (Trillium) has 2.30x more

To match 1x Cerebras WSE-3

NVIDIA A100 SXM

VRAM

0.55x

A100 SXM has 1.82x more

NVIDIA A40

VRAM

0.92x

A40 has 1.09x more

Google TPU v6e (Trillium)

VRAM

1.38x

Need 1.38x TPU v6e (Trillium)

To match 1x Google TPU v6e (Trillium)

NVIDIA A100 SXM

Compute (FP32-eq)

2.94x

Need 2.94x A100 SXM

FP32 Compute

23.54x

Need 23.54x A100 SXM

VRAM

0.40x

A100 SXM has 2.50x more

Memory Bandwidth

0.78x

A100 SXM has 1.27x more

NVIDIA A40

Compute (FP32-eq)

6.12x

Need 6.12x A40

FP32 Compute

12.27x

Need 12.27x A40

VRAM

0.67x

A40 has 1.50x more

Memory Bandwidth

2.30x

Need 2.30x A40

Cerebras WSE-3

VRAM

0.73x

WSE-3 has 1.38x more

Pricing

Price Type	A100 SXM	A40	WSE-3	TPU v6e (Trillium)
CAPEX (Street Price)	$15,000	—	—	—
OPEX (per hour)	$4.05/hr	—	—	—
Price per TFLOPs (FP32-eq)	$96	—	—	—