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 L40S	Google TPU v4	NVIDIA GeForce RTX 4070 Super	Etched Sohu
Architecture	Ada Lovelace	TPU v4	Ada Lovelace	Transformer ASIC
Form Factor	PCIe	Mezzanine	PCIe	PCIe
VRAM	48 GB	—	12 GB	144 GB
Memory Bandwidth	864 GB/s	—	504 GB/s	4,800 GB/s
TFLOPs (FP32)	91.6	—	35.5	10,000
TFLOPs (FP16)	733	—	71	—
TFLOPs	733	275	71	10,000
TFLOPs (FP8)	1,466	—	—	—
TDP	350 W	300 W	220 W	700 W
Launch Date	Oct 2023	May 2021	Jan 2024	Jun 2024

Efficiency Metrics

Metric	L40S	TPU v4	GeForce RTX 4070 Super	Sohu
TFLOPs per Watt (FP32-eq)	1.05	0.46	0.16	14.29
Memory Bandwidth per GB	18.0 GB/s	—	42.0 GB/s	33.3 GB/s

Performance Equivalence

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

To match 1x NVIDIA L40S

Google TPU v4

Compute (FP32-eq)

2.67x

Need 2.67x TPU v4

NVIDIA GeForce RTX 4070 Super

Compute (FP32-eq)

10.32x

Need 10.32x GeForce RTX 4070 Super

FP32 Compute

2.58x

Need 2.58x GeForce RTX 4070 Super

VRAM

4.00x

Need 4.00x GeForce RTX 4070 Super

Memory Bandwidth

1.71x

Need 1.71x GeForce RTX 4070 Super

Etched Sohu

Compute (FP32-eq)

0.04x

Sohu is 27.29x faster

FP32 Compute

0.01x

Sohu is 109.17x faster

VRAM

0.33x

Sohu has 3.00x more

Memory Bandwidth

0.18x

Sohu has 5.56x more

To match 1x Google TPU v4

NVIDIA L40S

Compute (FP32-eq)

0.38x

L40S is 2.67x faster

NVIDIA GeForce RTX 4070 Super

Compute (FP32-eq)

3.87x

Need 3.87x GeForce RTX 4070 Super

Etched Sohu

Compute (FP32-eq)

0.01x

Sohu is 72.73x faster

To match 1x NVIDIA GeForce RTX 4070 Super

NVIDIA L40S

Compute (FP32-eq)

0.10x

L40S is 10.32x faster

FP32 Compute

0.39x

L40S is 2.58x faster

VRAM

0.25x

L40S has 4.00x more

Memory Bandwidth

0.58x

L40S has 1.71x more

Google TPU v4

Compute (FP32-eq)

0.26x

TPU v4 is 3.87x faster

Etched Sohu

Compute (FP32-eq)

0.00x

Sohu is 281.69x faster

FP32 Compute

0.00x

Sohu is 281.69x faster

VRAM

0.08x

Sohu has 12.00x more

Memory Bandwidth

0.10x

Sohu has 9.52x more

To match 1x Etched Sohu

NVIDIA L40S

Compute (FP32-eq)

27.29x

Need 27.29x L40S

FP32 Compute

109.17x

Need 109.17x L40S

VRAM

3.00x

Need 3.00x L40S

Memory Bandwidth

5.56x

Need 5.56x L40S

Google TPU v4

Compute (FP32-eq)

72.73x

Need 72.73x TPU v4

NVIDIA GeForce RTX 4070 Super

Compute (FP32-eq)

281.69x

Need 281.69x GeForce RTX 4070 Super

FP32 Compute

281.69x

Need 281.69x GeForce RTX 4070 Super

VRAM

12.00x

Need 12.00x GeForce RTX 4070 Super

Memory Bandwidth

9.52x

Need 9.52x GeForce RTX 4070 Super

Pricing

Price Type	L40S	TPU v4	GeForce RTX 4070 Super	Sohu
CAPEX (Street Price)	$10,000	—	—	—
OPEX (per hour)	$1.50/hr	$3.00/hr	—	—
Price per TFLOPs (FP32-eq)	$27	—	—	—