Compare XPUs

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

Select XPUs to Compare

Clear all (5)Jump to results
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

Meta

MTIA v1

Microsoft

Maia 100

700 TFLOPs

Moore Threads

MTT S80

NVIDIA

A10

125 TFLOPs

NVIDIA

A100 SXM

312 TFLOPs

NVIDIA

A40

150 TFLOPs

NVIDIA

B200

2,250 TFLOPs

NVIDIA

GB200 NVL72

360,000 TFLOPs

NVIDIA

GB200 Superchip

5,000 TFLOPs

NVIDIA

GeForce RTX 4060 Ti

44.2 TFLOPs

NVIDIA

GeForce RTX 4070

58.2 TFLOPs

NVIDIA

GeForce RTX 4070 Super

71 TFLOPs

NVIDIA

GeForce RTX 4070 Ti

80.2 TFLOPs

NVIDIA

GeForce RTX 4070 Ti Super

88.2 TFLOPs

NVIDIA

GeForce RTX 4080

97.5 TFLOPs

NVIDIA

GeForce RTX 4080 Super

104.4 TFLOPs

NVIDIA

GeForce RTX 4090

165.2 TFLOPs

NVIDIA

GeForce RTX 5070

61.6 TFLOPs

NVIDIA

GeForce RTX 5070 Ti

88 TFLOPs

NVIDIA

GeForce RTX 5080

112.6 TFLOPs

NVIDIA

GeForce RTX 5090

209.5 TFLOPs

NVIDIA

H100 PCIe

1,513 TFLOPs

NVIDIA

H100 SXM

1,979 TFLOPs

NVIDIA

H200 PCIe

1,513 TFLOPs

NVIDIA

H200 SXM

1,979 TFLOPs

NVIDIA

L4

121 TFLOPs

NVIDIA

L40S

733 TFLOPs

NVIDIA

RTX 4000 Ada Generation

53.4 TFLOPs

NVIDIA

RTX 5000 Ada Generation

130.6 TFLOPs

NVIDIA

RTX 6000 Ada Generation

182.2 TFLOPs

NVIDIA

RTX PRO 6000 Blackwell Max-Q

125 TFLOPs

NVIDIA

RTX PRO 6000 Blackwell Server Edition

250 TFLOPs

NVIDIA

RTX PRO 6000 Blackwell Workstation Edition

250 TFLOPs

Qualcomm

Cloud AI 100

50 TFLOPs

Rebellions

ATOM

SambaNova

SN40L

Tenstorrent

Grayskull

200 TFLOPs

Tenstorrent

Wormhole

364 TFLOPs

Maximum of 5 XPUs can be compared at once. Deselect one to add another.

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

SpecificationNVIDIA L40SAMD MI325XSambaNova SN40LAWS Trainium2Etched Sohu
ArchitectureAda LovelaceCDNA 3.5Reconfigurable Dataflow Unit (RDU)Trainium Gen2Transformer ASIC
Form FactorPCIeOAM——PCIe
VRAM48 GB256 GB640 GB—144 GB
Memory Bandwidth864 GB/s6,000 GB/s——4,800 GB/s
TFLOPs (FP32)91.6175——10,000
TFLOPs (FP16)7331,400———
TFLOPs7331,400—68010,000
TFLOPs (FP8)1,4662,800———
TDP350 W800 W700 W300 W700 W
Launch DateOct 2023Oct 2024Jan 2024Nov 2023Jun 2024

Efficiency Metrics

MetricL40SMI325XSN40LTrainium2Sohu
TFLOPs per Watt (FP32-eq)1.050.88—1.1314.29
Memory Bandwidth per GB18.0 GB/s23.4 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

AMD MI325X
Compute (FP32-eq)
0.52x
MI325X is 1.91x faster
FP32 Compute
0.52x
MI325X is 1.91x faster
VRAM
0.19x
MI325X has 5.33x more
Memory Bandwidth
0.14x
MI325X has 6.94x more
SambaNova SN40L
VRAM
0.07x
SN40L has 13.33x more
AWS Trainium2
Compute (FP32-eq)
1.08x
Need 1.08x Trainium2
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 AMD MI325X

NVIDIA L40S
Compute (FP32-eq)
1.91x
Need 1.91x L40S
FP32 Compute
1.91x
Need 1.91x L40S
VRAM
5.33x
Need 5.33x L40S
Memory Bandwidth
6.94x
Need 6.94x L40S
SambaNova SN40L
VRAM
0.40x
SN40L has 2.50x more
AWS Trainium2
Compute (FP32-eq)
2.06x
Need 2.06x Trainium2
Etched Sohu
Compute (FP32-eq)
0.07x
Sohu is 14.29x faster
FP32 Compute
0.02x
Sohu is 57.14x faster
VRAM
1.78x
Need 1.78x Sohu
Memory Bandwidth
1.25x
Need 1.25x Sohu

To match 1x SambaNova SN40L

NVIDIA L40S
VRAM
13.33x
Need 13.33x L40S
AMD MI325X
VRAM
2.50x
Need 2.50x MI325X
Etched Sohu
VRAM
4.44x
Need 4.44x Sohu

To match 1x AWS Trainium2

NVIDIA L40S
Compute (FP32-eq)
0.93x
L40S is 1.08x faster
AMD MI325X
Compute (FP32-eq)
0.49x
MI325X is 2.06x faster
Etched Sohu
Compute (FP32-eq)
0.03x
Sohu is 29.41x faster

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
AMD MI325X
Compute (FP32-eq)
14.29x
Need 14.29x MI325X
FP32 Compute
57.14x
Need 57.14x MI325X
VRAM
0.56x
MI325X has 1.78x more
Memory Bandwidth
0.80x
MI325X has 1.25x more
SambaNova SN40L
VRAM
0.23x
SN40L has 4.44x more
AWS Trainium2
Compute (FP32-eq)
29.41x
Need 29.41x Trainium2

Pricing

Price TypeL40SMI325XSN40LTrainium2Sohu
CAPEX (Street Price)$10,000$45,000———
OPEX (per hour)$1.50/hr——$1.34/hr—
Price per TFLOPs (FP32-eq)$27$64———