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 128 XPUs • 5 selected

Alibaba

Hanguang 800

AMD

MI100

23.1 TFLOPs

AMD

MI210

181 TFLOPs

AMD

MI250X

383 TFLOPs

AMD

MI300X

1,307 TFLOPs

AMD

MI325X

1,400 TFLOPs

AMD

MI350X

2,100 TFLOPs

AMD

MI355X

5,300 TFLOPs

AMD

Radeon Pro V520

23.04 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

FuriosaAI

RNGD (Renegade)

256 TFLOPs

FuriosaAI

Warboy

Google

TPU v4

275 TFLOPs

Google

TPU v5e

197 TFLOPs

Google

TPU v5p

459 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	Groq LPU Inference Engine	AMD MI300X	NVIDIA Quadro P620	NVIDIA RTX A4500	NVIDIA GeForce RTX 3090
Architecture	TSP (Tensor Streaming Processor)	CDNA 3	Pascal	Ampere	Ampere
Form Factor	—	OAM	PCIe	PCIe	PCIe
VRAM	230 GB	192 GB	2 GB	20 GB	24 GB
Memory Bandwidth	—	5,300 GB/s	80 GB/s	640 GB/s	936 GB/s
TFLOPs (FP32)	—	163.4	1.386	23.7	35.6
TFLOPs (FP16)	—	1,307	—	—	—
TFLOPs	—	1,307	1.386	46.6	71
TFLOPs (FP8)	—	2,614	—	—	—
TDP	300 W	750 W	40 W	200 W	350 W
Launch Date	Feb 2024	Dec 2023	Feb 2017	Oct 2021	Sep 2020

Efficiency Metrics

Metric	LPU Inference Engine	MI300X	Quadro P620	RTX A4500	GeForce RTX 3090
TFLOPs per Watt (FP32-eq)	—	0.87	0.03	0.12	0.10
Memory Bandwidth per GB	—	27.6 GB/s	40.0 GB/s	32.0 GB/s	39.0 GB/s

Performance Equivalence

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

To match 1x Groq LPU Inference Engine

AMD MI300X

VRAM

1.20x

Need 1.20x MI300X

NVIDIA Quadro P620

VRAM

115.00x

Need 115.00x Quadro P620

NVIDIA RTX A4500

VRAM

11.50x

Need 11.50x RTX A4500

NVIDIA GeForce RTX 3090

VRAM

9.58x

Need 9.58x GeForce RTX 3090

To match 1x AMD MI300X

Groq LPU Inference Engine

VRAM

0.83x

LPU Inference Engine has 1.20x more

NVIDIA Quadro P620

Compute (FP32-eq)

471.50x

Need 471.50x Quadro P620

FP32 Compute

117.89x

Need 117.89x Quadro P620

VRAM

96.00x

Need 96.00x Quadro P620

Memory Bandwidth

66.25x

Need 66.25x Quadro P620

NVIDIA RTX A4500

Compute (FP32-eq)

28.05x

Need 28.05x RTX A4500

FP32 Compute

6.89x

Need 6.89x RTX A4500

VRAM

9.60x

Need 9.60x RTX A4500

Memory Bandwidth

8.28x

Need 8.28x RTX A4500

NVIDIA GeForce RTX 3090

Compute (FP32-eq)

18.41x

Need 18.41x GeForce RTX 3090

FP32 Compute

4.59x

Need 4.59x GeForce RTX 3090

VRAM

8.00x

Need 8.00x GeForce RTX 3090

Memory Bandwidth

5.66x

Need 5.66x GeForce RTX 3090

To match 1x NVIDIA Quadro P620

Groq LPU Inference Engine

VRAM

0.01x

LPU Inference Engine has 115.00x more

AMD MI300X

Compute (FP32-eq)

0.00x

MI300X is 471.50x faster

FP32 Compute

0.01x

MI300X is 117.89x faster

VRAM

0.01x

MI300X has 96.00x more

Memory Bandwidth

0.02x

MI300X has 66.25x more

NVIDIA RTX A4500

Compute (FP32-eq)

0.06x

RTX A4500 is 16.81x faster

FP32 Compute

0.06x

RTX A4500 is 17.10x faster

VRAM

0.10x

RTX A4500 has 10.00x more

Memory Bandwidth

0.13x

RTX A4500 has 8.00x more

NVIDIA GeForce RTX 3090

Compute (FP32-eq)

0.04x

GeForce RTX 3090 is 25.61x faster

FP32 Compute

0.04x

GeForce RTX 3090 is 25.69x faster

VRAM

0.08x

GeForce RTX 3090 has 12.00x more

Memory Bandwidth

0.09x

GeForce RTX 3090 has 11.70x more

To match 1x NVIDIA RTX A4500

Groq LPU Inference Engine

VRAM

0.09x

LPU Inference Engine has 11.50x more

AMD MI300X

Compute (FP32-eq)

0.04x

MI300X is 28.05x faster

FP32 Compute

0.15x

MI300X is 6.89x faster

VRAM

0.10x

MI300X has 9.60x more

Memory Bandwidth

0.12x

MI300X has 8.28x more

NVIDIA Quadro P620

Compute (FP32-eq)

16.81x

Need 16.81x Quadro P620

FP32 Compute

17.10x

Need 17.10x Quadro P620

VRAM

10.00x

Need 10.00x Quadro P620

Memory Bandwidth

8.00x

Need 8.00x Quadro P620

NVIDIA GeForce RTX 3090

Compute (FP32-eq)

0.66x

GeForce RTX 3090 is 1.52x faster

FP32 Compute

0.67x

GeForce RTX 3090 is 1.50x faster

VRAM

0.83x

GeForce RTX 3090 has 1.20x more

Memory Bandwidth

0.68x

GeForce RTX 3090 has 1.46x more

To match 1x NVIDIA GeForce RTX 3090

Groq LPU Inference Engine

VRAM

0.10x

LPU Inference Engine has 9.58x more

AMD MI300X

Compute (FP32-eq)

0.05x

MI300X is 18.41x faster

FP32 Compute

0.22x

MI300X is 4.59x faster

VRAM

0.13x

MI300X has 8.00x more

Memory Bandwidth

0.18x

MI300X has 5.66x more

NVIDIA Quadro P620

Compute (FP32-eq)

25.61x

Need 25.61x Quadro P620

FP32 Compute

25.69x

Need 25.69x Quadro P620

VRAM

12.00x

Need 12.00x Quadro P620

Memory Bandwidth

11.70x

Need 11.70x Quadro P620

NVIDIA RTX A4500

Compute (FP32-eq)

1.52x

Need 1.52x RTX A4500

FP32 Compute

1.50x

Need 1.50x RTX A4500

VRAM

1.20x

Need 1.20x RTX A4500

Memory Bandwidth

1.46x

Need 1.46x RTX A4500

Pricing

Price Type	LPU Inference Engine	MI300X	Quadro P620	RTX A4500	GeForce RTX 3090
CAPEX (Street Price)	—	$35,000	—	—	—
OPEX (per hour)	—	$10.40/hr	$0.06/hr	$0.34/hr	$0.11/hr
Price per TFLOPs (FP32-eq)	—	$54	—	—	—