Compare XPUs

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

Select XPUs to Compare

Clear all (7)Jump to results

Filter by Vendor

Showing 128 XPUs • 7 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	Meta MTIA v1	Enflame Technology CloudBlazer T20	Google TPU v5p	NVIDIA GeForce RTX 5070 Ti	NVIDIA GeForce RTX 4070 Ti Super	NVIDIA GeForce RTX 3090	NVIDIA A100 SXM
Architecture	Meta Training & Inference Accelerator	GCU (AI Graphics Unit)	TPU v5	Blackwell	Ada Lovelace	Ampere	Ampere
Form Factor	—	—	Mezzanine	PCIe	PCIe	PCIe	SXM
VRAM	128 GB	32 GB	95 GB	16 GB	16 GB	24 GB	80 GB
Memory Bandwidth	—	—	—	896 GB/s	672 GB/s	936 GB/s	2,039 GB/s
TFLOPs (FP32)	—	—	—	47	—	35.6	19.5
TFLOPs (FP16)	—	—	—	—	—	—	312
TFLOPs	—	—	459	133	88	71	312
TFLOPs (FP8)	—	—	—	—	—	—	—
TDP	400 W	250 W	400 W	300 W	285 W	350 W	400 W
Launch Date	May 2023	Nov 2021	Aug 2023	Feb 2025	Jan 2024	Sep 2020	May 2020

Efficiency Metrics

Metric	MTIA v1	CloudBlazer T20	TPU v5p	GeForce RTX 5070 Ti	GeForce RTX 4070 Ti Super	GeForce RTX 3090	A100 SXM
TFLOPs per Watt (FP32-eq)	—	—	0.57	0.22	0.15	0.10	0.39
Memory Bandwidth per GB	—	—	—	56.0 GB/s	42.0 GB/s	39.0 GB/s	25.5 GB/s

Performance Equivalence

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

To match 1x Meta MTIA v1

Enflame Technology CloudBlazer T20

VRAM

4.00x

Need 4.00x CloudBlazer T20

Google TPU v5p

VRAM

1.35x

Need 1.35x TPU v5p

NVIDIA GeForce RTX 5070 Ti

VRAM

8.00x

Need 8.00x GeForce RTX 5070 Ti

NVIDIA GeForce RTX 4070 Ti Super

VRAM

8.00x

Need 8.00x GeForce RTX 4070 Ti Super

NVIDIA GeForce RTX 3090

VRAM

5.33x

Need 5.33x GeForce RTX 3090

NVIDIA A100 SXM

VRAM

1.60x

Need 1.60x A100 SXM

To match 1x Enflame Technology CloudBlazer T20

Meta MTIA v1

VRAM

0.25x

MTIA v1 has 4.00x more

Google TPU v5p

VRAM

0.34x

TPU v5p has 2.97x more

NVIDIA GeForce RTX 5070 Ti

VRAM

2.00x

Need 2.00x GeForce RTX 5070 Ti

NVIDIA GeForce RTX 4070 Ti Super

VRAM

2.00x

Need 2.00x GeForce RTX 4070 Ti Super

NVIDIA GeForce RTX 3090

VRAM

1.33x

Need 1.33x GeForce RTX 3090

NVIDIA A100 SXM

VRAM

0.40x

A100 SXM has 2.50x more

To match 1x Google TPU v5p

Meta MTIA v1

VRAM

0.74x

MTIA v1 has 1.35x more

Enflame Technology CloudBlazer T20

VRAM

2.97x

Need 2.97x CloudBlazer T20

NVIDIA GeForce RTX 5070 Ti

Compute (FP32-eq)

3.45x

Need 3.45x GeForce RTX 5070 Ti

VRAM

5.94x

Need 5.94x GeForce RTX 5070 Ti

NVIDIA GeForce RTX 4070 Ti Super

Compute (FP32-eq)

5.22x

Need 5.22x GeForce RTX 4070 Ti Super

VRAM

5.94x

Need 5.94x GeForce RTX 4070 Ti Super

NVIDIA GeForce RTX 3090

Compute (FP32-eq)

6.46x

Need 6.46x GeForce RTX 3090

VRAM

3.96x

Need 3.96x GeForce RTX 3090

NVIDIA A100 SXM

Compute (FP32-eq)

1.47x

Need 1.47x A100 SXM

VRAM

1.19x

Need 1.19x A100 SXM

To match 1x NVIDIA GeForce RTX 5070 Ti

Meta MTIA v1

VRAM

0.13x

MTIA v1 has 8.00x more

Enflame Technology CloudBlazer T20

VRAM

0.50x

CloudBlazer T20 has 2.00x more

Google TPU v5p

Compute (FP32-eq)

0.29x

TPU v5p is 3.45x faster

VRAM

0.17x

TPU v5p has 5.94x more

NVIDIA GeForce RTX 4070 Ti Super

Compute (FP32-eq)

1.51x

Need 1.51x GeForce RTX 4070 Ti Super

VRAM

1.00x

GeForce RTX 4070 Ti Super has 1.00x more

Memory Bandwidth

1.33x

Need 1.33x GeForce RTX 4070 Ti Super

NVIDIA GeForce RTX 3090

Compute (FP32-eq)

1.87x

Need 1.87x GeForce RTX 3090

FP32 Compute

1.32x

Need 1.32x GeForce RTX 3090

VRAM

0.67x

GeForce RTX 3090 has 1.50x more

Memory Bandwidth

0.96x

GeForce RTX 3090 has 1.04x more

NVIDIA A100 SXM

Compute (FP32-eq)

0.43x

A100 SXM is 2.35x faster

FP32 Compute

2.41x

Need 2.41x A100 SXM

VRAM

0.20x

A100 SXM has 5.00x more

Memory Bandwidth

0.44x

A100 SXM has 2.28x more

To match 1x NVIDIA GeForce RTX 4070 Ti Super

Meta MTIA v1

VRAM

0.13x

MTIA v1 has 8.00x more

Enflame Technology CloudBlazer T20

VRAM

0.50x

CloudBlazer T20 has 2.00x more

Google TPU v5p

Compute (FP32-eq)

0.19x

TPU v5p is 5.22x faster

VRAM

0.17x

TPU v5p has 5.94x more

NVIDIA GeForce RTX 5070 Ti

Compute (FP32-eq)

0.66x

GeForce RTX 5070 Ti is 1.51x faster

VRAM

1.00x

GeForce RTX 5070 Ti has 1.00x more

Memory Bandwidth

0.75x

GeForce RTX 5070 Ti has 1.33x more

NVIDIA GeForce RTX 3090

Compute (FP32-eq)

1.24x

Need 1.24x GeForce RTX 3090

VRAM

0.67x

GeForce RTX 3090 has 1.50x more

Memory Bandwidth

0.72x

GeForce RTX 3090 has 1.39x more

NVIDIA A100 SXM

Compute (FP32-eq)

0.28x

A100 SXM is 3.55x faster

VRAM

0.20x

A100 SXM has 5.00x more

Memory Bandwidth

0.33x

A100 SXM has 3.03x more

To match 1x NVIDIA GeForce RTX 3090

Meta MTIA v1

VRAM

0.19x

MTIA v1 has 5.33x more

Enflame Technology CloudBlazer T20

VRAM

0.75x

CloudBlazer T20 has 1.33x more

Google TPU v5p

Compute (FP32-eq)

0.15x

TPU v5p is 6.46x faster

VRAM

0.25x

TPU v5p has 3.96x more

NVIDIA GeForce RTX 5070 Ti

Compute (FP32-eq)

0.53x

GeForce RTX 5070 Ti is 1.87x faster

FP32 Compute

0.76x

GeForce RTX 5070 Ti is 1.32x faster

VRAM

1.50x

Need 1.50x GeForce RTX 5070 Ti

Memory Bandwidth

1.04x

Need 1.04x GeForce RTX 5070 Ti

NVIDIA GeForce RTX 4070 Ti Super

Compute (FP32-eq)

0.81x

GeForce RTX 4070 Ti Super is 1.24x faster

VRAM

1.50x

Need 1.50x GeForce RTX 4070 Ti Super

Memory Bandwidth

1.39x

Need 1.39x GeForce RTX 4070 Ti Super

NVIDIA A100 SXM

Compute (FP32-eq)

0.23x

A100 SXM is 4.39x faster

FP32 Compute

1.83x

Need 1.83x A100 SXM

VRAM

0.30x

A100 SXM has 3.33x more

Memory Bandwidth

0.46x

A100 SXM has 2.18x more

To match 1x NVIDIA A100 SXM

Meta MTIA v1

VRAM

0.63x

MTIA v1 has 1.60x more

Enflame Technology CloudBlazer T20

VRAM

2.50x

Need 2.50x CloudBlazer T20

Google TPU v5p

Compute (FP32-eq)

0.68x

TPU v5p is 1.47x faster

VRAM

0.84x

TPU v5p has 1.19x more

NVIDIA GeForce RTX 5070 Ti

Compute (FP32-eq)

2.35x

Need 2.35x GeForce RTX 5070 Ti

FP32 Compute

0.41x

GeForce RTX 5070 Ti is 2.41x faster

VRAM

5.00x

Need 5.00x GeForce RTX 5070 Ti

Memory Bandwidth

2.28x

Need 2.28x GeForce RTX 5070 Ti

NVIDIA GeForce RTX 4070 Ti Super

Compute (FP32-eq)

3.55x

Need 3.55x GeForce RTX 4070 Ti Super

VRAM

5.00x

Need 5.00x GeForce RTX 4070 Ti Super

Memory Bandwidth

3.03x

Need 3.03x GeForce RTX 4070 Ti Super

NVIDIA GeForce RTX 3090

Compute (FP32-eq)

4.39x

Need 4.39x GeForce RTX 3090

FP32 Compute

0.55x

GeForce RTX 3090 is 1.83x faster

VRAM

3.33x

Need 3.33x GeForce RTX 3090

Memory Bandwidth

2.18x

Need 2.18x GeForce RTX 3090

Pricing

Price Type	MTIA v1	CloudBlazer T20	TPU v5p	GeForce RTX 5070 Ti	GeForce RTX 4070 Ti Super	GeForce RTX 3090	A100 SXM
CAPEX (Street Price)	—	—	—	—	—	—	$15,000
OPEX (per hour)	—	—	$5.00/hr	$0.12/hr	$0.12/hr	$0.11/hr	$4.05/hr
Price per TFLOPs (FP32-eq)	—	—	—	—	—	—	$96