NVIDIA A100 40GB vs NVIDIA H100 Full Specs

Summary

6,912 Shaders 1.41GHz	14,592 Shaders 1.75GHz
40GB HBM2e1.55TB/s	80GB HBM2e2.04TB/s
Peak·FP32·FP16 19.49 TFLOPS	Peak·FP32·FP16 51.22 TFLOPS
Form Factor PCIe Card	Form Factor PCIe Card
TDP 250W	TDP 350W
Power Connectors - 1x 8-Pin EPS	Power Connectors 1x 16-Pin 12VHPWR -

Highlights

Peak OPS

Peak OPS FP32 FP16 Tensor FP8-16 FP8-16 Tensor Sparse Tensor FP8-32 FP8-32 Tensor Sparse Tensor FP16-16 FP16-16 Tensor Sparse Tensor FP16-32 FP16-32 Tensor Sparse FP64 Tensor FP64 BF16 Tensor BF16 BF16 Tensor Sparse Tensor TF32 Tensor INT8 Tensor INT4 Memory Bandwidth TDP

A100 40GB2.5 POPSINT4 Tensor Sparse

x1

H1003.03 POPSINT8 Tensor Sparse

x1.21

Show Top Score

Theoretical Performance

Clock Speed1410MHz UC·Base·Boost·OC	Clock Speed1755MHz UC·Base·Boost·OC
Peak OPS 2.5 POPSINT4 Tensor Sparse	Peak OPS 3.03 POPSINT8 Tensor Sparse
-	Tensor FP8-16 1.51 PFLOPS FP8-16 Tensor Sparse 3.03 PFLOPS Tensor FP8-32 1.51 PFLOPS FP8-32 Tensor Sparse 3.03 PFLOPS
Tensor FP16-16 311.9 TFLOPS FP16-16 Tensor Sparse 623.7 TFLOPS Tensor FP16-32 311.9 TFLOPS FP16-32 Tensor Sparse 623.7 TFLOPS	Tensor FP16-16 756.4 TFLOPS FP16-16 Tensor Sparse 1.51 PFLOPS Tensor FP16-32 756.4 TFLOPS FP16-32 Tensor Sparse 1.51 PFLOPS
BF16 38.98 TFLOPS Tensor BF16 311.9 TFLOPS BF16 Tensor Sparse 623.7 TFLOPS	BF16 102.4 TFLOPS Tensor BF16 756.4 TFLOPS BF16 Tensor Sparse 1.51 PFLOPS
Tensor TF32 155.9 TFLOPS	Tensor TF32 378.2 TFLOPS
FP32 19.49 TFLOPS	FP32 51.22 TFLOPS
FP64 9.75 TFLOPS Tensor FP64 19.49 TFLOPS	FP64 25.61 TFLOPS Tensor FP64 47.28 TFLOPS
Tensor INT4 1.25 POPS	Tensor INT4 -
Tensor INT8 623.7 TOPS	Tensor INT8 1.51 POPS
Pixel Rate 225.6 GPixel/s	Pixel Rate 42.1 GPixel/s
Texture Rate 609.1 GTexel/s	Texture Rate 800.3 GTexel/s

Core Config

Shaders 6,912 Shaders	Shaders 14,592 Shaders
TMUs 432 TMUs	TMUs 456 TMUs
ROPs 160 ROPs	ROPs 24 ROPs
Tensor Cores 432 T-Cores	Tensor Cores 456 T-Cores
SMs 108 SMs	SMs 114 SMs

Clocks

Base Clock 765MHz	Base Clock 1.06GHz
Boost Clock 1.41GHz	Boost Clock 1.75GHz
Tensor Clock -	Tensor Clock 1.62GHz

Cache

L2 Cache 41MB shared

L2 Cache 51.2MB shared

Memory

40GB HBM2e	80GB HBM2e
Memory Bus 5120-bit	Memory Bus 5120-bit
Memory Speed 2.4GT/s	Memory Speed 3.2GT/s
Memory Bandwidth 1.55TB/s	Memory Bandwidth 2.04TB/s

Power & Thermals

TDP 250W

TDP 350W

Decoder

Decoder Model 5x NVDEC 4

Decoder Model 7x NVDEC 5

Card

Form Factor PCIe Card	Form Factor PCIe Card
PCIe 2-Slots	PCIe 2-Slots
Height 111 mm (4.37") Width 267 mm (10.51") Depth 40 mm (1.57")	Height 111 mm (4.37") Width 268 mm (10.55") Depth 40 mm (1.57")
Cooling Passive	Cooling Passive
Power Connectors 1x 8-Pin EPS	Power Connectors 1x 16-Pin 12VHPWR

Chip

Manufacturer NVIDIA	Manufacturer NVIDIA
Chip Designer NVIDIA	Chip Designer NVIDIA
Architecture Ampere	Architecture Hopper
Family Server	Family Server
Branding	Branding
Codename NV170	Codename NV180
Market Segment Server	Market Segment Server
Release Date Jun 28, 2021	Release Date Mar 22, 2022

Fabrication

Foundry TSMC	Foundry TSMC
Fabrication Node 7N	Fabrication Node 4N
Die Size 826mm²	Die Size 814mm²
Transistor Count 54.2 Billion	Transistor Count 80 Billion
Transistor Density 65.62 MTr/mm²	Transistor Density 98.28 MTr/mm²

Images

No images available

No images available

Competitors

NVIDIAA100 40GB vs GPU AMDInstinct MI300P NVIDIAA100 40GB vs GPU NVIDIARTX PRO 4500 Blackwell NVIDIAA100 40GB vs GPU NVIDIARTX PRO 5000 Blackwell Show 12 more

NVIDIAH100 vs GPU AMDInstinct MI300P NVIDIAH100 vs GPU NVIDIARTX PRO 5000 Blackwell NVIDIAH100 vs GPU NVIDIAB100 Show 12 more