NVIDIA GA100

GA100

Block Diagram

SM Diagram

NVIDIA's GA100 GPU uses the Ampere architecture and is made using a 7 nm production process at TSMC. With a die size of 826 mm² and a transistor count of 54,200 million it is a very big chip. GA100 does not support DirectX. For GPU compute applications, OpenCL version 3.0 and CUDA 8.0 can be used. It features 8192 shading units, 512 texture mapping units and 192 ROPs. Also included are 512 tensor cores which help improve the speed of machine learning applications.

Further reading: Ampere Architecture Whitepaper

Graphics Processor

GPU Name: GA100

Codename: NV170

Architecture: Ampere

Foundry: TSMC

Process Size: 7 nm

Transistors: 54,200 million

Density: 65.6M / mm²

Die Size: 826 mm²

Released: May 14th, 2020

Graphics Features

DirectX: N/A

OpenGL: N/A

OpenCL: 3.0

Vulkan: N/A

CUDA: 8.0

Shader Model: N/A

Render Config

Shading Units: 8192

TMUs: 512

ROPs: 192

SM Count: 128

FP16 Units: 32768

FP64 Units: 4096

INT4 Units: 131072

INT8 Units: 65536

INT32 Units: 8192

Tensor Cores: 512

SFUs: 2048

TPCs: 64

GPCs: 8

Tex L1 Cache: 64 KB per SM

L1 Cache: 192 KB per SM

L2 Cache: 81920 KB

Register File: 32768 KB

Max. TDP: 400 W

All Ampere GPUs

NVIDIA GPU Architecture History

1998-2000 Fahrenheit
1999-2005 Celsius
2001-2003 Kelvin
2003-2005 Rankine
2003-2013 Curie
2006-2010 Tesla
2007-2013 Tesla 2.0
2010-2016 Fermi
2010-2013 VLIW Vec4
2010-2016 Fermi 2.0
2012-2018 Kepler
2013-2015 Kepler 2.0
2014-2017 Maxwell
2014-2019 Maxwell 2.0
2016-2021 Pascal
2017-2020 Volta
2018-2022 Turing
2020-2022 Ampere
2022 Hopper
2022-2023 Lovelace

Graphics cards using the NVIDIA GA100 GPU

Name	Chip	Memory	Shaders	TMUs	ROPs	Base Clock	Boost Clock	Memory Clock
NVIDIA A100 SXM4 40 GB		40 GB	6912	432	160	1095 MHz	1410 MHz	1215 MHz
NVIDIA GRID A100B		48 GB	6912	432	192	900 MHz	1005 MHz	1215 MHz
NVIDIA GRID A100A		48 GB	6912	432	192	900 MHz	1005 MHz	1215 MHz
NVIDIA A100 PCIe		40 GB	6912	432	160	765 MHz	1410 MHz	1215 MHz
NVIDIA A100 SXM4 80 GB		80 GB	6912	432	160	1275 MHz	1410 MHz	1593 MHz
NVIDIA A30 PCIe		24 GB	3584	224	96	930 MHz	1440 MHz	1215 MHz
NVIDIA PG506-232		24 GB	3584	224	96	930 MHz	1440 MHz	1215 MHz
NVIDIA A100 PCIe 80 GB		80 GB	6912	432	160	1065 MHz	1410 MHz	1593 MHz
NVIDIA PG506-242		24 GB	3584	224	96	930 MHz	1440 MHz	1215 MHz
NVIDIA CMP 170HX	GA100-105F-A1	16 GB	4480	280	128	1140 MHz	1410 MHz	1458 MHz