NVIDIA GA104
Block Diagram
SM Diagram
NVIDIA's GA104 GPU uses the Ampere architecture and is made using a 8 nm production process at Samsung. With a die size of 392 mm² and a transistor count of 17,400 million it is a large chip. GA104 supports DirectX 12 Ultimate (Feature Level 12_2). For GPU compute applications, OpenCL version 2.0 and CUDA 8.5 can be used. Additionally, the DirectX 12 Ultimate capability guarantees support for hardware-raytracing, variable-rate shading and more, in upcoming video games. It features 6144 shading units, 192 texture mapping units and 96 ROPs. Also included are 192 tensor cores which help improve the speed of machine learning applications. The GPU also contains 48 raytracing acceleration cores.
Further reading:
Ampere Architecture Whitepaper 
Graphics Processor
- GPU Name
- GA104
- Codename
- NV174
- Architecture
- Ampere
- Foundry
- Samsung
- Process Size
- 8 nm
- Transistors
- 17,400 million
- Density
- 44.4M / mm²
- Die Size
- 392 mm²
- Released
- Sep 1st, 2020
Graphics Features
- DirectX
- 12 Ultimate (12_2)
- OpenGL
- 4.6
- OpenCL
- 2.0
- Vulkan
- 1.2
- CUDA
- 8.5
- Shader Model
- 6.5
- PureVideo HD
- VP11
- VDPAU
- Feature Set k
Render Config
- Shading Units
- 6144
- TMUs
- 192
- ROPs
- 96
- SM Count
- 48
- FP16 Units
- 6144
- FP64 Units
- 96
- INT32 Units
- 3072
- Tensor Cores
- 192
- RT Cores
- 48
- SFUs
- 768
- TPCs
- 24
- GPCs
- 6
- Tex L1 Cache
- 64 KB per SM
- L1 Cache
- 128 KB per SM
- L2 Cache
- 4096 KB
- Max. TDP
- 250 W
All Ampere GPUs
- NVIDIA GA100
- NVIDIA GA102
- NVIDIA GA104
- NVIDIA GA106
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
- 2016-2020 Pascal
- 2017-2020 Volta
- 2018-2020 Turing
- 2020-2021 Ampere
Graphics cards using the NVIDIA GA104 GPU
| Name | Chip | Memory | Shaders | TMUs | ROPs | Base Clock | Boost Clock | Memory Clock |
|---|---|---|---|---|---|---|---|---|
NVIDIA GeForce RTX 3070
|
GA104-300-A1 | 8 GB | 5888 | 184 | 96 | 1500 MHz | 1725 MHz | 1750 MHz |
| NVIDIA GeForce RTX 3070 Ti | GA104-400-A1 | 16 GB | 6144 | 192 | 96 | 1410 MHz | 1695 MHz | 2000 MHz |
| NVIDIA GeForce RTX 3060 Ti | GA104-200-A1 | 8 GB | 4864 | 152 | 80 | 1410 MHz | 1665 MHz | 1750 MHz |
| NVIDIA GeForce RTX 3070 Mobile | 8 GB | 5888 | 184 | 96 | 1110 MHz | 1545 MHz | 1500 MHz | |
| NVIDIA GeForce RTX 3070 Max-Q | 8 GB | 5888 | 184 | 96 | 780 MHz | 1290 MHz | 1250 MHz | |
| NVIDIA GeForce RTX 3070 Max-Q | 8 GB | 5888 | 184 | 96 | 780 MHz | 1245 MHz | 1250 MHz |
GA104 GPU Notes
| Speculation |