Nvidia GPU Architecture
来自Jack's Lab
(版本间的差异)
(→Maxwell Micro Architecture) |
(→Pascal Micro Architecture) |
||
第247行: | 第247行: | ||
[[文件:Pascal-GP104-arch.png | 950px]] | [[文件:Pascal-GP104-arch.png | 950px]] | ||
+ | |||
+ | |||
+ | * 1 SM: 2 Warp Scheduler (2 instruction dispatchers per Warp) | ||
+ | * 1 Warp: 32 CUDA cores + 16 DPU + 8 Load/Store Units + 8 SPU | ||
+ | |||
+ | * 1 SM: 64 CUDA cores + 32 DPU + 16 Load/Store Units + 16 SPU | ||
<br> | <br> |
2017年3月23日 (四) 13:28的版本
目录 |
1 Overview
The GPU architecture is built around a scalable array of Streaming Multiprocessors (SM)
The key components of a SM:
- CUDA cores (ALU + FPU)
- Double Precision Units (DPU)
- Special Function Units (SPU)
- Load/Store Units (LD/ST)
- Register File
- Shared Memory/L1 Cache
- Warp Scheduler
2 Fermi Micro Architecture
The Fermi architecture was the first complete GPU computing architecture to deliver the features required for the most demanding HPC applications.
- 1 SM: 32 CUDA cores + 16 Load/Store Unit + 4 SPU + 2 Warp Scheduler
- 1 SM: 2 Warps
- 1 Warp: 16 CUDA cores + 16 Load/Store Unit(shared) + 4 SPU(shared) + [32 threads context ?]
- Handle 48 warps per SM for a total of 1536 (48x32) threads resident in a single SM at a time [48 Warps context ?]
- 1 CUDA core: 1 ALU + 1 FPU
- Register file is 32KB
GTX480:
- 15 SM (32 CUDA cores/SM)
- 480 CUDA cores
- 1345 GFLOPs
- 40 nm
- 3.2 billion transistors
- GTX480 250Watts
2.1 Fermi Cards
2.1.1 GeForce 400 Series
- Release date: April 12, 2010
- Codename: GF10x
- Architecture: Fermi
- Models:
- GeForce Series
- GeForce GT Series
- GeForce GTS Series
- GeForce GTX Series
- Fabrication process and transistors:
- 260M 40nm (GT218)
- 585M 40 nm (GF108)
- 1.170M 40 nm (GF106)
- 1.950M 40 nm (GF104)
- 1.950M 40 nm (GF114)
- 3.200M 40 nm (GF100)
- Cards:
- Entry-level GT420 GT430
- Mid-range GT440 GTS450 GTX460
- High-end GTX465 GTX470
- Enthusiast GTX480 (2010.3, 3.2 billion Transistors, 15 SMs, 1536MB, 1345 GFLOPS, 250W)
2.1.2 GeForce 500 Series
- Release date: 8 November 2010
- Codename: GF11x
- Architecture: Fermi
- Models:
- GeForce Series
- GeForce GT Series
- GeForce GTX Series
- Fabrication process and transistors:
- 292M 40nm (GF119)
- 585M 40 nm (GF108)
- 1.170M 40 nm (GF116)
- 1.950M 40 nm (GF114)
- 3.000M 40 nm (GF110)
- Cards:
- Entry-level 510 GT520 GT530
- Mid-range GT545 GTX550Ti GTX560 GTX560Ti
- High-end GTX570 GTX580 GTX590(2011.3, 2x3 billion transistors, 32 SMs, 2x1536MB, 2488GFLOPS, 365W)
3 Kepler Micro Architecture
Released in the fall of 2012
- 1 SM: 4 Warps Scheduler (2 instruction dispatchers per Warp)
- 1 Warp: [32 threads context ?]
- 1 SM: 192 CUDA cores + 64 DPU (shared) + 32 Load/Store Unit (shared) + 32 SPU (shared) + 4 Warp Scheduler
- Handle 64 warps/SM for a total of 2048 (64x32) threads resident in a single SM at a time [64 Warps context ?]
- Register file size is 64K
K20X:
- 14 SM
- 2688 CUDA cores, 6GB
- 3.935 TFLOPs / DPU: 1.312 TFLOPs
- 28 nm
- 235Watts
GTX690:
- 2x8 SM
- 3072 CUDA cores
- 2x2.8TFLOPs
- 2x3.54 billion transistors
- 300Watts (2012.4)
3.1 Kepler Cards
3.1.1 GeForce 600 series
- Release date: March 22, 2012
- Codename: GK10x
- Models
- GeForce Series
- GeForce GT Series
- GeForce GTX Series
- Fabrication process and transistors
- 292M 40 nm (GF119)
- 585M 40 nm (GF108)
- 1.170M 40 nm (GF116)
- 1.950M 40 nm (GF114)
- 1.270M 28 nm (GK107)
- 1.270M 28 nm (GK208)
- 2.540M 28 nm (GK106)
- 3.540M 28 nm (GK104)
- Cards:
- Entry-level GT610 GT620 GT630 GT640
- Mid-range GTX650 GTX650Ti GTX650Ti Boost GTX 660
- High-end GTX660Ti GTX670
- Enthusiast GTX680 GTX690
3.1.2 GeForce 700 series
- Release date: May 2013
- Codename: GK110 GK208
- Models:
- GeForce Series
- GeForce GT Series
- GeForce GTX Series
- Fabrication process and transistors:
- 585M 28 nm (GF117)
- 1.020M 28 nm (GK208)
- 1.270M 28 nm (GK107)
- 3.540M 28 nm (GK104)
- 7.080M 28 nm (GK110)
- Cards
- Entry-level: GeForce GT 705 GeForce GT 710 GeForce GT 720 GeForce GT 730 GeForce GT 740 GeForce GTX 745
- Mid-range: GeForce GTX 750 GeForce GTX 750 Ti GeForce GTX 760 192-Bit GeForce GTX 760 GeForce GTX 760 Ti
- High-end: GeForce GTX 770 GeForce GTX 780
- Enthusiast: GeForce GTX 780 Ti GeForce GTX Titan GeForce GTX Titan Black GeForce GTX Titan Z
4 Maxwell Micro Architecture
- 1 SM (SMM): 4 Warp Scheduler (2 instruction dispatchers per Warp)
- 1 Warp: 32 CUDA cores + 1 DPU + 8 Load/Store Units + 8 SPU
- 1 SM (SMM): 128 CUDA cores + 4 DPU + 32 Load/Store Units + 32 SPU
GTX980:
- 16 SM (SMM)
- 2048 CUDA cores
- 4612 GFLOPs / DPU: 144 GFLOPs
- 28 nm
- 5.2 billion transistors
- 165W
TITAN X:
4.1 Maxwell Cards
4.1.1 GeForce 900 series
- Release date: September 2014
- Codename: GM20x
- Models
- GeForce Series
- GeForce GT Series
- GeForce GTX Series
- Cards
- Mid-range GTX950 / GTX960
- High-end GTX970 / GTX980
- Enthusiast GTX980 Ti / GTX Titan X
5 Pascal Micro Architecture
- 1 SM: 2 Warp Scheduler (2 instruction dispatchers per Warp)
- 1 Warp: 32 CUDA cores + 16 DPU + 8 Load/Store Units + 8 SPU
- 1 SM: 64 CUDA cores + 32 DPU + 16 Load/Store Units + 16 SPU
5.1 Pascal Cards
5.1.1 GeForce 1000 series
- Release date: May 2016
- Codename: GP10x
- Models
- GeForce GTX Series
- Fabrication process and transistors:
- 3.3B 14 nm (GP107)
- 4.4B 16 nm (GP106)
- 7.2B 16 nm (GP104)
- 12B 16 nm (GP102)
- Cards:
- Entry-level: GTX1050 / GTX1050 Ti
- Mid-range: GTX1060
- High-end: GTX1070 / GTX1080(2016.5, 2560 CUDA cores, 20 SMs, 16nm, 7.2billion, 8GB, 8.2TFLOPs / DPU: 0.257TFLOPs, 180Watts)
- Enthusiast: GTX1080 Ti / NVIDIA Titan X(2016.8, 3584 CUDA cores, 28 SMs, 16nm, 12billion, 12GB, 10TFLOPs / DPU: 0.317TFLOPs, 250Watts)
6 Reference
- https://en.wikipedia.org/wiki/Fermi_(microarchitecture)
- https://en.wikipedia.org/wiki/GeForce_400_series
- https://en.wikipedia.org/wiki/GeForce_500_series
- https://en.wikipedia.org/wiki/GeForce_600_series
- https://en.wikipedia.org/wiki/GeForce_700_series
- https://en.wikipedia.org/wiki/GeForce_800M_series