Architecture
NVIDIA Ampere 8nm
Most Powerful Variant
GA102
Cores : TMUs : ROPs
10752 : 336 : 96
Typical Memory Size
49152 MB
Typical Memory Bus
384 -bit
GA102 GPU
Overview
Manufacturer
NVIDIA
Release Date
October 5th, 2020
Graphics Processing Unit
Architecture
Ampere
Fabrication Process
8 nm
Die Size
628 mm2
Transistors Count
28.3B
Tensor Cores
336
RT Cores
84
CUDAs
10752
TMUs
336
ROPs
96
Clocks
Base Clock
1455 MHz
Memory Configuration
Memory Size
49152 MB
Memory Type
GDDR6
Memory Bus Width
384-bit
Memory Bandwidth
768.0 GB/s
Performance
Pixel Fillrate
172.8 GPixels/s
Texture Fillrate
604.8 GTexel/s
SPFP Performance
38.7 TFLOPS
Performance per W
129 GFLOPS/W
Performance per mm2
61.6 GFLOPS/mm2
Physical
Interface
PCI-Express 4.0 x16
Typical Thermal Design Power
300 W
SLI Support
2-way
API Support
DirectX
12.2
Vulkan
1.2
OpenGL
4.6
OpenCL
3.0
| Model | GPU Model | Unified Cores | Boost Clock | Memory Clock | Memory Config. | |
|---|---|---|---|---|---|---|
 | NVIDIA A40 | GA102 | TBC | TBC | TBC | TBC |
 | NVIDIA RTX A6000 | GA102 | ||||
 | NVIDIA GeForce RTX 3090 | GA102-300 | ||||
 | NVIDIA GeForce RTX 3080 Ti | GA102-225 | ||||
| | NVIDIA A10 | GA102-890 | ||||
 | NVIDIA GeForce RTX 3080 | GA102-200 | ||||
| | NVIDIA GeForce RTX 3080 LHR | GA102-202 | ||||
| | NVIDIA CMP 90HX | GA102 | TBC | TBC | TBC | TBC |
 | NVIDIA RTX A5000 | GA102 |