Architecture
NVIDIA Ampere 7nm
Most Powerful Variant
GA100-884
Cores : TMUs : ROPs
6912 : 432 : 0
Typical Memory Size
81920 MB
Typical Memory Bus
5120 -bit
GA100 GPU
Overview
Manufacturer
NVIDIA
Release Date
November 16th, 2020
Graphics Processing Unit
Architecture
Ampere
Fabrication Process
7 nm FinFET
Die Size
826 mm2
Transistors Count
54.2B
CUDAs (FP16)
27648
CUDAs (FP64)
3456
Tensor Cores
432
CUDAs
6912
TMUs
432
Clocks
Base Clock
1065 MHz
Memory Configuration
Memory Size
81920 MB
Memory Type
HBM2
Memory Bus Width
5120-bit
Memory Bandwidth
2,048.0 GB/s
Performance
Texture Fillrate
609.1 GTexel/s
SPFP Performance
19.5 TFLOPS
Performance per W
48.7 GFLOPS/W
Performance per mm2
23.6 GFLOPS/mm2
Physical
Interface
SXM 4.0
Typical Thermal Design Power
400 W
API Support
DirectX
12.2
Vulkan
1.2
OpenGL
4.6
OpenCL
2.1
| Model | GPU Model | Unified Cores | Boost Clock | Memory Clock | Memory Config. | |
|---|---|---|---|---|---|---|
 | NVIDIA HGX A100 (16-GPU) | GA100 | ||||
 | NVIDIA DGX A100 | GA100 | ||||
 | NVIDIA HGX A100 (8-GPU) | GA100 | ||||
 | NVIDIA HGX A100 (4-GPU) "Redstone" | GA100 | ||||
 | NVIDIA A100 SXM 80GB | GA100-884 | ||||
 | NVIDIA A100 PCIe 80GB | GA100-883AA | ||||
 | NVIDIA EGX A100 | GA100 | TBC | TBC | TBC | TBC |
 | NVIDIA A100 SXM 40GB | GA100-884 | ||||
| | NVIDIA A100 PCIe 40GB | GA100-883AA | ||||
| | NVIDIA CMP 170HX | GA100-100 | ||||
 | NVIDIA A30 | GA100 |