NVIDIA HGX A100 (16-GPU) Professional Computing Solution

16x NVIDIA Tesla A100

Product status: Released | Report Error
Overview
Manufacturer
NVIDIA
Original Series
Tesla Ampere
Announcement Date
May 14th, 2020
Release Date
May 14th, 2020
Graphics Processing Unit
GPU Model
16× GA100
Architecture
Ampere
Fabrication Process
7 nm FinFET
Die Size
16× 826 mm2
Transistors Count
16× 54.2B
Transistors Density
65.6M TRAN/mm2
CUDAs
16× 6912 (110592)
CUDAs (FP16)
16× 27648 (442368)
CUDAs (FP64)
16× 3456 (55296)
Tensor Cores
16× 432 (6912)
SM
16× 108 (1728)
GPCs
16× 7 (112)
TMUs
16× 432 (6912)
Clocks
Boost Clock
1410 MHz
Memory Clock
1215 MHz
Effective Memory Clock
2430 Mbps
Memory Configuration
Memory Size
16× 40960 (655360) MB
Memory Type
HBM2
Memory Bus Width
16× 5120 (81920)-bit
Memory Bandwidth
16× 1,555.2 (16) GB/s
VideoCardz.com

Physical
Interface
SXM 4.0
Power Connectors
-
Board Power
13000 W
API Support
DirectX
12 Ultimate
Vulkan
1.2
OpenGL
4.6
OpenCL
2.1

Performance
Texture Fillrate
9.7 TTexel/s
SPFP Performance
311.9 TFLOPS
Performance per W
24 GFLOPS/W
Performance per mm2
23.6 GFLOPS/mm2





  Model Clocks Memory
NVIDIA HGX A100 (16-GPU)
110592 Unified Cores
1410 / 1215 MHz 640 GB HBM2
NVIDIA DGX A100
55296 Unified Cores
1410 / 1215 MHz 320 GB HBM2
NVIDIA HGX A100 (8-GPU)
55296 Unified Cores
1410 / 1215 MHz 320 GB HBM2
NVIDIA HGX A100 (4-GPU)
27648 Unified Cores
1410 / 1215 MHz 160 GB HBM2
NVIDIA A100
6912 Unified Cores
1410 / 1215 MHz 40 GB HBM2
NVIDIA EGX A100
6912 Unified Cores
1410 / 1215 MHz 40 GB HBM2
  Model Clocks Memory
NVIDIA HGX A100 (8-GPU)
55296 Cores
1410 / 1215 MHz 320 GB HBM2
NVIDIA HGX A100 (4-GPU)
27648 Cores
1410 / 1215 MHz 160 GB HBM2
NVIDIA HGX A100 (16-GPU)
110592 Cores
1410 / 1215 MHz 640 GB HBM2
NVIDIA EGX A100
6912 Cores
1410 / 1215 MHz 40 GB HBM2
NVIDIA DGX A100
55296 Cores
1410 / 1215 MHz 320 GB HBM2
NVIDIA A100
6912 Cores
1410 / 1215 MHz 40 GB HBM2