Infrastructure atlas
NVIDIA/Ampere

A100 Tensor Core

The established Ampere workhorse for training, inference, analytics, and HPC, still common across cloud fleets.

GPUAmperetraininginferenceCUDA
A100 Tensor Core hardware

01 / Memory

80 GB HBM2e

SXM or PCIe variants

02 / Bandwidth

2.04 TB/s

80 GB SXM

03 / Power

Up to 400 W

SXM TDP

Precision & peak compute

PrecisionPublished peakBasis
FP16 Tensor312 TFLOPSDense
FP16 Tensor624 TFLOPSSparse
INT8 Tensor624 TOPSDense
INT8 Tensor1,248 TOPSSparse

Peak FLOPS and TOPS are theoretical hardware figures. Sparse and dense modes, chip and system totals, and form factors are not interchangeable.

NVIDIA A100 specifications

Technical specification

Silicon & memory
Architecture
Ampere
Memory
40 / 80 GB HBM2e
FP16 Tensor
312 TFLOPS dense · 624 sparse
MIG
Up to 7 isolated GPU instances
System & software
Form factors
SXM or PCIe
Scale-up
NVLink up to 600 GB/s
Software
CUDA, TensorRT, NCCL, MIG
Best fit
Mature training and inference fleets