ProductA100 / NVIDIA A100

A100 GPU

A data-center GPU from NVIDIA based on the Ampere architecture, widely used for AI training and inference. It supports precision formats including TF32 and BF16, and is available with 40GB or 80GB of HBM2e memory. Its tensor cores and high memory bandwidth make it a common baseline for large-scale deep learning workloads.

Timeline (last 90 days)

Oct 15

At 60% sparsity, AugReg ViT-B/16 achieves up to a 1.80x throughput improvement and a 1.26x reduction in single-query latency on an A100 GPU.

Unverified50%

All Facts (1)

Unverified

At 60% sparsity, AugReg ViT-B/16 achieves up to a 1.80x throughput improvement and a 1.26x reduction in single-query latency on an A100 GPU.

50%

Source Articles

Your Transformer Is Secretly Elastic: Dynamic Compression Without RetrainingOct 15