inference/research
verifiedhigh · 0.90updated 2026-10-10

NVIDIA A100 Tensor Core GPU datasheet

spec-sheet

Fields

id
sources/a100canon-nv-a100-datasheet
Publisher
NVIDIA
Kind
spec-sheet
Published
2022-05
Accessed
2026-10-03
notes
Capacity-unstated A100 copy of [[sources/nv-a100-datasheet]], added 2026-10-10 for [[accelerators/nvidia-a100-sxm]]. SAME DOCUMENT, SAME URL. Covers A100 80GB PCIe and 80GB SXM4. Its 80GB SXM column prints the SAME compute figures as the Ampere white paper's A100 column (FP64 9.7, FP64 Tensor 19.5, FP32 19.5, TF32 156, BF16/FP16 312, INT8 624), which is the second independent first-party statement that the compute is capacity-independent and is why the capacity-unstated record can state those values. All Tensor Core figures are the with-sparsity numbers; the dense figure is half. The sheet does NOT print a memory bus width, only the aggregate 2039 GB/s.

Notes

Capacity-unstated A100 copy of [[sources/nv-a100-datasheet]], added 2026-10-10 for [[accelerators/nvidia-a100-sxm]]. SAME DOCUMENT, SAME URL. Covers A100 80GB PCIe and 80GB SXM4. Its 80GB SXM column prints the SAME compute figures as the Ampere white paper's A100 column (FP64 9.7, FP64 Tensor 19.5, FP32 19.5, TF32 156, BF16/FP16 312, INT8 624), which is the second independent first-party statement that the compute is capacity-independent and is why the capacity-unstated record can state those values. All Tensor Core figures are the with-sparsity numbers; the dense figure is half. The sheet does NOT print a memory bus width, only the aggregate 2039 GB/s.

Cross-references

This record references no other records.