Ampere data-center GPU; introduced Multi-Instance GPU partitioning.
CUDA cores
6,912
Memory
80GB HBM2e
Bandwidth
2,039GB/s
Power
up to 400 W (SXM)
Overview
- Brand
- NVIDIA
- Segment
- Data center
- Series
- Ampere data center
- Architecture
- Ampere
- Status
- Legacy
- Launch
- 2020
- Launch price (US)
- —
GPU
- Process
- TSMC 7N
- Transistors
- 54.2 billion
- Die size
- 826 mm²
- Streaming multiprocessors
- 108
- CUDA cores
- 6,912
Memory
- Capacity
- 80 GB
- Type
- HBM2e
- Bandwidth
- 2,039 GB/s
Performance
- FP16 / BF16
- 312 TF dense
- FP64
- 9.7 TF (19.5 TF Tensor)
Board & platform
- Power
- up to 400 W (SXM)
Variants & versions
A100 40 GBHBM2 at about 1.55 TB/s.
Common questions
How much memory does the A100 have?
The A100 has 80 GB of HBM2e, giving 2,039 GB/s of bandwidth.
When was the A100 released?
2020.