NVIDIA DGX A100 640GB
rack
8× A100
NVSwitch + NVLink
2020
FP64
0.1
PFLOPS
FP32
0.2
PFLOPS
Power
6.5
kW Total
Memory
640
GB Total
DGX A100 with 640GB total GPU memory
FP64
0.08
PFLOPS
11.9 TFLOPS/kW
FP32
0.16
PFLOPS
24.0 TFLOPS/kW
TF32
1.25
PFLOPS
192.0 TFLOPS/kW
FP16
2.50
PFLOPS
384.0 TFLOPS/kW
System Details
GPU Configuration
GPU Model: NVIDIA
A100
SXM4 80GB
GPU Count: 8 GPUs
Architecture: Ampere
Interconnect: NVSwitch + NVLink
System Specifications
Form Factor: rack
Total Power: 6.5 kW
Total Memory: 640 GB
Memory Bandwidth: 16312 GB/s
Precision Performance Breakdown
| Precision | System Performance | Per GPU | Efficiency |
|---|---|---|---|
| FP64 | 0.078 PFLOPS | 9.7 TFLOPS | 11.9 TFLOPS/kW |
| FP32 | 0.156 PFLOPS | 19.5 TFLOPS | 24.0 TFLOPS/kW |
| TF32 | 1.248 PFLOPS | 156.0 TFLOPS | 192.0 TFLOPS/kW |
| FP16 | 2.496 PFLOPS | 312.0 TFLOPS | 384.0 TFLOPS/kW |
| BF16 | 2.496 PFLOPS | 312.0 TFLOPS | 384.0 TFLOPS/kW |
| INT8 | 4.992 PFLOPS | 624.0 TFLOPS | 768.0 TFLOPS/kW |
Powered by NVIDIA A100
This system utilizes 8 × NVIDIA A100 SXM4 80GB GPUs, each delivering exceptional performance for AI and HPC workloads.
Per GPU TDP
400W
Per GPU Memory
80 GB
Process Node
7nm
Architecture
Ampere
Documentation & Resources
workstation-datasheet-dgx-spark-gtc25-spring-nvidia-us-3716899-web.pdf
NVIDIA • 2025-10-16
NVIDIA DGX Documentation
System guides, deployment resources
Typical Use Cases
Large Language Model Training
Distributed Deep Learning
Multi-GPU Inference
HPC Simulations
Scientific Computing
The NVIDIA DGX A100 640GB with 8× A100 GPUs is designed for enterprise-scale workloads requiring 0.1 PFLOPS of compute power with NVSwitch + NVLink interconnect for efficient multi-GPU communication.