PNY NVIDIA A100 40GB Graphics Card (TCSA100M-PB)

Architecture     

Ampere

Process Size      

7nm | TSMC

Transistors         

54 Billion

Die Size

826 mm2

CUDA Cores      

6912

Streaming Multiprocessors

108

Tensor Cores | Gen 3    

432

Multi-Instance GPU (MIG) Support         

Yes, up to seven instances per GPU

FP64     

9.7 TFLOPS

FP64 Tensor Core           

19.5 TFLOPS

FP32     

19.5 TFLOPS

TF32 Tensor Core            

156 TFLOPS | 312 TFLOPS*

BFLOAT16 Tensor Core 

312 TFLOPS | 624 TFLOPS*

FP16 Tensor Core           

312 TFLOPS | 624 TFLOPS*

INT8 Tensor Core            

624 TOPS | 1248 TOPS*

INT4 Tensor Core            

1248 TOPS | 2496 TOPS*

NVLink 2-Way Low Profile,

2-Slot

NVLink Interconnect     

600 GB/s Bidirectional

GPU Memory   

40 GB HBM2e

Memory Interface          

5120-bit

Memory Bandwidth      

1555 GB/s

System Interface             

PCIe 4.0 x16

Thermal Solution            

Passive

vGPU Support  

NVIDIA Virtual Compute Server with MIG support

Secure and Measured Boot Hardware Root of Trust       

CEC 1712

NEBS Ready       

Level 3

Power Connector           

8-pin CPU

Maximum Power Consumption

250 W