Académique Documents
Professionnel Documents
Culture Documents
DESIGNED FOR
DESIGNED FOR DESIGNED FOR
Deep Learning Deep Learning Inference and Video
HPC and Deep Learning
Trancoding
Up to Up to
3X
faster time-to-solution
4X
higher throughput for
40X
higher energy efficiency than CPUs
over P100 mixed workloads for inference
Ultimate deep learning training Low power, low profile optimized for scale
Most versatility for mixed HPC workloads
performance out deep learning inference deployment
>> 125 TeraFLOPS of tensor operations for >> 112 TeraFLOPS of tensor operations for >> 22 TeraOPS of INT8 inference
deep learning deep learning performance
>> 15.7 TeraFLOPS of single-precision >> 14 TeraFLOPS of single-precision >> 5.5 TeraFLOPS of single-precision
performance performance performance
>> 7.8 TeraFLOPS of half- precision >> 7 TeraFLOPS of half-precision >> 1 decode and 2 encode video engines
performance performance >> 50 W/75 W power
>> 300 GB/s NVIDIA NVLink Interconnect >> 900 GB/s memory bandwidth >> Low profile form factor
>> 900 GB/s memory bandwidth >> 16 GB of HBM2 memory
>> 16 GB of HBM2 memory
8-way hybrid cube mesh (HGX) 2-4 GPUs per node 1-2 GPUs per node
© 2017 NVIDIA Corporation. All rights reserved. NVIDIA, the NVIDIA logo, and Tesla are trademarks and/or registered trademarks of NVIDIA Corporation in the
U.S. and other countries. All other trademarks and copyrights are the property of their respective owners.