Where to find Xavier Performance in FLOPs?

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
#1

Hi,

I like to find some material with GPU performance for Jeston Xavier.
Where can I find it?

NVIDIA announces Volta V100 runs single precision as like https://www.nvidia.com/en-us/data-center/tesla-v100/.
NVIDIA TESLA V100 SPECIFICATIONS

Tesla V100 for NVLink
Tesla V100 for PCIe
PERFORMANCE
with NVIDIA GPU Boost™
DOUBLE-PRECISION
7.8 teraFLOPS

SINGLE-PRECISION
15.7 teraFLOPS

DEEP LEARNING
125 teraFLOPS

DOUBLE-PRECISION
7 teraFLOPS

SINGLE-PRECISION
14 teraFLOPS

DEEP LEARNING
112 teraFLOPS

Thanks in advance

#2

Hi chjung90, please refer to this FAQ entry: https://developer.nvidia.com/embedded/faq#xavier-performance

The Jetson AGX Xavier webinar also contains performance information:

https://devtalk.nvidia.com/default/topic/1041822/jetson-agx-xavier/nvidia-webinar-mdash-jetson-agx-xavier-and-the-new-era-of-autonomous-machines/

#3

Hi chjung90

Additionally, we have gathered a bunch of different sources and summarized them in our developer’s wiki. You can find information on the Xavier GPU (and other processors) here:


https://developer.ridgerun.com/wiki/index.php?title=Xavier/Processors/GPU/Description

Hope it helps.

#4

My empirical measurements on Xavier AGX with CUBLAS SGEMM and DGEMM suggest a peak FP32 rate of ~1.4 TF/s and a peak FP64 rate of 44 GF/s, which correspond to 2 and 64 FMA/clock/SM. This is the same ratio as TU102 and very different from GV100.

I can’t find this publicly documented anywhere.