What is TOPS of Tx2 board?

Hi I have a Tx2 Board and am SW engineer.
So I want to provide my customer with the performance of Tx2.

So What is TOPS of Tx2 which is processed per second?
(* TOPS is Tera Operations Per Second.)

Hi sfireman,

Please refer to Technical Specifications from: https://developer.nvidia.com/embedded/develop/hardware

Thank you for your answer.

I have reviewed TX2 performance from “https://developer.nvidia.com/embedded/develop/hardware” and it is 1.33 TFLOPS of TX2.

My customers want to know the TOPS of TX2, not TFLOPS. So They want to compare between TX2 and others.

Please share the TOPS information of TX2.

TX2 doesn’t support INT8, so the TX2 performance is given in TFLOPS (FP16). Xavier supports INT8, so Xavier performance is shown as TOPS. You can compare the two in the same way.

Further to your reply, I have some questions below.

If we can compare FLOPS and TOPS in the same way like you said, please check if the sentence is correct below.

[Exemples]
A. Nvidia TX2 Board : 1.33 TFLOPS
B. Another Board : 1.8 TOPS

The performance of B is better because the B board has a higher value than the A board.

[Question-1] Could we compare the performance between two boards like this way?
[Question-2] Please let me know what tools measure the TOPS and TFLOPS.

Hi sfireman,

In general, the performance can be represent theoretically to X TFLOPS for FP16 or 2X TOPS for INT8 if both type are supported, but due to TX2 can’t support INT8, so don’t have the data.
To do the comparison, you may need to have the same benchmark for FP16 for both TX2 and another board or try to measure the deep learning inference throughput (images per second), such as GoogLeNet deep image recognition network to compare.

Hi kayccc,

Thank you for your kind description.
There are the TOPS among your XAVIER products and it is the values below.

  1. XAVIER NX : 21 TOPS
  2. AGX XAVIER 8GB : 20 TOPS
  3. AGX XAVIER : 32 TOPS

Do you have the data that measured TFLOPS among the three products above?
If you have, please share the TFLOPS data. And then I could explain the difference of performance to my customer.

Hi sfireman,

Please find the Technical Specifications by clicking the “COMPARE MODULE” from AI-Powered Autonomous Machines at Scale | NVIDIA Jetson AGX Xavier

Thanks