NVIDIA Jetson Xavier Development Kit: Under 30 Watts, 8-Core ARMv8.2, 512 Core Volta

[phoronix]

Phoronix: NVIDIA Jetson Xavier Development Kit: Under 30 Watts, 8-Core ARMv8.2, 512 Core Volta

The NVIDIA Jetson Xavier Development Kit is pretty darn exciting with having eight ARMv8.2 cores, a 512-core Volta GPU, 16GB of LPDDR4, and under 30 Watt power use...

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-Jetson-Xavier-Dev-Kit

[sykobee]

Would be interesting to see this go up against one of the Ryzen Embedded Mini-STX boards.

[Jabberwocky]

Hardware looks like it's to die for, the software you'd probably have to hold a gun to my head to use it. Let's not talk about the price!

[squash]

Is Phoronix going to start publishing tensorflow benchmarks? it's more or less usable with AMD now, and it would be interesting to see how a ML environment on the new Jetson board compares to typical desktop hardware.

[Michael]

Is Phoronix going to start publishing tensorflow benchmarks? it's more or less usable with AMD now, and it would be interesting to see how a ML environment on the new Jetson board compares to typical desktop hardware.

Is Phoronix going to start publishing tensorflow benchmarks? it's more or less usable with AMD now, and it would be interesting to see how a ML environment on the new Jetson board compares to typical desktop hardware.

There are one or two tensor benchmarks in PTS, but need to see if there are any other new ones available that could be easily and reliably incorporated.

[hoohoo]

Hardware looks like it's to die for, the software you'd probably have to hold a gun to my head to use it. Let's not talk about the price!

What do you dislike about the software?

[coder]

There are one or two tensor benchmarks in PTS, but need to see if there are any other new ones available that could be easily and reliably incorporated.

TensorFlow?

The tricky part might be the non-free libraries you need for a properly-accelerated Nvidia build. Also, I have no experience with TensorFlow, but other frameworks require separate builds (and sometimes even separate forks) for different hardware backends.

[coder]

Is Phoronix going to start publishing tensorflow benchmarks? it's more or less usable with AMD now, and it would be interesting to see how a ML environment on the new Jetson board compares to typical desktop hardware.

For the same money, a desktop build will be faster.
For general-purpose rendering or GPU compute, a GTX 1050 is faster.
For roughly equivalent inferencing performance, you'd probably need a GTX 1080.

https://en.wikipedia.org/wiki/Tegra#Xavier

Remember, it's main thing is compact & power-efficient compute.

[Michael]

TensorFlow?

The tricky part might be the non-free libraries you need for a properly-accelerated Nvidia build. Also, I have no experience with TensorFlow, but other frameworks require separate builds (and sometimes even separate forks) for different hardware backends.

Yes a basic tensorflow test is present though that one I believe is currently just configured for CPU use http://openbenchmarking.org/test/pts/tensorflow