NVIDIA Jetson AGX Xavier Benchmarks - Incredible Performance On The Edge

[phoronix]

Phoronix: NVIDIA Jetson AGX Xavier Benchmarks - Incredible Performance On The Edge

Each year it's been quite fascinating to see the advance of NVIDIA's Tegra-powered Jetson developer boards with their increasing GPU and CPU capabilities. With the NVIDIA Jetson AGX Xavier that began shipping at the start of this quarter (as well as the AGX Xavier Module now shipping as of this month), there is a tremendous performance upgrade compared to the previous-generation Jetson TX2. I have been benchmarking the Jetson AGX Xavier the past number of weeks and continue to be surprised by its performance potential for relatively low power that makes it suitable for robotics and other AI applications. Here is a bulk of the initial benchmarks I've been running on the NVIDIA Jetson AGX with its 512-core Volta GPU and eight ARMv8.2 Carmel CPU cores.

http://www.phoronix.com/vr.php?view=27320

[enihcam]

At 15W, X86-64 can do a better job.

[ldesnogu]

That's again not enabling all the cores of the TX2 it seems.

phoronix you should read this https://www.jetsonhacks.com/2017/03/...velopment-kit/

Run "nvpmodel 0" before doing any measurements.

[Michael]

That's again not enabling all the cores of the TX2 it seems.

phoronix you should read this https://www.jetsonhacks.com/2017/03/...velopment-kit/

Run "nvpmodel 0" before doing any measurements.

The max performance mode was used on the Jetson boards tested.

[schmidtbag]

And here I thought the TX2 and XU4 had good performance hahaha.

At 15W, X86-64 can do a better job.

Not unanimously.

[csvance]

Hi Michael,

One important metric for autonomous vehicles / robotics applications is using a batch size of one for TensorRT inferences. The results of these inferences are most valuable in real-time, and very often there is only a single camera in question. In applications with more cameras, batch size is usually equal to the number of cameras.

[dungeon]

At 15W, X86-64 can do a better job.

Yeah, this maybe could be compared to let say... against Ryzen embedded V1605B, that have 15W typical TDP.

https://www.reddit.com/r/Amd/comment...d_arrived_amd/

For about $460 as shipping costs from HK.

Or to higher one UDOO Bolt or to Ryzen 5 mobile or whatever

[squash]

Super cool board but wowza my wallet will overrule my excitement!

[AndyChow]

If you play around with tensor things, this is unique. It will easily beat a threadripper 2990wx. I don't like nVidia, but I want this so bad.

[wizard69]

Yeah, this maybe could be compared to let say... against Ryzen embedded V1605B, that have 15W typical TDP.

https://www.reddit.com/r/Amd/comment...d_arrived_amd/

For about $460 as shipping costs from HK.

Or to higher one UDOO Bolt or to Ryzen 5 mobile or whatever

An ARM board of this class can certainly use a few modern X86 implementations for comparison. Personally I’m not seeing a lot of value in this comparison as Raspberry PI is in A PI is in a totally different class.

I still wonder about NVidias ability to sell this chip. It would be interesting to hear about some of the design ins they have achieved for the series.