NVIDIA Jetson TX2 Linux Benchmarks

[schmidtbag]

Ad some comparison with x86 and Core i3 and cheap Rapsberry 3, some Games would be nice too. I know that Michael ignoring Android at all, this board is suited for it.

I agree that comparing to some other platforms would be nice (though the Pi3 is extremely weak in comparison - there are other ARM boards that might be better to compare to) but no, this board is not suitable for Android, at all. It has features that, to my knowledge, Android doesn't use. In fact I haven't yet found anything from Nvidia that mentions Android at all, but they do mention Linux:

http://www.nvidia.com/object/embedded-systems-dev-kits-modules.html

https://developer.nvidia.com/embedded/jetpack

http://elinux.org/Jetson

[rhysk]

I agree that comparing to some other platforms would be nice (though the Pi3 is extremely weak in comparison - there are other ARM boards that might be better to compare to) but no, this board is not suitable for Android, at all. It has features that, to my knowledge, Android doesn't use. In fact I haven't yet found anything from Nvidia that mentions Android at all, but they do mention Linux:

Android works fine on the predecessor Tegra X1. The Nvidia SHIELD TV (2015 and 2017 editions) are Tegra X1-based systems that run Android.

[schmidtbag]

Android works fine on the predecessor Tegra X1. The Nvidia SHIELD TV (2015 and 2017 editions) are Tegra X1-based systems that run Android.

I don't deny that Android can/does run on these devices. I'm aware Tegra in general was targeted toward Android. But these Jetson boards are not suited for it; they are intended for Linux use.

[Xaero_Vincent]

That Jetson is a sexy ARM beast. I'd be interested to see gaming performance with it. Obviously ARM compiled FOSS games.

I wonder how feasible would to get x86 emulation working on it like QEMU or Eltechs. Would be cool to figure out a way to get 3D acceleration working with x86 emulation.

[coder]

I can't help but be a bit disappointed. I expected the GPU performance difference to be on the order of Maxwell vs. Pascal, but it seems about half that? I guess the TX1 was a 20 nm part, IIRC.

This should be compared to Intel's Goldmont i guess... to something like this $99 x86:

https://www.newegg.com/Product/Produ...82E16813157731

Well, it'd be an interesting CPU core comparison, but the GPUs really aren't comparable.

Considering the price of this thing, one of the mobile Kaby Lake SKUs would make a lot more sense. In the iGPU department, I'd still give this thing an edge over anything other than the Iris Pro graphics. TX2 might still win in power-efficiency, though.

IMO, the main reason people would use this board is for robotics or other cases where they need a fast, power-efficient GPU in a small form factor.

[dungeon]

Well, it'd be an interesting CPU core comparison, but the GPUs really aren't comparable.

Yeah but CPU tasks should be somewhat comparable, i expect that it would be mostly slower than that Apollo and somewhere faster i guess... and GPU sounds to me probably somewhat 2.5 maybe 3 times faster than that

For those who wants comparison with RPi3 on CPU see there... somewhat about 4 times faster

http://hackaday.com/2017/03/14/hands...edded-devices/

It is not ideal comparison as PI does not run 64bit arm, but OK probably in best case scenario maybe we can count on 3 times on average at least

[cjcox]

It was fun just taking those few benchmarks through some results on openbenchmarking. Those results show the TX2 being 2 to 3 times faster than the RPi 3 (not 4). Also nice to see where it could beat the AMD 5150 and E1's as well as the lower end older (but not terribly so) Celerons. Would be nice to have a complete set of benchmarks, but this was still good enough for a comparison.

[pipe13]

I can't help but be a bit disappointed. I expected the GPU performance difference to be on the order of Maxwell vs. Pascal, but it seems about half that? I guess the TX1 was a 20 nm part, IIRC.

Well, it'd be an interesting CPU core comparison, but the GPUs really aren't comparable.

Considering the price of this thing, one of the mobile Kaby Lake SKUs would make a lot more sense. In the iGPU department, I'd still give this thing an edge over anything other than the Iris Pro graphics. TX2 might still win in power-efficiency, though.

IMO, the main reason people would use this board is for robotics or other cases where they need a fast, power-efficient GPU in a small form factor.

More, "...they need a fast, power-efficient GPU in a small form factor" with low-latency high-bandwidth IO.

I was speaking to a potential customer about these things over the weak-end. His outfit does SOTA vibration R&D and QA testing, where data collection bandwidth and latency are critical. I don't know if the USB3 on JTX2 will be sufficient for him. It might, but if not the camera port should be if he can tie into it without too much pain.

He probably doesn't need the GPU/Cuda, but any links to the IO aspect of Jetson or other Tegra or other embedded boards will be welcome.

[coder]

More, "...they need a fast, power-efficient GPU in a small form factor" with low-latency high-bandwidth IO.

I was speaking to a potential customer about these things over the weak-end. His outfit does SOTA vibration R&D and QA testing, where data collection bandwidth and latency are critical. I don't know if the USB3 on JTX2 will be sufficient for him. It might, but if not the camera port should be if he can tie into it without too much pain.

He probably doesn't need the GPU/Cuda, but any links to the IO aspect of Jetson or other Tegra or other embedded boards will be welcome.

If you're not going to use the GPU, then I don't really get why you'd use this board. I'd probably just use an x86 embedded board, like Up Board (http://www.up-board.org/up/) or some industrial x86 PC.

BTW, maybe you already saw this, but here's more info (incl. links to datasheets, if you look carefully):

https://devblogs.nvidia.com/parallel...lligence-edge/

[jinova]

Can anyone tell me if this board or the tx1 have opencl support and up to what version? I would buy it in a millisecond if it did.