I got jetson just before and tried to execute n-body.
But it seems the performance is not good than I expected.
The performance value is 4.xx Gflops during this simulation.
Is that the normal operation??
Are there any solution or work around to improve this situation?
Many thanks,
With zero effort or tuning, I took an old code of mine and hit 90 GFLOPS on a piece of code that is full of transcendental functions, so I think that the performance of the N-body example means nothing unless what you’re doing matches what it does precisely… I am particularly suspicious of the graphics performance on the current software due to the fact that on my board, the “compiz” process eats about 15% of the CPU even when the machine is totally idle with no graphics draws occuring. I’m hoping that subsequent Jetson OS images will eliminate that and provide us with an X config that looks a little more like what we’re used to on the previous Carma / Kayla boards, or on x86, but we’ll see. In any case, I wouldn’t read too much meaning into the performance of N-body in particular.
Thanks for your comment. I used “top” command and tried to check system resource in my jetson.
During n-body simulations, the CPU resource is 90 %. It seems GPU acceleration could not work correctly.
Anyway I should reset my board and try to install each package and settings.
I doubt that resetting or reloading anything will make any difference, I think that you’re just seeing some of the impact of the particular windowing system code in the current version of the system software, which is also related to the “compiz” thing I mentioned above. I am hoping all of this will get better in subsequent versions of the L4T OS distribution for Jetson.