A Quick Test Of NVIDIA's "Carmel" CPU Performance

[AlexSt]

First of all sorry for not directly using OpenBenchmarking, I need to learn it first :-)

On my TX2 [with all 6 cores enabled and set to max speed] I get this: 386 seconds
If I get back to default mode [with only 4 cores, the 2 Denver cores disabled, clock setting might be maxed]: 568 (4 threads) to 539 (16 threads) seconds

i think Michael should be (in theory) be quite aware of the power-save-to-performance options that system does offer when he was doing his testing. but you are right - as the bar graphs on page 2 of the article are indicating a value of 675 seconds for the TX2 there is something not plausible with either your or his numbers. might it be the version of the benchmark you used on your box?

[AlexSt]

i am missing the primesieve numbers for Tinkerboard and TX2. As those boards are otherwise relatively close to the device under test those missing test numbers would be good helpers for getting the full picture.

as TX2 has a mixed core architecture and further Xavier has an advanced version of them it might be further very interesting to run the various single cores against each other, not only full SoCs. As clock rate is tune-able thus a clock rate variety also might help - because often only the core clock changes but not necessarily the speed of the memory interface. Some folks might even be interested in how power consumption is affected by various such sleep mode and other power consumption affecting changes. (e.g. drone and robot usage, trimmed for a better run time...)

[buggystick]

Recently got my Jetson AGX Xavier delivered and my results are quite a bit better than this initial quick test. Hopefully Michael's upcoming review will shed some light on this.

https://openbenchmarking.org/result/1812170-SK-1809248RA04

[coder]

Recently got my Jetson AGX Xavier delivered and my results are quite a bit better than this initial quick test. Hopefully Michael's upcoming review will shed some light on this.

https://openbenchmarking.org/result/1812170-SK-1809248RA04

Thanks, but to what does "w/L4T 31.1" refer?