HeroBox Celeron J4125 single core performance good

HeroBox Celeron J4125 single core performance is twice that of a Raspberry Pi400, and only 60% worse than Intel i7-8650U CPU, at least for the “compute prime magic square 1 million times” I measured. Details in this posting:

Especially the only 60% worse than i7 CPU surprised me, since two single core benchmarks I based my HeroBox buying decision on showed 100% worse than i7:

I bought HeroBox for headless Intel single core cpu applications (like prime factorization, or my planar graph playground), and am positively surprised from CPU power sofar.

I never heard HeroBox fans sofar, neither under Win10, nor under Ubuntu 22.04 Desktop, although I pushed cores with more than 10 minutes of cpuburn, see “A problem with fans” thread in this forum, I am restricted to 2 URLs as new user.

While I plan to use HeroBox mostly headless, Ubuntu 22.04 Desktop looks nice on 1368x768 HDMI monitor, after running “gnome-control-center” terminal command and correcting display resolution to 1368x768.

Performance varies from app to app. In this case, prime factorization is mostly a CPU-bound benchmark, while there are other use cases when memory bandwidth is more important. In these cases, the difference is much bigger in favor of intel core processors.

@HermannSW You mention that you are using your Herobox headless. I was planning on doing the same, but hesitated to buy the Herobox because I read in the forum that it does not support auto-poweron after power failure/recovery (see HeroBox doesn't wake up automatically
BIOS Autostart CHUWI HeroBox )

Is that feature working for you, or you need to turn it on manually after it loses power?

@huantx HeroBox did not start on powering off and then on.
I entered Bios and changed under Boot tab to “Power on”.
Now after powercycling HeroBox boots as soon as it gets power.
So no problem with remote headless work.

This is from a photo of Bios Screen I took when I got HeroBox:

The search for prime magic square has all primes precomputed, basically the code is a huge number of nested loops testing for a 3x3 magic square. So yes, memory is not a big problem, and therefore J4125 behaves better.

I just tested 6-coloring (planar graphs always have a 4 coloring, but 6-coloring is simple algorithm needing linear time only) on a 500,000 vertices maximal planar graph:

i7 8 seconds:

$ time ./c++_6coloring ../graphs/500000.a 2>err >> out

real	0m7.939s
user	0m7.794s
sys	0m0.127s

J4125 14.7 seconds, 83.75% slower than i7:

hermann@j4125:~/planar_graph_playground/c++$ time ./c++_6coloring ../graphs/500000.a 2>err >> out

real	0m14.701s
user	0m14.449s
sys	0m0.249s

So J4125 is worse than “60% worse than i7” before, but still better than “100% worse than i7” according the two single core performance benchmarks, and that for 168$ only: