Benchmarks Of The $129 8-Core 64-bit ARM Development Board

[phoronix]

Phoronix: Benchmarks Of The $129 8-Core 64-bit ARM Development Board

Announced last month was the HiKey 8-core 64-bit ARM development board being based upon the HiSilicon Cortex-A53 SoC. This HiKey board came out of 96Boards as the first certified board by the Linaro Community Board Group. I happen to have some early benchmarks of this eight-core AArch64 development board running Linaro/Debian...

http://www.phoronix.com/scan.php?page=news_item&px=HiKey-8-Core-64-bit-ARM-Bench

[sykobee]

Is there a nice simple single-page link to compare this board with the R-Pi, C120, Odroid, etc?

[ARM SoC]

I am not sure those benches were compiled correctly (especially LAMMPS). Memory contention alone wouldn't explain the discrepancy. Core count maybe... We will dig into the codes over here and see what we can find out. Note that many of the codes, if you want to compare ARMv7 runs, won't compile right now. HPCC for instance. Probably just config issues. Stay tuned...

[uid313]

Theme

What is the theme featured on the screenshot in the article?

[jhenke]

What is the theme featured on the screenshot in the article?

Looks like GNOME Terminal in dark theme mode.

[uid313]

Looks like GNOME Terminal in dark theme mode.

Yeah, but I was wondering about the GTK theme.

[lixiao]

Typo

It's a Kirin 620 SOC, not 6220.

[darkblu]

I am not sure those benches were compiled correctly (especially LAMMPS). Memory contention alone wouldn't explain the discrepancy. Core count maybe... We will dig into the codes over here and see what we can find out. Note that many of the codes, if you want to compare ARMv7 runs, won't compile right now. HPCC for instance. Probably just config issues. Stay tuned...

First rule of benchmarking when results come surprising in some shape or form, is to verify the benchmark run is valid. That includes verifying the generated code (if possible, since that requires sufficient ISA knowledge) and trying to reproduce the results with different tools, like a different compiler version, or a different compiler altogether (hint: clang). Of course all that takes time, and runs against the busy schedules of most benchmarking sites.