Calxeda ARM Multi-Core Scaling Linux Performance

[phoronix]

Phoronix: Calxeda ARM Multi-Core Scaling Linux Performance

After delivering some AMD Vishera multi-core scaling benchmarks for reference on Saturday, here's some similar tests conducted from a Calxeda ECX-1000 quad-core ARM server node...

http://www.phoronix.com/vr.php?view=MTIxNjY

[FourDMusic]

Could someone explain to me why in the CLOMP benchmark, if you normalize the speed up by the number of cores, the ratio grows with the number of cores? I would have thought it would decrease due to overhead.

[chuckula]

Could someone explain to me why in the CLOMP benchmark, if you normalize the speed up by the number of cores, the ratio grows with the number of cores? I would have thought it would decrease due to overhead.

CLOMP seems to be a weird outlier especially when you look at the other benchmarks and you see that none of them come even close to perfect 4x scaling (which is to be expected due to overhead).

Larabel's methods are not always that precise. Additionally, since these chips basically exist for the sole purpose of being cheap web-servers, none of the benchmarks that he's posted are even remotely relevant to what these servers will actually do in the real world. Some Apache/Nginx, SSL, and database benchmarks would be the only things that would count.

[kathlyyoun]

Could someone explain to me why in the CLOMP benchmark