While digging around for some information, I came across this link. It claimed the C2 would only be running at 1.536 GHz, since at the two faster modes benchmark results (they are compiling code) would be the same and no further increasing.
I did some tests with 7z and sysbench (primes up to 10000) myself and am unfortunately experiencing the same issue.
I confirmed this on two systems, updated to the latest official Ubuntu levels. Frequency was set with cpufreq-set -u $FREQ and queried with cpufreq-info.
Please find my results below:
- Code: Select all
7z b, total MIPS
avg. of 3 x sysbench --num-threads=4 --test=cpu run
kHZ seconds @4threads seconds @1threads
100000 40 151
250000 15.3 61
500000 7.6 30
1000000 3.76 14.9
1296000 2.9 11.5
1536000 2.46 9.64
1752000 2.43 9.7
2016000 2.43 9.64
While the results scale almost linearly between 0.25, 0.5, 1 and 1.5GHz, they hit a wall there.
Regardless if run on 4 or just 1 threads.
Quickly scanning the forums here I couldn't find anything on the topic yet, just Exploring CPU clock speeds on the C2, trying to overclock to 2.2GHz with no visible effect, now making sense in that light...
Any idea whats going one here?
Are the devices really running at 2GHz? If so, what is that barrier at 1.5GHz, why is there no performance increase at the last two clock settings?
Running just one thread didn't benefit, I'd say the prime computation isn't bound by memory bandwidth, but what else?
I own 8 C2 myself, in our BOINC team we have dozens of them computing for science. Of course we are trying to maximize their performance. Usually these are actively cooled and could certainly run at a bit more voltage (if needed) and full throttle. If they could deliver some 20-30% more performance, by benefiting from the last two clock steps, that would be awesome...