| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Suggestions / Feedback
Thread: Project differences, benchmarks and points |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 24
|
|
| Author |
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Just my 2 cents of authority, the rest you can slug out with your antipode:
----------------------------------------Re this and the other thread: http://www.worldcommunitygrid.org/forums/wcg/...amp;lastpage=yes#lastpost - rDCF is badly affected by any one short series of highly variable run times (See the FAQ's). Depending on the combination of projects, particular on multicore, it can shoot up rapidly. - Points per hour claim would only be accurate if 100% CPU efficiency is reached. BOINC client will claim anything dumbly based on recorded time versus benchmark. The project servers look at the efficiency. If running multiple CEP's simultaneous, it will effect e.g. FAAH, while running simultaneous. This is simply because hardware has limits on certain resources, thus queueing kicking in. Some hardware is better at integer some at float, in combination with the particular OS. Now [a long essay could be inserted here], but read the developers are toying with showing 'elapsed' time from 6.6. At least on the alpha I have there is a difference looking in the properties view. We can stand here and stamp our feet and demand 100% efficiency, but you wont get it. It's not practical to code ad infinitum, and near impossible to achieve due the widest variety of participating hardware/os combinations. Kremmen your list means little to me. I've no idea what hardware is doing the job. If you get 51% on HCC, something is either consistently badly overclaiming or your computer is basically a sure dog at integers. End of my 2 cents (and I get 113% in mean on RICE, off my quad  )
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Mar 16, 2009 8:57:29 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Kremmen your list means little to me. I've no idea what hardware is doing the job. Yes, you do, because I stated that up front. They are all P3 or Athlon machines. To be more specific, mostly P3/800, P3/866 and P3/1000. The 800 and 866 figures are slightly better than the P3/1000, by varying degrees. (I would assume because the FSB is the same and anything that hammers memory will therefore get very little gain using a 1GHz machine over 800MHz, even though the benchmark figure will be higher.) Data is from machines when they were running boinc close to 100% of the time. I have no multi-processor machines any more, but used to have a twin-P3/866 machine. Obviously, the figures between projects will differ. My point (as asked at the beginning) was that if we knew what behaviour causes the huge differences between projects, we'd get an idea of which projects are better suited to which hardware. e.g. If we know that a particular project accesses memory a great deal, a fast FSB may be more important than a higher clock speed. Alternatively, one can check empirically for a particular family of hardware, as I've done. |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges: 
|
OK, we know more about the hardware.
----------------------------------------What about the software now? OS family, 32/64-bit mode, Boinc version... After everything which has been written on the subject of crediting differences in this forum it is surprising that you didn't realize that these factors matter much... Read you later. Jean. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
What about the software now? OS family, 32/64-bit mode, Boinc version... Various Debian linux versions, various Boinc versions. No significant differences occur as a result of OS, including when I've upgraded from 2.4 to 2.6 kernels. My results are for the benchmark results from 5.8.15 as it was available from this site. The only difference that I've observed between Boinc versions, as regards points, is that benchmark results vary a great deal depending on where Boinc comes from (i.e. how is was compiled). However, this proportionately changes all results equally. (... unless it puts them so far out of sync with everyone else as to trigger a different points calculation method. I've always ignored such outliers in the averages above.) |
||
|
|
|