| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
| World Community Grid Forums
|
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 23
|
|
| Author |
|
|
MAB The Frisian
Cruncher Joined: Oct 12, 2008 Post Count: 22 Status: Offline |
I wonder which OS delivers the highest average.
I'm using W2K on my dual core systems, but heard that running XP or higher gets more out of the procs. Does that mean the Boinc-client runs faster on XP or higher and therefore gets more results during the same amount of time ? And which should I choose ? XP 32, XP 64 or Vista ? If there is already a topic about this, I 'd appreciate the link. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The OS doesn't make a significant difference.
It's very hard to measure performance accurately, because of a number of factors: different projects have different performance characteristics, and each work unit is different. It is impossible to compare the same work unit on different operating systems under controlled conditions. This won't stop many people giving you their informed and uninformed opinions, though, based on their own observations. The bottom line is this: the uncertainties are too great for any minor difference to be measurable. CPU and memory have a far greater effect. If you do notice a discrepancy, it probably means something is amiss. Please tell us if you notice anything like this. |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
I wonder which OS delivers the highest average. It depends if you are thinking of returning more credits or of actually doing more work. If you are thinking of producing more work there seems to be a much better performance (about 30 %) in 64-bit mode for HCC WUs. I observed it under Ubuntu (Linux) but I think that it should be the same with 64-bit Windows, the reason being that, at least for some of the 64-bit processors, the processor is much more efficient on integer computation than in 32-bit mode. Since there is no noticeable difference of performance between 64 and 32 modes for floating point computations WUs which are running essentially in floating point mode (all except HCC) do not show a noticeable difference. However, if you are thinking of getting more credits it is no longer as beautiful. For various reasons which are difficult to overcome work done under Linux systems is often less credited than it would be under Windows, therefore the benefit is mitigated for HCC in 64-bit mode, and it would probably be negative in 32-bit mode as well as for all other WUs in both modes. Does that mean the Boinc-client runs faster... Even if it were running 100 times faster that would not matter because the Boinc client is only organizing and managing the work and uses very little CPU power. The hard work is done by the science applications, and their performance depends on the type of computation they are doing and how efficiently they have been designed and coded (see above). And, as said by Didactylos, everything I have written here is based on observation (much, but still...) and reasoning, but not on inside knowledge of the science application or the systems. And not on exhaustive testing as well since it is not possible to run the same WU in different environments under normal operating conditions. Edit: Clearly said, "based on much observation" means statistical observation on lots of results produced under Ubuntu 64-bit and XP 32-bit. Cheers. Jean. ---------------------------------------- [Edit 1 times, last edit by JmBoullier at Oct 15, 2008 12:23:59 PM] |
||
|
|
MAB The Frisian
Cruncher Joined: Oct 12, 2008 Post Count: 22 Status: Offline |
It's not that I wouldn't like more credits
---------------------------------------- , but the most important thing is I get as much work done as efficient as possible with my systems.I use a GA-MA78GM-S2H / 4050e and a K9A2GM-FIH / 4850e both with 2Gig DDR2 and a 80G HD. At the moment they both run W2K. If it would make a difference for WCG to run XP 64 on both systems I would kill two flyes at once cause XP makes better use of the procs. [Edit 1 times, last edit by MAB The Frisian at Oct 15, 2008 1:46:08 PM] |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
If you feel ready for the move, just do it, if I am still allowed to say.
---------------------------------------- When they get HCC WUs they will love it, and for the other projects they should do at least as well as with W2K, I think. Good luck for the move. Jean. |
||
|
|
MAB The Frisian
Cruncher Joined: Oct 12, 2008 Post Count: 22 Status: Offline |
Perhaps it's an idea to collect/post your floating points and integer MIPS per OS/processor combination ?
----------------------------------------For instance : W2K/4050e = 2045 floating point MIPS / 3651 integer MIPS. W2K/4850e = 2534 ............................. 4537 ................... Suppose other things have a certain influence but you could use it as an indication. Start up in save mode and test ? K9A2GM-FIH / 4850e (save mode) running W2K/CPU-Control/Outpost gives : 2542/4670 GA-MA78GM-S2H / 4050e (save mode) running XP Pro X64/CPU-Control/Outpost gives : 2130/3870 [Edit 3 times, last edit by MAB The Frisian at Oct 29, 2008 5:53:52 PM] |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
Just two quick comments (I am gnawing an other big bone right now
---------------------------------------- ):1. You would have to add the clock frequency to the OS/processor combo... 2. I have seen this kind of effort started on the site of a very big team with lots of members more excited with getting credits than with what they are computing. And even there it sinked into oblivion after a few posts: too many particular cases make any consolidation/organization effort not sustainable. Cheers. Jean. |
||
|
|
MAB The Frisian
Cruncher Joined: Oct 12, 2008 Post Count: 22 Status: Offline |
Mine are running at standard speeds. Bought these procs to save energy.
----------------------------------------But yes, it has to be posted as well. And it's not necessary to post a lot of data. Just enough to validate the different results and show the difference in a chart. [Edit 3 times, last edit by MAB The Frisian at Oct 16, 2008 5:24:57 PM] |
||
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
Just two quick comments (I am gnawing an other big bone right now ):1. You would have to add the clock frequency to the OS/processor combo... 2. I have seen this kind of effort started on the site of a very big team with lots of members more excited with getting credits than with what they are computing. And even there it sinked into oblivion after a few posts: too many particular cases make any consolidation/organization effort not sustainable. Cheers. Jean. Hello Jean; I don't know if you were talking about us at Xs but even if not I'd like to comment on it. You get many different kinds of people on the teams. The purists who don't care about points at all but are just there for the work. Those are a minority from what I've seen. Good folks but overall in the minority. Then you see the people whose first motivation is the work but also like the thrill of competition. Those compose the majority and even in that group there are differences in the amount that an individual is "into" the competitions. I'd toss myself into that group. I'm here for the work but I do enjoy competing but not as an end all. In the last 2 years I've probably put over 5 million points on other members and other teams names in the spirit of good will or in the case of one team here to say that I was touched by the loss of a child. Then last you do have a small group that would run an app that counted tiddly-winks if it gave points. These programs I consider worthless but it is not my place in life to tell others what to do. I can only suggest and do my sales pitch for WCG and hope some listen. Now as to what windows OS score better than others it seems from what I've seen that Vista Ultimate 64 and win2k8 64 score the best. I run neither at this time. Vista is persona non grata in my home. Too much work to make it into what it should have been when released. I run winxp 32 on two machines, win2K3 server on the two clovers and winXP 64 on my harpertown machine. I'll have a Nehalem machine up and running tomorrow but unfortunately I can't put it on WCG until the NDA lifts on November 16 as Boinc grabs numbers on the processors efficiency and lists them here even when your machines are "hidden": http://boincstats.com/stats/host_cpu_stats.php?pr=wcg&st=0 If you know of a way to get around that I'd be interested in hearing it as this machine is a beast in terms of computational power but I did agree not to post numbers or pictures until the NDA lifts. Thanks for reading. ![]() |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
Hello Movieman!
----------------------------------------Oh no, I was not talking of the XS team. I was talking of a very big team with very big results (both at global level) and not of a medium size team with big results. I am kidding you but it's true, I am not following XS's forum.My point was not to give an opinion on these "credit crunchers", it was to say that even people as motivated as they are quickly give up on making such a consolidation. MAB The Frisian said it's not necessary to post a lot of data. Just enough to validate the different results and show the difference in a chart. Regarding our motivations it seems that we have several things in common. I would toss myself into the same group as yours and I don't put the blame on credit crunchers (there are not that many anyway, one or two scores maximum in this very big team). And I wish to avoid Vista as long as reasonably possible, which is why I could only make assumptions that 64-bit Windows should show very good work throughput AND good crediting too. Which you confirmed, thanks. Last, about the confidentiality of your machines I don't think you are right, but if you don't want to take any risk it is certainly safer to wait until the end of your NDA. I think that machines you can see in BoincStats even when you have asked WCG to keep them hidden are machines which have crunched at least once on other Boinc grids. Those which have always crunched for WCG only are still not visible in BoincStats. I have just checked with several big members of my team (where most members are WCG only) and either I could not see any machine at all, or I could only see one or two out of much bigger farms. If you asked WCG to keep your machines hidden it does not put machine information in the files it sends to BoincStats, and I do not see how BoincStats would be able to pick it from WCG in that case. Cheers. Jean. |
||
|
|
|