| 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: 9
|
|
| Author |
|
|
mclaver
Veteran Cruncher Joined: Dec 19, 2005 Post Count: 566 Status: Offline Project Badges:
|
I am sure this has been asked before but I just installed an Intel I7 965 and in most cases my Claimed is signifcantly higher than my granted approaching over 40% higher. I do not see this large of a difference on any of my other machines. IS there something unique about an I7, to casue this?
----------------------------------------![]() ![]() ![]() |
||
|
|
GIBA
Ace Cruncher Joined: Apr 25, 2005 Post Count: 5374 Status: Offline |
I am sure this has been asked before but I just installed an Intel I7 965 and in most cases my Claimed is signifcantly higher than my granted approaching over 40% higher. I do not see this large of a difference on any of my other machines. IS there something unique about an I7, to casue this? Welcome to club, it is a boring fact in WCG overall. Happens same thing with me with mine QX9770 Extreme since last year in all projects and actually especially in CEP (too bad !). When I ran overclocked the difference increase. Only WU's from NRW project appear be more equalized in Claimed vs Granted for me, but many times it not happens too. My differences appear have a little bit less magnitude than yours but it is so huge to not notice. ![]()
Cheers ! GIB@
![]() Join BRASIL - BRAZIL@GRID team and be very happy ! http://www.worldcommunitygrid.org/team/viewTeamInfo.do?teamId=DF99KT5DN1 |
||
|
|
petehardy
Senior Cruncher USA Joined: May 4, 2007 Post Count: 318 Status: Offline Project Badges:
|
Hi mclaver,
----------------------------------------If you look here BOINCstats you'll see that your (I want one) i7 is getting nearly double the amount of credit/week than your other quads. Both my quads - Phenom 9600 XP32 and Phenom 9850 Vista64 overclaim significantly, it's a quad thing! I still want one Pete ![]() "Patience is a virtue", I can't wait to learn it! |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
My guess is that mclaver runs 8 tasks in this i7 and what you see is the effect of hyperthreading on claimed points, at least under Windows. A little before the i7 have been officially delivered I have run some tests in my old P4 HT to see what really happens when using it as a 2-core processor, i.e. when running two WUs together. I was more interested with effects on the global daily throughput than with points but I learned a lot too in this area.
----------------------------------------To make it short, with HCC WUs which are more or less all the same size - one WU alone was taking about 10 hours of runtime (and wallclock time too) and was claiming, say, 70 credits (as in most other machines) - two simultaneous WUs were taking about 15 hours of runtime each (and wallclock too) and were claiming 100 credits. For the throughput this is fine since in 30 hours I can process 3 WUs in single mode and 4 in pseudo dual-core mode. But regarding credits and runtime it is rather unfair because this machine is now getting 2 days of runtime per day instead of one, although it is doing only 25 % more work. Fortunately, for credits the unfairness is lowered because in most cases these WUs are paired with "normal ones" and they are generally getting the "normal" credits (70 in this example) or only a little more. If HT has the same effect in the i7 as in the P4 HT** that explains that you are consistently getting much less credits than you are claiming, but globally more than if you were running only 4 WUs at a time (my P4 has seen its RAC jump considerably). Cheers. Jean. ** From tests posted by Movieman it seems that Windows is not better with runtime in the i7 than in the P4. Since runtime is actually called CPU time a given task should use more or less the same "CPU time" to be processed, whether it takes 10 hours of wallclock or 15. |
||
|
|
mclaver
Veteran Cruncher Joined: Dec 19, 2005 Post Count: 566 Status: Offline Project Badges:
|
You are correct that the I7 is hyperthreading and I have 8 tasks running at 13%/12% each 24/7. Nothing else runs on this machine so it is doing a fair amount of work. I just installed it on 12/21. It is not overclocked. According to BOINCSTATS, it is producing almost twice the work of my Intel Q9450 if I look at credit/week, although there is only one week of data. The average credits have not been stablized yet because it has only been running for a little more a week.
----------------------------------------I am not exactly understanding your example, but is claimed credits artificailly high for an I7, and since you get the same credit as everyone else that processed that WU, the credit granted is "fair" for the work done? ![]() ![]() ![]() |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello mclaver,
I have to wait to see how hyperthreading on the i7 works out, but it is possible that it will claim credit at the full rate for each thread, even though a hyperthreading thread will only run at 60%-65% of full speed. Lawrence |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
but is claimed credits artificailly high for an I7 Clearly, yes, claimed credits with virtual cores like with hyperthreading under Windows is artificially high. My P4 has best benchmarks of 1220 fpops and 2360 iops with one task, and its last one with two tasks is 1100 and 2360 respectively! With runtimes about 50% longer in my case you can figure out the discrepancy! When this machine crunches Rice WUs it is claiming about 70 credits per WU but thanks to the way the real work done is evaluated by knreed's formulas these WUs are consistently getting about 50 credits (which might be already too much, maybe, but I cannot know precisely). If/when the proportion of virtual cores vs real ones is increasing, either because of more i7 running 8 tasks or because members reading this decide to push their old P4 HTs, then the effect of lower granted credits might decrease (when a WU is crunched by 2 members with virtual cores) and there might be some inflation of results. Cheers. Jean. |
||
|
|
mclaver
Veteran Cruncher Joined: Dec 19, 2005 Post Count: 566 Status: Offline Project Badges:
|
So I need to be paired up with another machine hyperthreading and I can get lots of extra credit!!!!!
----------------------------------------Since the Rice seems pretty consistent, all of my work units show cpu time of about 10 hours. The claim credit is between 215 and 233, the granted credit is between 154 and 176. How does this compare to your example of 70 per work unit for rice? ![]() ![]() ![]() |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
That's not what he said, but it's missing the point, anyway. NRW work units take 10 hours on any computer, but are awarded credit according to how much work was actually done in those 10 hours: more for a faster computer, less for a slower one.
|
||
|
|
|