| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: FightAIDS@Home
Thread: Windows 7 / 64 bit / Lower Credit? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 60
|
|
| Author |
|
|
Hypernova
Master Cruncher Audaces Fortuna Juvat ! Vaud - Switzerland Joined: Dec 16, 2008 Post Count: 1908 Status: Offline Project Badges: 
|
I dunno if it can help, but I have run until Gold on FAAH with machines on Win7 64 bit and I7 950 or 975 CPU's. What I do as these machines will run 24/7, is to deactivate already in the motherboard Bios (Asus 1366 motherboards) all settings that correspond to power or thermal optimization.
----------------------------------------In Win7 Settings I put everything to full performance with no limits. Never had problems.  [Edit 1 times, last edit by Hypernova at Feb 4, 2010 12:02:52 PM] |
||
|
|
Highwire
Cruncher Joined: Aug 18, 2006 Post Count: 39 Status: Offline Project Badges: 
|
Hmm can't imagine, unless Windows has changed way they manage process scheduling very significantly (and in a way that seems to trouble FAAH specifically), how BIOS could affect this.
When you say 'no problems' : you might not realise you've problems unless you were looking for them. What sort of time / claim / given do you have? Your I7 950 would be ballpark with mine I'd imagine, which for 32 bit Boinc gives benchmarks ballpark ~ 3170 / 6780 at current clock (fluctuates a little of course). I'm just running HCC now on it so watching to see how machine credit responds. |
||
|
|
JmBoullier
Former Community Advisor Normandy - France Joined: Jan 26, 2007 Post Count: 3716 Status: Offline Project Badges:
|
What is suggested by Hypernova is to force the processor's frequency at its nominal value at BIOS level. I don't know if that will eliminate your problem (personally I bet no) but that will eliminate a range of possible issues like power management etc...
----------------------------------------Check your BIOS advanced settings for various parameters which could influence. For example in my motherboard I have set the frequency multiplier to "Manual" and at its nominal value 9. This way "nobody" will ever be able to play with frequency at software level. Seen from my chair the most visible difference between HCC and FAAH WUs at system level is the necessary RAM. HCC WUs need little RAM and FAAH much more. Unfortunately I see no information at all about RAM in your machine in all this thread. Maybe it is time to post the start up messages of your client here, that would avoid that we ask you pieces of information one by one and, who knows, that could also allow any reader of this thread to notice a clue. Read you later. Jean. |
||
|
|
Highwire
Cruncher Joined: Aug 18, 2006 Post Count: 39 Status: Offline Project Badges:
|
Well my bios settings were already tweaked for a small overclock, it's a q9550 processor + gigabyte motherboard. Given the CPU temps were high I'm hard to convince that whatever core was running FAAH only, under W7 only (remember this ran for .. > year before W7 fine), running slowly, but showing 100% core use in Task Manager? I'll double check next time I reboot. I didn't change them when I installed W7. I mean is it even possible to lower the clock on some cores while keeping others full?
Memory: I've 4GB of good ram in there, so there is actually more memory available in 64 bit - I doubt the swap file has had much use. Startup (is on other computer) .. 5.10.45 .. 4gb/8gb. . interestingly that's still showing old OS name, as I kept BOINC folder on same path / drive and just fire it up on either OS as is dual boot. surely doesn't require full install again - it's mid jobs. Surely FAAH doesn't somehow look at the OS the client seen one time in history (it could easily check itself if it looked) and behave massively differently? I'd be very surprised if it did that - that'd be pretty wierd. I'm certainly not only person in world who dual boots boinc. flags .. fpu tsc pae nx sse sse2 pni mmx I'm just so surprised that OS seems to have such a bearing on what is I thought a number crunching program where the OS should be fairly irrelevant. |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
Re advanced power settings, max was 100% min was 5%. As to what minimum means, quick search ofthe web and it appears .. nobody knows?! I can't think myself what that means. Max is obvious to anyone familiar with speedstep etc, but minimum?? Well, don't know if this is the correct answer, but would guess it basically means the cpu will throttle-down to 5% for various reasons... ... and one of these reasons is if only idle-threads is using most of the cpu-resources in a system... So, while the cpu doesn't neccessarily throttle-down to 5%, it's a good chance it is throttling part of the time, and this is consistent with your very high cpu-times. So, to fix, set the Min to 100%.  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Highwire
Cruncher Joined: Aug 18, 2006 Post Count: 39 Status: Offline Project Badges:
|
"and one of these reasons is if only idle-threads is using most of the cpu-resources in a system..."
----------------------------------------Well this certainly isn't my understanding of how windows used this sort of setting, idle is just a name for a number, a number used in the base + thread priority algorithm to do with their round robin who gets cpu cycle next allocation. And it doesn't match up with HCC whch runs at same priority, or any of the other evidence, though I could be wrong. They (MS) seem in no hurry to explain it, I downloaded a power management PDF from MS and no explanation. Also if going into low power mode completely screws the claimed time, something is clearly wrong in there. I've upped to 100% but not hopeful. That computer is just running HCC now. In fact .. I tell you what I might throw every task WCG does at it and see which is behaving and which isn't. Edit: Since I switched to HCC / no FAAH credit graph is shooting up again. [Edit 1 times, last edit by Highwire at Feb 4, 2010 5:38:16 PM] |
||
|
|
Highwire
Cruncher Joined: Aug 18, 2006 Post Count: 39 Status: Offline Project Badges:
|
Well the scores are in. There's nothing in my BIOS that I can see relevant to power. Upping the Minimum Processor State from 5% to 100%, as expected, fixed absolutely nothing and has zero effect.
----------------------------------------I've been running every task WCG offers and here are results: Task Hours Claimed / Given CMD2_ 0324-HSP7CA.clustersOccur-1VYH_ C.clustersOccur_ 26_ 115628_ 117267_ 0-- 3.40 68.6 / 67.3 CMD2_ 0324-HSP7CA.clustersOccur-2BXH_ A.clustersOccur_ 27_ 35749_ 36186_ 1-- 2.37 47.9 / 46.3 X0000089021318200707021410_ 1-- 3.49 70.6 / 77.4 X0000089020430200707021424_ 0-- 3.52 71.2 / 69.1 CMD2_ 0321-1NW3_ A.clustersOccur-2C6C_ A.clustersOccur_ 4_ 13202_ 13510_ 0-- W 4.30 87.0 / 100.5 R00527_ 3183d5a3815260f2db8c532ac2985406_ 02_ 003_ 0-- 7.02 141.9 / 159.5 R00526_ f799c737a0300434538116d8d045ef93_ 00_ 001_ 19-- 7.01 141.7 / 156.6 nc819_ 00076_ 19-- 3.91 78.5 / 63.1 CMD2_ 0316-MYH14.clustersOccur-1JK9_ B.clustersOccur_ 27_ 63513_ 65043_ 0-- 4.93 99.0 / 103.2 CMD2_ 0316-RS20A.clustersOccur-2ANW_ A.clustersOccur_ 0_ 11967_ 22838_ 0-- 3.31 66.4 / 64.1 R00526_ 5c0d535bec5af90df9b9362f98e66678_ 02_ 004_ 4-- 7.01 140.7 / 174.5 faah10576_ ZINC31848546_ xMut_ md07200_ 00_ 1-- 11.76 236.2 / 126.8 X0000088991099200706110832_ 0-- 3.61 72.6 / 71.4 X0000088990201200706110847_ 0-- 3.62 72.7 / 72.7 Spot the odd one out? The only one misbehaving is FAAH. It's quite clear that FAAH is the ONLY TASK misbehaving here, on a fresh (2 weeks old) W7 64 bit ultimate install. It was funny watching the FAAH amongst the other tasks, it had a 'To Completion' of about 5.5 hours but while every other task was counting down, it's completion time would just sit there, and came in at 11.76 (remember I'm seeing 20+ hours for these). I ran a a diagnostic on the tasks and the FAAH was reporting as using the same number of CPU cycles per second as the other tasks, and a similar number of CPU cycles total for a task that had run about the same length of time. God knows what it's doing, but it's certainly FAAH that has the problem here. I'm afraid I can't justify running FAAH on my main crunching computer when I knows it's going to be wasting CPU by a factor of 2x + so sorry but until it's fixed its got to be HCC/HFCC :( EDIT: My HFCC aren't valid yet so not in above list. Now I check them live they may be misbehaving as well?! Is there anything FAAH/HFCC have in common that other tasks don't? Compiler? Libraries? It's CPU cycle usage is like FAAH : trundling away exact same as a 'normal' task just slow. Most odd. [Edit 1 times, last edit by Highwire at Feb 6, 2010 3:23:54 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
comparing run times for FAAH/HFCC to HCMD/NRW is apples to oranges. HCMD and NRW have built in max run times and FAAH/HFCC (and HCC/HPF) run till done. My average points/task for FAAH/HFCC are 2x the average of HCMD/NRW and average run times also 2 to 1
|
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
comparing run times for FAAH/HFCC to HCMD/NRW is apples to oranges. HCMD and NRW have built in max run times and FAAH/HFCC (and HCC/HPF) run till done. My average points/task for FAAH/HFCC are 2x the average of HCMD/NRW and average run times also 2 to 1 Actually, if you looks on his granted credit, you'll see while for most of the non-FAAH-tasks, his claimed credit is within 10% of the granted credit. The single FAAH-task on the other hand is 86% off... Now, a single result is much too few to draw any conclusions, since granted credit can be "all over the place". Still, it is something that would be interesting to investigate more closely, to see if there is a problem somewhere or not... "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
There's not a single HFCC in that list [AutoDock based science as is FAAH without wingman]. Given both these sciences have a mini benchmark included that when too well executed causes credit undergrants i.e. the jobs execute less to what the mini benchmark infers, I'd not conclude anything until seen a sizeable series of both HFCC and FAAH. Given that actual FAAH project credit grant per hour is at the center of WCG averages together with HFCC, think the device is just not doing all-round, but as said could not possibly derive anything until a series is tested.
----------------------------------------btw, was that 1 FAAH job run without wingman? If it was, could be the claim differential was so great that the outlier rule kicked in... that credit granted to both per the device nearest it's recent historical average [for that science]. --//--
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
|