News:

Members can see all forum boards and posts. Non members can only see a few boards.  If you have forgotten your password use this link to change it.
https://forum.boinc-australia.net/index.php?action=reminder

Main Menu

Optimised apps

Started by veebee, February 12, 2009, 07:12:52 AM

Previous topic - Next topic

veebee

Optimised apps for MW are killing !

I have never had a problem with any other project.... run down the cache, stop Boinc client, copy app info xml and the app file to the project folder, then restart client and allow the project t get new work..the new app runs and the world is mellow.

But I have done the above so many times now, and keep ending up with the "intel x86 " 0.17  file downloading with the fresh batch of WU's, and then the regular crunching of 50 minute 0.17 WU's begins all over again...and they are "high priority..which doesn't allow me to finally use CUDA for some of these other faster cuda apps.

Can anyone pick what I may be doing wrong..?

thanksamillyun

Rocoll


veebee

I found despite the app-info XML file MW still uses the standard app.
I had to stop BOINC and delete the standard app.... only then did it see the optimised app as directed by the XML.

To get cuda working go to the lunatics web site and look in the forums for version 7 of the cuda optimised app.

Cruncher Pete

From Milkyway News Archive:

I've upated the alpha release of the v0.18 application. It should now compile correctly. It has some changes for a small checkpointing bug and more performance improvements. I'm going to let other people try it out for a bit before I update the stock application. So if all goes well stock should upgrade to 0.18 in a couple days. After we've updated the stock application to v0.18 and we see that it's working correctly and we have optimized versions of it available, I'm going to stop awarding credit to pre 0.18 compiled applications, because we don't have any other way to get people to upgrade their optimized applications. I'll post a news item as a warning and provide links to appropriate v0.18 applications so that everyone has time to upgrade their applications before I make the switch.

Wang Solutions

veebee are you having problems in Windows or Linux, and which BOINC version?

veebee

Windows, 32 bit  I have gone back to 6.4.5 as I had nothing but trouble with 6.6.x.

As it is now, 5 MW WU's run at one time which is "ok", but if I run SETI, only one SETI WU will run and the climate model.

Furlozza

In the config file, have a feeling that you should explain to your boy that there is just one GPU and four CPUs.... and I think the file that does that is mentioned somewhere in this board..... brb

"<avg_ncpus>-1</avg_ncpus>
<max_ncpus>-1</max_ncpus>
<coproc>
<type>CUDA</type>
<count>1</count>" as per Rocoll, but I'd change the avg and max numbers for the cpus to 4 and THEN see how it runs.

Tixx

#6
QuoteI've updated the stock application for linux and os x to v0.18. The code is in the code release directory as mw_v0.18d. This has a couple new performance improvements (thanks to our users) which has it running 30% faster, or more. This should bring our stock application (and some of the other optimized versions of it) in line with the fastest optimized apps we're seeing out there. Because of this, in the next day or so we'll have another (much loved) credit modification. I'll most likely be dropping the credit rate by about 30-50% but at the same time removing the credit limit. Due to the nature of the performance improvements, older hosts should see better performance improvements than newer ones, so for most people the credit/hr should be the same, while faster hosts might lose some credit or gain some credit depending on where they are compared to the speed limit.

any idea on what their "speed limit" gauge is set to? :)

yoda

Quote from: Tixx on February 16, 2009, 04:47:40 PM
any idea on what their "speed limit" gauge is set to? :)

108 credits per core per hour

Tixx

yeah but they are removing the cap, so they must benchmark it to a ghz amount instead..

for example they would benchmark 108 credits done by a single core in 1hr at 2.1ghz speed
then anyone at higher clock speeds should be able to clock more credits per hour?

the whole milky credit juggling thing i still havent figured out :)


kashi

#9
I understand your confusion.  Unless I have misunderstood then I think Travis has made a large miscalculation there or he is hoping that nobody says anything. If he drops the speed limit and reduces the credit by 50% to 15 credits per task, then a fast Windows host that takes for example 5 minutes will score, cough, cough, very highly.  :wink

Still that's nothing because the GPU client on a HD4870 takes only 9 seconds, so if WUs are worth 15 credits then the cr/hr will be 6,000 until the daily WU limit is reached after 2.5 hours.   :coolgleam:

I have a HD3850 GPU ready to take advantage of any temporary bonus/error such as this but my computer refuses to operate. They renewed the electricity cable to my house on Saturday and I powered down from constant PrimeGrid crunching. When I tried to restart an hour later fans repeatedly spin for 5 seconds, stop for 5 seconds, but no banana. Computer is kaput.  :cry2:

I have done some troubleshooting and it is either the motherboard or the CPU gone to electrical heaven. Most disheartening. When I feel better about it tomorrow I will try my old Pentium D 935 and see what happens.

Currently I am soldiering on with my old A64 3200 doing some MilkyWay, but the days for easy MilkyWay credits for these old boat anchors are just about to finish I think. I'm more disappointed that my raid on PrimeGrid has been so horribly terminated.

yoda

#10
Quote from: kashi on February 16, 2009, 09:56:54 PM
When I tried to restart an hour later fans repeatedly spin for 5 seconds, stop for 5 seconds, but no banana. Computer is kaput.  :cry2:

You tried disconnecting the power cable, leaving it out for 5 seconds or so, then plug in and power up again?  Sometimes that sorts out the problem.

But yes, the Windows apps are faster than even 64bit Linux, so I might have to see how I can get hold of some Windows licences if that's the case!  The X3350 at 3.4 GHz could probably do them in 3 minutes too iun Windows, so that's 80 WU an hour or (at 15 credits per WU) 1200 an hour / 28,800 a day on the one quad.  Hmmm.

kashi

Thanks for the suggestion, I'll give it a try tomorrow.

I really hope it's not the CPU, the place I bought it from said they had stock, took my money and then didn't supply for a month. I particularly don't want to have anything to do with them ever again.

Back on topic a bit, yes if Travis wants to restrict fast Windows machines to 108 cr/hr or less he will have to reduce the credit to 5.4, not 15. Owners of slower machines and possibly Linux users will not be too happy about that. He'll possibly choose to do it in stages to cushion the shock for many MilkyWay contributors.

Ah well I can't try the GPU app yet anyway without 64-bit Windows, so I won't really be missing anything there. It's just the thought of doing a WU in 25 seconds or so excited my imagination. Hmm, that's 15K credit (1,000 WU daily limit) in 7 hours at 15 per WU.

kashi

New optimised Linux v18d apps from speedimic. :)

http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=566&nowrap=true#11020

Excuse double post, but I thought this deserved it.

Tixx

Just ran a couple stock 0.19 milky wus..

on a duocore 2.7ghz they ran just under 22min..
so its almost double improvement on the old stock app.

however, hopefully we'll get the new optimized app out to bring them down to 4min again :)

Diablo

Windows 0.19 optimised apps are up (http://zslip.com/)