Welcome to MilkyWay@home

3rd.in - optimized apps

Message boards : Number crunching : 3rd.in - optimized apps
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 27 · 28 · 29 · 30 · 31 · 32 · 33 . . . 40 · Next

AuthorMessage
Ironworker16
Avatar

Send message
Joined: 31 Jan 09
Posts: 31
Credit: 69,908,565
RAC: 0
Message 30663 - Posted: 13 Sep 2009, 13:26:54 UTC - in response to Message 30640.  
Last modified: 13 Sep 2009, 13:28:21 UTC

Most important change from the scientific side is an improved accuracy (better than the stock app ;), which exactly matches the results of the new 0.20 ATI versions (have to run the last tests on that, will be posted tomorrow if nothing goes wrong).


Well this is different, I'll be looking forward to the start of a new work week a Monday.

Thanks CP
ID: 30663 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile GalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
Message 30693 - Posted: 13 Sep 2009, 20:06:45 UTC
Last modified: 13 Sep 2009, 20:07:36 UTC

My 0.20 versions for CPUs are ready and available for download here.


Thanks Cluster Physik / Gipsel

Downloads are also available here; http://www.brilliantsite.com/milkyway

ID: 30693 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 30706 - Posted: 13 Sep 2009, 21:58:56 UTC - in response to Message 30640.  

My 0.20 versions for CPUs are ready and available for download here.

Most important change from the scientific side is an improved accuracy (better than the stock app ;), which exactly matches the results of the new 0.20 ATI versions (have to run the last tests on that, will be posted tomorrow if nothing goes wrong).

But it also got slightly faster and there are now 64bit versions in the package which add some more percents to the performance (about 15% on a Phenom with a 2stream WU). And as always, read the included readme.txt!

It's almost tomorrow here (2 minutes of today are left). I guess it will be okay to post a link to the GPU version 0.20 then.

Changes are as already mentioned a higher accuracy with results exactly matching my latest CPU versions which have increased the accuracy above that of the stock app. These differences in the result are all well within the limits given by the project, so I think it is okay to implement it unilaterally so to say. But it should be possible to tighten these limits when the changes are ported to the stock app as well as the CUDA app (Travis and Anthony were given notice of those changes some days ago). This should enable less noisy results which may improve the convergence of the searches.

But I guess most are concerned by the speed of the new versions. I invite everyone to compare their credits/day with the 0.19 versions before the credit reduction and now with the 0.20 version after the reduction ;)

And really important not only this time: Read the readme.txt included in the package!
The included app_info.xml is already made for the new 6.10.x BOINC clients (hope there will be a good working version within in a few days). If you use another one, you may need to remove the <coproc></coproc> statements and adjust the <avg_ncpus> value (i.e. raise it to your needs).
ID: 30706 · Rating: 0 · rate: Rate + / Rate - Report as offensive
STE\/E

Send message
Joined: 29 Aug 07
Posts: 486
Credit: 576,516,164
RAC: 36,864
Message 30710 - Posted: 13 Sep 2009, 22:16:57 UTC

By reading in this Thread I take it the 0.20 Applications is for the CPU's only & won't work with the GPU's ???
ID: 30710 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Labbie
Avatar

Send message
Joined: 29 Aug 07
Posts: 327
Credit: 116,463,193
RAC: 0
Message 30712 - Posted: 13 Sep 2009, 22:26:16 UTC - in response to Message 30710.  

By reading in this Thread I take it the 0.20 Applications is for the CPU's only & won't work with the GPU's ???


The post just before yours has the link to the ATI apps.



Calm Chaos Forum...Join Calm Chaos Now
ID: 30712 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 30714 - Posted: 13 Sep 2009, 22:29:15 UTC - in response to Message 30710.  

By reading in this Thread I take it the 0.20 Applications is for the CPU's only & won't work with the GPU's ???

And what the hell is linked in the post directly before yours???

Is my English that bad? ;)
ID: 30714 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile [AF>Occitania] Meteore31

Send message
Joined: 11 Dec 07
Posts: 11
Credit: 95,037,645
RAC: 0
Message 30715 - Posted: 13 Sep 2009, 22:32:37 UTC

with boinc 6.10.4 and yours 0.20 gpu, it crunches only 1 wu at the same time, is it ok ? before it was 3 wus at the same time...
ID: 30715 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Raistmer*

Send message
Joined: 27 Jun 09
Posts: 85
Credit: 39,805,338
RAC: 0
Message 30716 - Posted: 13 Sep 2009, 22:34:02 UTC - in response to Message 30714.  


x accumulates more in graphics RAM, fewer reductions but more RAM needed

Could you specify how much is needed now, please?


- uses boinc_critical_sections so the app isn't interrupted while it owns a GPU

Does it mean that global mutex doesn't used now or it still implemented?
ID: 30716 · Rating: 0 · rate: Rate + / Rate - Report as offensive
JAMC

Send message
Joined: 9 Sep 08
Posts: 96
Credit: 336,443,946
RAC: 0
Message 30717 - Posted: 13 Sep 2009, 22:38:01 UTC - in response to Message 30715.  

with boinc 6.10.4 and yours 0.20 gpu, it crunches only 1 wu at the same time, is it ok ? before it was 3 wus at the same time...


Me too with 6.10.3- 1 WU at a time on single 4870 and 2 WU's at a time on double 4870 rigs... tried n2 but still 1 WU at a time... XP Home
ID: 30717 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile GalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
Message 30719 - Posted: 13 Sep 2009, 22:40:19 UTC


Ummmmmm. The 0.20ati is...... fast! :)



ID: 30719 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Raistmer*

Send message
Joined: 27 Jun 09
Posts: 85
Credit: 39,805,338
RAC: 0
Message 30720 - Posted: 13 Sep 2009, 22:40:21 UTC - in response to Message 30717.  

with boinc 6.10.4 and yours 0.20 gpu, it crunches only 1 wu at the same time, is it ok ? before it was 3 wus at the same time...


Me too with 6.10.3- 1 WU at a time on single 4870 and 2 WU's at a time on double 4870 rigs... tried n2 but still 1 WU at a time... XP Home


Try to set this one to 0.5

<coproc>
<type>ATI</type>
<count>1</count>
</coproc>
ID: 30720 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Crunch3r
Volunteer developer
Avatar

Send message
Joined: 17 Feb 08
Posts: 363
Credit: 258,227,990
RAC: 0
Message 30721 - Posted: 13 Sep 2009, 22:43:02 UTC - in response to Message 30715.  
Last modified: 13 Sep 2009, 22:44:14 UTC

with boinc 6.10.4 and yours 0.20 gpu, it crunches only 1 wu at the same time, is it ok ? before it was 3 wus at the same time...


OMG ! Are you people even reading what he posted ?

And really important not only this time: Read the readme.txt included in the package!
The included app_info.xml is already made for the new 6.10.x BOINC clients (hope there will be a good working version within in a few days). If you use another one, you may need to remove the <coproc></coproc> statements and adjust the <avg_ncpus> value (i.e. raise it to your needs).


Join Support science! Joinc Team BOINC United now!
ID: 30721 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 30722 - Posted: 13 Sep 2009, 22:48:05 UTC - in response to Message 30716.  


x accumulates more in graphics RAM, fewer reductions but more RAM needed

Could you specify how much is needed now, please?

The RAM usage depends on the WU size, but for the currently largest WUs it is about 70MB.


- uses boinc_critical_sections so the app isn't interrupted while it owns a GPU

Does it mean that global mutex doesn't used now or it still implemented?

It is still using mutexes (otherwise more than one WU per GPU wouldn't work). The boinc_critical_sections are only used to disallow BOINC to suspend a task while a kernel runs. In the worst case this would block all other instances to use the GPU. Some people have reported this exact problem, therefore I put it in and hope those problems are gone now.
ID: 30722 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 30723 - Posted: 13 Sep 2009, 22:51:50 UTC - in response to Message 30721.  

with boinc 6.10.4 and yours 0.20 gpu, it crunches only 1 wu at the same time, is it ok ? before it was 3 wus at the same time...

OMG ! Are you people even reading what he posted ?

And really important not only this time: Read the readme.txt included in the package!
The included app_info.xml is already made for the new 6.10.x BOINC clients (hope there will be a good working version within in a few days). If you use another one, you may need to remove the <coproc></coproc> statements and adjust the <avg_ncpus> value (i.e. raise it to your needs).

I guess it's not that important that I forgot to mention how to raise the number of running WUs per GPU with that <coproc> section. It wouldn't have been read either way! *lol*
ID: 30723 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Raistmer*

Send message
Joined: 27 Jun 09
Posts: 85
Credit: 39,805,338
RAC: 0
Message 30724 - Posted: 13 Sep 2009, 22:51:51 UTC - in response to Message 30722.  


x accumulates more in graphics RAM, fewer reductions but more RAM needed

Could you specify how much is needed now, please?

The RAM usage depends on the WU size, but for the currently largest WUs it is about 70MB.


- uses boinc_critical_sections so the app isn't interrupted while it owns a GPU

Does it mean that global mutex doesn't used now or it still implemented?

It is still using mutexes (otherwise more than one WU per GPU wouldn't work). The boinc_critical_sections are only used to disallow BOINC to suspend a task while a kernel runs. In the worst case this would block all other instances to use the GPU. Some people have reported this exact problem, therefore I put it in and hope those problems are gone now.


Thanks!
And yes, I've seen that problem on my host. MW processing was blocked until tasks begin to fail due "too long processing time" limit.

ID: 30724 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 30725 - Posted: 13 Sep 2009, 22:55:12 UTC - in response to Message 30719.  
Last modified: 13 Sep 2009, 22:55:32 UTC

Ummmmmm. The 0.20ati is...... fast! :)

Yep :D
ID: 30725 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Crunch3r
Volunteer developer
Avatar

Send message
Joined: 17 Feb 08
Posts: 363
Credit: 258,227,990
RAC: 0
Message 30726 - Posted: 13 Sep 2009, 22:59:16 UTC - in response to Message 30723.  


I guess it's not that important that I forgot to mention how to raise the number of running WUs per GPU with that <coproc> section. It wouldn't have been read either way! *lol*


Naja, dann gibt es wenigstens was zu lachen :D
Der erste der das dann raus findet,bekommt einen blechpokal als belohnung :p




Join Support science! Joinc Team BOINC United now!
ID: 30726 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 30727 - Posted: 13 Sep 2009, 23:01:51 UTC - in response to Message 30726.  

I guess it's not that important that I forgot to mention how to raise the number of running WUs per GPU with that <coproc> section. It wouldn't have been read either way! *lol*

Naja, dann gibt es wenigstens was zu lachen :D
Der erste der das dann raus findet,bekommt einen blechpokal als belohnung :p

Raistmer hat die Lösung doch schon gepostet ;)
ID: 30727 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Dan T. Morris
Avatar

Send message
Joined: 17 Mar 08
Posts: 165
Credit: 410,228,216
RAC: 0
Message 30728 - Posted: 13 Sep 2009, 23:03:05 UTC - in response to Message 30725.  

Getting a lot of bad wu. I am using plain 6.10.4 and the new cp ati ap.

Anyone else having this error problem?

I have 2 hd 4870 in a q6600 intel. with 4 geg ram.


ID: 30728 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Labbie
Avatar

Send message
Joined: 29 Aug 07
Posts: 327
Credit: 116,463,193
RAC: 0
Message 30729 - Posted: 13 Sep 2009, 23:05:14 UTC - in response to Message 30728.  

Getting a lot of bad wu. I am using plain 6.10.4 and the new cp ati ap.

Anyone else having this error problem?

I have 2 hd 4870 in a q6600 intel. with 4 geg ram.



did you run out your cache before changing the app? I had that problem yesterday with the CPU app on one box. You may need to reset the project too.

Calm Chaos Forum...Join Calm Chaos Now
ID: 30729 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Previous · 1 . . . 27 · 28 · 29 · 30 · 31 · 32 · 33 . . . 40 · Next

Message boards : Number crunching : 3rd.in - optimized apps

©2024 Astroinformatics Group