Welcome to MilkyWay@home

3rd.in - optimized apps


Advanced search

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

To post messages, you must log in.

Previous · 1 . . . 31 · 32 · 33 · 34 · 35 · 36 · 37 . . . 40 · Next

AuthorMessage
Chris S
Avatar

Send message
Joined: 20 Sep 08
Posts: 1387
Credit: 186,726,858
RAC: 0
100 million credit badge10 year member badge
Message 30852 - Posted: 14 Sep 2009, 18:30:09 UTC

I think it might be worth while pointing out that each and every PC is an individual one off. There are various combinations and versions of motherboards, processors, Bios's, graphics cards, memory, etc etc. All these factors are different in each PC, and when running something like Boinc, make a significant difference.

Each cruncher needs to work out for themselves what particular combination of software and hardware works for them. Boinc Developers can only try to deal with what they think is a common denominator, when they issue a new Boinc client.

I have to say that I think there have been far too many versions of the Boinc client issued over the last couple of years, and sometimes it appears for no good reason. I'm still using 6.4.7 and it works for me. And I will continue to do so until I am convinced there is a good reason to change.
ID: 30852 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
50 million credit badge10 year member badgeextraordinary contributions badge
Message 30859 - Posted: 14 Sep 2009, 19:19:00 UTC - in response to Message 30832.  

0.20 errors after 1 sec (((

Look to this thread. You appear to use the same OS/driver combination which may be to blame. Changing the driver version should help.
ID: 30859 · Rating: 0 · rate: Rate + / Rate - Report as offensive
STE\/E

Send message
Joined: 29 Aug 07
Posts: 486
Credit: 573,620,762
RAC: 0
500 million credit badge10 year member badge
Message 30861 - Posted: 14 Sep 2009, 19:25:33 UTC - in response to Message 30852.  

I think it might be worth while pointing out that each and every PC is an individual one off. There are various combinations and versions of motherboards, processors, Bios's, graphics cards, memory, etc etc. All these factors are different in each PC, and when running something like Boinc, make a significant difference.

Each cruncher needs to work out for themselves what particular combination of software and hardware works for them. Boinc Developers can only try to deal with what they think is a common denominator, when they issue a new Boinc client.

I have to say that I think there have been far too many versions of the Boinc client issued over the last couple of years, and sometimes it appears for no good reason. I'm still using 6.4.7 and it works for me. And I will continue to do so until I am convinced there is a good reason to change.


I've said the same thing a few times myself Chris, but some people think that what works for them should work for you. I take everything people say into consideration and then throw out 99.9% of it out because it doesn't work for me most of the time.

I'm having trouble myself with the 0.20 Application and don't even want to touch the 6.10.xx series of Clients for awhile after this mornings FUBAR with them. I have the 0.20 Application running but with the older 6.6.20 Client which I've been using for awhile with success on all my Box's ATI & NVIDIA.
ID: 30861 · Rating: 0 · rate: Rate + / Rate - Report as offensive
ProfileKWSN imcrazynow
Avatar

Send message
Joined: 22 Nov 08
Posts: 136
Credit: 319,414,799
RAC: 0
300 million credit badge10 year member badge
Message 30870 - Posted: 14 Sep 2009, 19:59:01 UTC - in response to Message 30792.  

Paul,
Look at your wire size. If it is a 12Ga. wire it is fine to put a 20Amp breaker on. If it is 14Ga. wire you should stay with the 15Amp breaker.
You should be able to find somewhere on the wire jacket what size it is.
By the wire jacket I mean it will have 3 wires in the jacket. 1 bare copper(ground) 1 Black(hot) 1 White (Neutral). If your home is newer construction (last 5 years or so) the jacket will be color coded. Orange = 10Ga. Yellow = 12Ga. White = 14Ga.
If your house is old construction it may have all 12 ga. wire to your 120 volt circuits.
This applies to the USA. I don't know if simmilar anywhere else.
Hope this helps.

4870 GPU
4870 GPU
ID: 30870 · Rating: 0 · rate: Rate + / Rate - Report as offensive
ProfileBeyond

Send message
Joined: 15 Jul 08
Posts: 383
Credit: 501,817,790
RAC: 0
500 million credit badge10 year member badge
Message 30881 - Posted: 14 Sep 2009, 20:33:49 UTC

Cluster Physik, thanks much for the hard work on the v.20 app. It solved the video crashes that I was having regularly with v.19f.
Using an HD 4770 card, WinXP32 & v9.8 drivers.
ID: 30881 · Rating: 0 · rate: Rate + / Rate - Report as offensive
ProfileKWSN imcrazynow
Avatar

Send message
Joined: 22 Nov 08
Posts: 136
Credit: 319,414,799
RAC: 0
300 million credit badge10 year member badge
Message 30897 - Posted: 14 Sep 2009, 23:11:59 UTC

Maybe i'm missing something or just a little dense but, I want to install the new app on BOINC 6.4.7. Below is the entire contents of the readme file that was part of the download...


Known issue:

Requires MSVC++ 2005 SP1 runtime package, which may not be installed. You should download it directly from Microsoft.

32Bit:
http://www.microsoft.com/downloads/details.aspx?FamilyID=200b2fd9-ae1a-4a14-984d-389c36f85647&displaylang=en

64Bit:
http://www.microsoft.com/downloads/details.aspx?familyid=EB4EBE2D-33C0-4A47-9DD4-B9A6D7BD44DA&displaylang=en

=========================================================================================

Changelog:

=========================================================================================


0.20: 13th Sep 2009
- improves the accuracy of the integral and likelihood calculation to exactly match the
results of my 0.20 GPU variant
- some speed tuning (should gain about 10% compared to 0.19)
- introduces a 64bit variant which is about 15% faster (tested on an AMD Phenom)
- removes the compilations for all different SSE#.# sublevels as there was no consistent gain
................................................................................
There was no Brook.dll and the app info file has no line to change the number of running units. I want 3 or 4 running at the same time.
I really hate to seem dense but if anybody can tell me what I need to do I would appreciate it. I've already downloaded and installed MSVC++ .

4870 GPU
4870 GPU
ID: 30897 · Rating: 0 · rate: Rate + / Rate - Report as offensive
STE\/E

Send message
Joined: 29 Aug 07
Posts: 486
Credit: 573,620,762
RAC: 0
500 million credit badge10 year member badge
Message 30899 - Posted: 14 Sep 2009, 23:30:31 UTC
Last modified: 14 Sep 2009, 23:38:24 UTC

I didn't have to install anything extra on my 64-Bit Windows Box's. I just put the Proper 0.20 App for a Windows 64-Bit OS & then either leave it the way it is for v6.10.xx Clients or take out the lines:

<coproc>
<type>ATI</type>
<count>1</count>
</coproc>

And put your old command lines from your 0.19f or whatever Previous Application you were using in place of the <coproc> lines for the Pre v6.10.xx Clients:

<avg_ncpus>0.X</avg_ncpus>
<max_ncpus>X</max_ncpus>
<cmdline>nX</cmdline>
ID: 30899 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Japi

Send message
Joined: 22 Feb 09
Posts: 445
Credit: 31,866,641
RAC: 0
30 million credit badge10 year member badge
Message 30901 - Posted: 14 Sep 2009, 23:46:57 UTC - in response to Message 30813.  

Decided to give 0.20 another try after seeing Verstapps post above. So I d/l vcredist, install, reboot and try again. Same result. So then I installed BOINC 6.10.3. Still getting the same results, all WUs error out with the 'output file absent'. I guess it's back to 0.19g for me.

Once more into the breach... This time upgraded Catalyst from 9.5 to 9.9. Now it works like a charm, seems that the culprit was Cat9.5. So now the system is Vista64, Cat9.9, BOINC 6.10.3 and 0.20 ATI app. Might even try to use 6.10.4 again.

Thank you Cluster Physik! :)
ID: 30901 · Rating: 0 · rate: Rate + / Rate - Report as offensive
ProfileDan T. Morris
Avatar

Send message
Joined: 17 Mar 08
Posts: 165
Credit: 410,228,216
RAC: 0
300 million credit badge10 year member badge
Message 30904 - Posted: 15 Sep 2009, 0:01:48 UTC - in response to Message 30901.  

warning 6.10.4 bad. I promise you won't like what is does. Just my opinion..


ID: 30904 · Rating: 0 · rate: Rate + / Rate - Report as offensive
ProfileKWSN imcrazynow
Avatar

Send message
Joined: 22 Nov 08
Posts: 136
Credit: 319,414,799
RAC: 0
300 million credit badge10 year member badge
Message 30906 - Posted: 15 Sep 2009, 0:19:35 UTC - in response to Message 30899.  

Thanks PoorBoy.! I'll give that a whirl. However.... bet you knew that was coming. I was using 19f,with that was Brook.dll. I noticed with 19g the Brook file was Brook64.dll. Will that make any difference as far as performance or do you know? Maybe CP can say.
PS.. Hows your grandson doing? I lost track of the thread at boinc stats. Our thoughts and prayers are still with him.

4870 GPU
4870 GPU
ID: 30906 · Rating: 0 · rate: Rate + / Rate - Report as offensive
STE\/E

Send message
Joined: 29 Aug 07
Posts: 486
Credit: 573,620,762
RAC: 0
500 million credit badge10 year member badge
Message 30907 - Posted: 15 Sep 2009, 0:28:53 UTC

The 0.20 Application comes together with a 32 & 64 Bit version, so just use the appropriate one for your OS. As far as I know the 64-Bit Versions are faster. My Grandson is in good Spirits, not much more to report on him at him at the moment though.
ID: 30907 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
50 million credit badge10 year member badgeextraordinary contributions badge
Message 30909 - Posted: 15 Sep 2009, 0:36:34 UTC - in response to Message 30906.  
Last modified: 15 Sep 2009, 0:53:45 UTC

Maybe i'm missing something or just a little dense but, I want to install the new app on BOINC 6.4.7. Below is the entire contents of the readme file that was part of the download...

Looks like you downloaded the CPU version.

I noticed with 19g the Brook file was Brook64.dll. Will that make any difference as far as performance or do you know? Maybe CP can say.

It will make the difference if it runs or not ;)
You can't use the older brook.dll and rename it and use it with the new app. I've done some modifications to it, which are required for the ATI app to run correctly. One should use the brook file which is supplied with the specific version one wants to run (most likely 0.20).
ID: 30909 · Rating: 0 · rate: Rate + / Rate - Report as offensive
ProfileKWSN imcrazynow
Avatar

Send message
Joined: 22 Nov 08
Posts: 136
Credit: 319,414,799
RAC: 0
300 million credit badge10 year member badge
Message 30913 - Posted: 15 Sep 2009, 1:28:46 UTC

You were exactly right CP. I made a boneheaded mistake by not paying attention. I've downloaded the GPU version this time. Thank You!
PoorBoy,
Glad to hear he's in good spirits. A good attitude can go a very long way to helping one heal. We'll keep him in our prayers.
Randy

4870 GPU
4870 GPU
ID: 30913 · Rating: 0 · rate: Rate + / Rate - Report as offensive
John Clark

Send message
Joined: 4 Oct 08
Posts: 1734
Credit: 64,228,409
RAC: 0
50 million credit badge10 year member badge
Message 30953 - Posted: 15 Sep 2009, 13:57:01 UTC
Last modified: 15 Sep 2009, 14:07:56 UTC

Now that my AG ATI HD3850 has crunched a few caches using the Gispel 0.19 Win32-SSE2 version, I decided to update to the new 0.20 32 bit version.

I run Win XP Pro, BOINC 6.4.7 and the ATI 8.12 Catalyst.

As soon as I restarted BOINC Manager I got the following in the /messages/ tab -

15/09/2009 14:52:00||[error] Missing a ATI coprocessor
15/09/2009 14:52:00||[error] Missing a ATI coprocessor
15/09/2009 14:52:00||[error] Missing a ATI coprocessor
15/09/2009 14:52:00||[error] Missing a ATI coprocessor
15/09/2009 14:52:00||[error] Missing a ATI coprocessor
15/09/2009 14:52:02|Milkyway@home|Sending scheduler request: Requested by user. Requesting 0 seconds of work, reporting 0 completed tasks
15/09/2009 14:52:24||Project communication failed: attempting access to reference site
15/09/2009 14:52:25||Internet access OK - project servers may be temporarily down.
15/09/2009 14:52:27|Milkyway@home|Scheduler request failed: Failure when receiving data from the peer
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor
15/09/2009 14:52:27||[error] Missing a ATI coprocessor


The cache of MW work is there and there is no calculations.

I am reverting to 0.19, which was working a minute ago.

Less than 30 seconds ago I reinstalled 0.19 and MW is GPU crunching 3 x 0.19 WUs fine.

That raises the question as to what might be going wrong to prevent the Gipsel 0.20 client working?

I have made no alterations or additions to the app_info file, just left it as downloaded and unzipped.
Go away, I was asleep


ID: 30953 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
50 million credit badge10 year member badgeextraordinary contributions badge
Message 30954 - Posted: 15 Sep 2009, 14:18:15 UTC - in response to Message 30953.  

I run Win XP Pro, BOINC 6.4.7 and the ATI 8.12 Catalyst.

As soon as I restarted BOINC Manager I got the following in the /messages/ tab -

15/09/2009 14:52:00||[error] Missing a ATI coprocessor
[..]
That raises the question as to what might be going wrong to prevent the Gipsel 0.20 client working?

I have made no alterations or additions to the app_info file, just left it as downloaded and unzipped.

The red part ist what you have done wrong. When you guys will finally start to read the damn readme included in the package? Just to cite it:

System requirements:
[..]
- BOINC client 6.10.x or higher for ATI support, but works also with older versions,
see below the changelog how to change the supplied app_info.xml for older clients

[..]
Configuration for older BOINC clients (before 6.10.x):

The included app_info.xml declares the app as GPU app for the new 6.10.x clients. This is done with a <coproc></coproc> statement, which may confuse older versions as they don't recognize the type of required GPU. Theses clients may refuse to run the app in such a case. To get it running, just remove the lines
<coproc>
...
</coproc>
and everything between those tags. It is twice in the .xml file, you have to remove both. After that you should adjust the value <avg_ncpus>0.05</avg_ncpus> (just above the coproc statement) to something higher (depends on your needs, but I would recommend 0.5 or even 1).


Setting the <count> value within those <coproc> statement to something below 1 enables to run more than 1 WU per GPU with 6.10.3 and up (it is not working with 6.10.0).
ID: 30954 · Rating: 0 · rate: Rate + / Rate - Report as offensive
John Clark

Send message
Joined: 4 Oct 08
Posts: 1734
Credit: 64,228,409
RAC: 0
50 million credit badge10 year member badge
Message 30958 - Posted: 15 Sep 2009, 15:15:50 UTC
Last modified: 15 Sep 2009, 15:16:52 UTC

Oh, dear! How very very embarassing!

Stands at the naughty desk and rreads the new text files bundled with Gipsel's 0.20 client.

Now come careful sissor work (not allowed a knife as not old enough) and removal of 2 coproc statements, then modifying theav_ncpu statements and adding to the cmdline statements

Restarting BOINC, and all is sweatness and light.

Many thanks Clusterr Physik

Now standing, facing the wall, in the naughty corner until the teacher calls me for tea!
;))
Go away, I was asleep


ID: 30958 · Rating: 0 · rate: Rate + / Rate - Report as offensive
ProfilePaul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
100 million credit badge10 year member badge
Message 30959 - Posted: 15 Sep 2009, 15:22:49 UTC - in response to Message 30954.  

Setting the <count> value within those <coproc> statement to something below 1 enables to run more than 1 WU per GPU with 6.10.3 and up (it is not working with 6.10.0).

Does it actually run more than one at the same time? Or just pretend to?
ID: 30959 · Rating: 0 · rate: Rate + / Rate - Report as offensive
STE\/E

Send message
Joined: 29 Aug 07
Posts: 486
Credit: 573,620,762
RAC: 0
500 million credit badge10 year member badge
Message 30960 - Posted: 15 Sep 2009, 15:25:57 UTC

It's so easy a Caveman like me can do it ... :P
ID: 30960 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
50 million credit badge10 year member badgeextraordinary contributions badge
Message 30964 - Posted: 15 Sep 2009, 15:55:00 UTC - in response to Message 30959.  

Setting the <count> value within those <coproc> statement to something below 1 enables to run more than 1 WU per GPU with 6.10.3 and up (it is not working with 6.10.0).

Does it actually run more than one at the same time? Or just pretend to?

You are right, the graphics driver (and the CAL interface) can't handle to really run kernels from different contexts (different WUs) at the same time on one GPU. But it is possible to hide the (small) CPU part of one WU during the execution of the GPU part of a second one by switching between the WUs on a GPU in a round robin fashion. This may shave off a second or so per WU (of course not off the GPU time, but the wallclock time may be quite close to 2*GPU time [for equal WUs], meaning it is an effective gain), if you run two at once.
ID: 30964 · Rating: 0 · rate: Rate + / Rate - Report as offensive
ProfilePaul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
100 million credit badge10 year member badge
Message 30972 - Posted: 15 Sep 2009, 17:14:07 UTC - in response to Message 30964.  

Setting the <count> value within those <coproc> statement to something below 1 enables to run more than 1 WU per GPU with 6.10.3 and up (it is not working with 6.10.0).

Does it actually run more than one at the same time? Or just pretend to?

You are right, the graphics driver (and the CAL interface) can't handle to really run kernels from different contexts (different WUs) at the same time on one GPU. But it is possible to hide the (small) CPU part of one WU during the execution of the GPU part of a second one by switching between the WUs on a GPU in a round robin fashion. This may shave off a second or so per WU (of course not off the GPU time, but the wallclock time may be quite close to 2*GPU time [for equal WUs], meaning it is an effective gain), if you run two at once.

Not broke, not going to fix it ...

Seems the gain is not worth the potential pain or problems ...

Thanks for the explanation...

Detour:

Took a look at my cost to re-engine the Q9300 with an i7 and a pair of HD4870 cards ... ah well, a guy can dream ...
ID: 30972 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Previous · 1 . . . 31 · 32 · 33 · 34 · 35 · 36 · 37 . . . 40 · Next

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

©2019 Astroinformatics Group