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 . . . 34 · 35 · 36 · 37 · 38 · 39 · 40 · Next

AuthorMessage
steveplanetary
Avatar

Send message
Joined: 19 May 09
Posts: 16
Credit: 20,273
RAC: 0
Message 31909 - Posted: 3 Oct 2009, 5:18:14 UTC - in response to Message 31837.  

Nuadormrac, before I replaced the 0.19 stock version with the 0.20 version downloaded from zslip.com I looked for an app_info.xml file, but couldn't find one. So I put the 0.20 app_info file in the ...Boinc/projects/milkway.cs.rpi.edu/milkyway folder. When the first MW WU downloaded after I upgraded, BOINC Manager listed the application as milkyway 0.20. I doen't think the server has strictly 0.19 WUs that require the 0.19 app. I think the server downloads the same WUs regardless of which app you're using. It sounds like you have both apps in the ...Boinc/projects/milkway.cs.rpi.edu/milkyway folder. If so, I would at least move the 0.19 app to another folder. The server (or the Boinc client) may see it as the default.

Like you, I used CPU-Z to learn the most advanced instruction set my CPU is capable of.

when I went to parse together the 2 app_info.xml files to include entries for both one listing followed by the other; things seemed to not exactly go well together (the run times were suggestive of it going back to a stock client).


Maybe that wasn't such a good idea. :)

Cheers,

Steve
ID: 31909 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Misfit
Avatar

Send message
Joined: 27 Aug 07
Posts: 915
Credit: 1,503,319
RAC: 0
Message 31964 - Posted: 4 Oct 2009, 19:13:49 UTC

Looking forward to v0.20 SSSE3. Any ETA?
me@rescam.org
ID: 31964 · 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 31969 - Posted: 4 Oct 2009, 21:34:28 UTC - in response to Message 31964.  

Looking forward to v0.20 SSSE3. Any ETA?

Never ;)

All the different SSE levels won't change the performance as for the vectorization SSE2 is basically enough. The 64bit variants are extracting more performance than any additional (S)SSEy.x version could do.
ID: 31969 · Rating: 0 · rate: Rate + / Rate - Report as offensive
steveplanetary
Avatar

Send message
Joined: 19 May 09
Posts: 16
Credit: 20,273
RAC: 0
Message 32023 - Posted: 6 Oct 2009, 11:34:29 UTC - in response to Message 31964.  

Looking forward to v0.20 SSSE3. Any ETA?


But if you really want SSE3 for Win32 it's available at www.zslip.com.
ID: 32023 · Rating: 0 · rate: Rate + / Rate - Report as offensive
John Clark

Send message
Joined: 4 Oct 08
Posts: 1734
Credit: 64,228,409
RAC: 0
Message 32026 - Posted: 6 Oct 2009, 13:45:41 UTC - in response to Message 32023.  

That's for CPUs only, not ATI GPUs.
Go away, I was asleep


ID: 32026 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Paul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
Message 32046 - Posted: 6 Oct 2009, 20:14:35 UTC - in response to Message 32026.  

That's for CPUs only, not ATI GPUs.

If you are running on a GPU, SSE of any flavor is not going to do much of anything to improve run times. In essence the GPU runs an instruction set that is similar in effect as to the SSE instructions. So, all the calculations that could take advantage of this instruction set are run on the GPU ... the remaining code on the CPU will get no benefit as this code is mostly moving data on and off the GPU itself.
ID: 32046 · Rating: 0 · rate: Rate + / Rate - Report as offensive
NX-01

Send message
Joined: 26 Jul 09
Posts: 6
Credit: 48,518
RAC: 0
Message 32055 - Posted: 6 Oct 2009, 21:55:55 UTC - in response to Message 31969.  

Looking forward to v0.20 SSSE3. Any ETA?

Never ;)

All the different SSE levels won't change the performance as for the vectorization SSE2 is basically enough. The 64bit variants are extracting more performance than any additional (S)SSEy.x version could do.

So 0.20 SSE3 should be faster than 0.19 SSSE3, right?

ID: 32055 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Nuadormrac

Send message
Joined: 11 Sep 08
Posts: 22
Credit: 9,081,761
RAC: 0
Message 32060 - Posted: 7 Oct 2009, 0:54:29 UTC - in response to Message 31909.  

Actually, hitherto it's been using the 0.19 app; things got a little whonky, so did a reset project, then just put the 0.19 in as that's all that was there, after cleaning it out (it wasn't even processing WUs). Now that's a hmm, because with the seti optimized apps, when they transitioned from 1 science app version to another, having entries for both in the app_info file worked perfectly well, and on download it just told it which to use.

Either way, as it gave 0.19 work from all indications that's what I put in. However, the project is out of results ready to send, so once it comes back up I'll see. Some indication should definitely come when they want us to switch, but I'm guessing there was a backlog of 0.19 work present or something for which it continued sending results.

Perhaps this no work available now, could be replaced with only 0.20 coming up after? Or maybe it's just another in a long line of server problems (the website access timed out once or twice, and one access to the project gave

10/6/2009 8:41:16 PM Milkyway@home Sending scheduler request: To fetch work.
10/6/2009 8:41:16 PM Milkyway@home Requesting new tasks
10/6/2009 8:41:38 PM Project communication failed: attempting access to reference site
10/6/2009 8:41:39 PM Internet access OK - project servers may be temporarily down.
10/6/2009 8:41:41 PM Milkyway@home Scheduler request failed: Couldn't connect to server


I'll see after it shows more then 0 results ready to send on server status page.
ID: 32060 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile arkayn
Avatar

Send message
Joined: 14 Feb 09
Posts: 999
Credit: 74,932,619
RAC: 0
Message 32213 - Posted: 10 Oct 2009, 1:18:57 UTC

I have my site updated with the newer 0.20b ATI apps now.

http://www.arkayn.us/milkyway/index.html
ID: 32213 · 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 32234 - Posted: 10 Oct 2009, 13:30:36 UTC - in response to Message 32213.  

I have my site updated with the newer 0.20b ATI apps now.

http://www.arkayn.us/milkyway/index.html

0.20b doesn't work, at least not for me. Since you've taken 0.20 off your site I take it that you think 0.20b will work for all?


ID: 32234 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile arkayn
Avatar

Send message
Joined: 14 Feb 09
Posts: 999
Credit: 74,932,619
RAC: 0
Message 32242 - Posted: 10 Oct 2009, 18:15:55 UTC - in response to Message 32234.  

They are not off the site, just put in the older folder.
http://www.arkayn.us/milkyway/older/index.htm


0.20b has my card screaming right now.
ID: 32242 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile uBronan
Avatar

Send message
Joined: 9 Feb 09
Posts: 166
Credit: 27,520,813
RAC: 0
Message 32243 - Posted: 10 Oct 2009, 18:21:02 UTC

I tried the new 0.20b on my 4850 and to my surprise all units failed
And started crash ?!? when i put back the 0.20ATI it runs normal

I have been testing many things since then but i am baffled why it fails

Its new, its relative fast... my new bicycle
ID: 32243 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Not2Nutz

Send message
Joined: 15 Aug 09
Posts: 9
Credit: 318,640
RAC: 0
Message 32244 - Posted: 10 Oct 2009, 19:08:39 UTC - in response to Message 32213.  

I have my site updated with the newer 0.20b ATI apps now.

http://www.arkayn.us/milkyway/index.html


I just installed 20b and it seems to be working fine, at least, on the first few MilkyWay WUs.

The previous v20 actually deleted itself (the app) when I first started Boinc. (see my previous post.) So I switched back to 19g which worked fine and ran 3 WUs at a time, just uning the GPU.

Upon installing 20b and starting Boinc it instantly deleted all of the MilkyWay WUs I already had queued to run, about 20 of them. That was worrisome. I don't know if this is standard practice following an upgrade. But I would like to know if it is, as it seems to be a strange behavior, like the previous v20.

So after those WUs disappeared, I waited a few minutes and another 10 were downloaded and started, but are running only one at a time. They are taking about 3-5 seconds a piece to finish. That is on an HD3870 with a dual-quadcore Xeon at 2.66GHz.

So far, 20b looks good!

n2n
ID: 32244 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Gary Roberts

Send message
Joined: 1 Mar 09
Posts: 56
Credit: 1,984,937,499
RAC: 0
Message 32266 - Posted: 11 Oct 2009, 6:40:26 UTC - in response to Message 32244.  

I just installed 20b and it seems to be working fine, at least, on the first few MilkyWay WUs.

Of course it works fine. It works fine for me as well and it would probably work fine for the vast majority if only they would really, really read the readme and not do anything until they fully covered and understood each point. It really is all there in the readme.

The previous v20 actually deleted itself (the app) when I first started Boinc. (see my previous post.) So I switched back to 19g which worked fine and ran 3 WUs at a time, just uning the GPU.

No it didn't delete itself. When you use AP (anonymous platform) you are telling BOINC precisely which application it has to use for specific tasks 'branded' with specific version numbers. V20b is described in the new app_info.xml and not V20. Therefore the AP mechanism will delete V20 because app_info.xml doesn't need it any more. This is exactly as intended to clean up old and no longer required versions of apps. I really don't understand why if V20b was working OK (as you say in your first statement), you got worried about V20 and then went back to a V19 variant???

Upon installing 20b and starting Boinc it instantly deleted all of the MilkyWay WUs I already had queued to run, about 20 of them. That was worrisome. I don't know if this is standard practice following an upgrade. But I would like to know if it is, as it seems to be a strange behavior, like the previous v20.

I presume you are now talking about an upgrade from V19 back to V20b. What you describe above is exactly what AP is supposed to do if you haven't made arrangements within the app_info.xml file to properly handle the differently 'branded' tasks. There are no differences in the tasks themselves but they do become 'branded' with the version number of the app in existence at the time they were downloaded. This 'branding' is stored in your state file (client_state.xml). There is no difference in version 'branding' for V20 and V20b. Both are <version_num>20</version_num> under AP. There is a big difference between V19x and V20x however and all you needed to do to prevent trashing your V19 'branded' tasks was to duplicate the <app_version> clause in app_info.xml so that both 'brandings' were specified as being allowed to be crunched by the new V20b app. After all, computers are extremely literal - they do exactly what they are told :-). It's just we humans that write the instructions that always get it wrong - we're not very good at reading and writing precise instructions :-).

So after those WUs disappeared, I waited a few minutes and another 10 were downloaded and started, but are running only one at a time. They are taking about 3-5 seconds a piece to finish. That is on an HD3870 with a dual-quadcore Xeon at 2.66GHz.

I presume you mean "3-5 seconds less a piece to finish."

So far, 20b looks good!

Precisely!! :-).

I think a lot of people reporting failures have got the _ati.exe and _amd.exe versions mixed up because they haven't properly read the readme. Using AP is not easy - particularly when you eventually want to bail out and go back to a fixed up stock app. You can research AP somewhere in the stuff on boinc.berkeley.edu. Stick 'anonymous platform boinc' into google and I'm sure you'll get lots of stuff to read and digest :-).

Cheers,
Gary.
ID: 32266 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile uBronan
Avatar

Send message
Joined: 9 Feb 09
Posts: 166
Credit: 27,520,813
RAC: 0
Message 32278 - Posted: 11 Oct 2009, 11:58:56 UTC
Last modified: 11 Oct 2009, 12:00:39 UTC

After fiddling with the darn catalyst drivers and many tests with settings all run smooth with 0.20b now.
I still wonder why once in certain timeframe the dual card system seems to be messed by unknown factor with the driver.
While running 1 card does run super stable but if 2 .... ppffff they keep you busy all the time.
Especially when a new release is done from the apps from collatz or MW xD
I think CP has found a new way of keep us busy ;)

Sidenote the new 0.20b version does make a significant impact on 4830/4770 cards if you are up to the trouble its worth it :D
Its new, its relative fast... my new bicycle
ID: 32278 · 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 32296 - Posted: 11 Oct 2009, 19:42:05 UTC


Version 0.20b for ATI optimized apps is now on brilliantsite.com

All zip files in the downloads begin with ati_ to denote they are for ati GPU processing (not CPU). Some of the file names end with _amd.zip which denote that amdcal* drivers are required. See the readme.txt in each download.


ID: 32296 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Not2Nutz

Send message
Joined: 15 Aug 09
Posts: 9
Credit: 318,640
RAC: 0
Message 32327 - Posted: 12 Oct 2009, 15:03:44 UTC - in response to Message 32266.  

"I presume you mean "3-5 seconds less a piece to finish."

No, actually, it was 3-5 "minutes" to complete each WU.


My mistake in using seconds instead of minutes.

After running 20b for two days I have decided that I prefer the way 19g runs. It seems to be quite faster, going thru 3 WUs at a time. So I am going to switch back.


"No it didn't delete itself."

And I want to make THIS point perfectly clear. The v20 application code was instantly DELETED. You can hedge or quibble in your explanation all you want. Whether it was by design, programing, or program error, or the Man-In-The-Moon, I REALLY DON'T CARE. "DELETED" means that it was gone within seconds WITHOUT EXPLANATION or note in the Log. I saw this behavior with my own eyes. And I am NOT mistaken. Version 20 DELETED the application code immediately each of the 5 times I started BOINC with it installed. That's the way it worked (?) on my machine; others may have experienced something else. For ME, Version 20 was a waste of time.


"if you haven't made arrangements within the app_info.xml file to properly handle the differently 'branded' tasks. "

I don't do anything with the XML app_info file, let alone "make arrangements". I just copy it into the correct place and run. I ASSUMED that this application would be ready-to-go when posted. If not, then it defeats the entire plan of upgrading to using it. Although I don't miind turning my machinee over for use my a project, I personally, don't have the time to screw around with this stuff.


I think there needs to be some/more QA (quality assurance) done on these applications prior to their release.


All I tried to accomplish by posting was to provide a little feedback that might be useful and/or encouraging. I really HATE it when someone uses my post to make a speech and to preach to me. And I don't appreciate the tone of the response that I received. Sounds very arrogant.

Just my opinion...

n2n

ID: 32327 · 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 32335 - Posted: 12 Oct 2009, 19:27:37 UTC - in response to Message 32327.  

I ASSUMED that this application would be ready-to-go when posted.


That is a good assumption, all the versions have worked 'out of the box' for me with options to tweak things. There will always be misunderstandings and explanations and detail that we have to struggle with at times. But then there always some who want to get the best for all of us ;)


ID: 32335 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Divide Overflow
Avatar

Send message
Joined: 16 Feb 09
Posts: 109
Credit: 11,089,510
RAC: 0
Message 32346 - Posted: 13 Oct 2009, 5:45:31 UTC - in response to Message 32327.  

I don't do anything with the XML app_info file, let alone "make arrangements". I just copy it into the correct place and run. I ASSUMED that this application would be ready-to-go when posted.
...
I think there needs to be some/more QA (quality assurance) done on these applications prior to their release.

You'd probably be better off sticking with the standard project applications. You made some pretty poor assumptions.
ID: 32346 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile TomaszPawel
Avatar

Send message
Joined: 9 Nov 08
Posts: 41
Credit: 92,786,635
RAC: 0
Message 32403 - Posted: 15 Oct 2009, 20:09:07 UTC
Last modified: 15 Oct 2009, 20:20:12 UTC

OK it it time...

It is my first Radeon, and it is HD5870.

I install BM 6.10.13, drivers 1.4.427, and astronomy_0.20b_ATI_x64_ati opt.

OS is Windows 7 RC 64bit

Everything works very well!

Example:

<core_client_version>6.10.13</core_client_version>
<![CDATA[
<stderr_txt>
Running Milkyway@home ATI GPU application version 0.20b (Win64, CAL 1.4) by Gipsel
instructed by BOINC client to use device 0
CPU: Intel(R) Core(TM)2 Quad  CPU   Q8200  @ 2.33GHz (4 cores/threads) 2.8701 GHz (234ms)

CAL Runtime: 1.4.427
Found 1 CAL device

Device 0: ATI Radeon HD5800 series (Cypress) 1024 MB local RAM (remote 831 MB cached + 831 MB uncached)
GPU core clock: 850 MHz, memory clock: 300 MHz
1600 shader units organized in 20 SIMDs with 16 VLIW units (5-issue), wavefront size 64 threads
supporting double precision

Starting WU on GPU 0

main integral, 320 iterations
predicted runtime per iteration is 61 ms (33.3333 ms are allowed), dividing each iteration in 2 parts
borders of the domains at 0 800 1600
Calculated about 8.22242e+012 floatingpoint ops on GPU, 1.23583e+008 on FPU. Approximate GPU time 20.6122 seconds.

probability calculation (stars)
Calculated about 3.34818e+009 floatingpoint ops on FPU.

WU completed.
CPU time: 2.51162 seconds,  GPU time: 20.6122 seconds,  wall clock time: 21.768 seconds,  CPU frequency: 2.87016 GHz

</stderr_txt>
]]>


Default app info is in use:

<app_info>
 <app>
 <name>milkyway</name>
 </app>
 <file_info>
  <name>astronomy_0.20b_ATI_x64_ati.exe</name>
  <executable/>
 </file_info>
 <file_info>
  <name>brook64.dll</name>
  <executable/>
 </file_info>

 <app_version>
  <app_name>milkyway</app_name>
  <version_num>20</version_num>
    <flops>1.0e11</flops>
    <avg_ncpus>0.05</avg_ncpus>
    <max_ncpus>1</max_ncpus>
    <coproc>
      <type>ATI</type>
      <count>1</count>
    </coproc>
    <cmdline></cmdline>
  <file_ref>
   <file_name>astronomy_0.20b_ATI_x64_ati.exe</file_name>
   <main_program/>
  </file_ref>
  <file_ref>
   <file_name>brook64.dll</file_name>
  </file_ref>
 </app_version>

</app_info>


But I see this in GPU-Z:



As you see GPU Load is droping when WU is ending and new WU is starting.

So I read ReadMe and there "are several options one can put in there"
<cmdline></cmdline>

I am just starting ATI crunching... so i ask experts for sugestion how to change the option in order to minimize drop of gpuload in counting... The best would be non stop 99% load :)

Now Boinc is crunching one MW task, ends it, and start another...

Now it looks like HD5870 is crunching 900 - 905 WU at 6 hours... so it is 192420 - 193489 c per 24 hours.
A proud member of the Polish National Team

COME VISIT US at Polish National Team FORUM

ID: 32403 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Previous · 1 . . . 34 · 35 · 36 · 37 · 38 · 39 · 40 · Next

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

©2024 Astroinformatics Group