Welcome to MilkyWay@home

Testing ATI Application Availability

Message boards : Number crunching : Testing ATI Application Availability
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 3 · 4 · 5 · 6 · 7 · Next

AuthorMessage
Profile [AF>Occitania>Lengadocia...

Send message
Joined: 30 Mar 08
Posts: 25
Credit: 75,915,107
RAC: 0
Message 35609 - Posted: 13 Jan 2010, 19:54:48 UTC
Last modified: 13 Jan 2010, 19:55:35 UTC

After some tests on BOINC 6.10.25 W7 64bits and ATI appl 0.21.

I suppose that we loose possibility to process 2 WU at a time without app_info.xml.
Am I right?

I see a decrease in GPU load between each WU which is covered with 2 WUs at a time.

Switching back to 0.20b stops downloading of WUs (not tried more than an hour).
Editing app_info.xml seens not to work (perhaps I forgot something).
ID: 35609 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile cornel
Avatar

Send message
Joined: 28 Feb 09
Posts: 38
Credit: 10,200,014
RAC: 0
Message 35610 - Posted: 13 Jan 2010, 20:49:11 UTC
Last modified: 13 Jan 2010, 20:57:30 UTC

Jean Paul, I also had his video card until about six months ago. The problem is it cannot crunch, therefore it is useless here. I searched the shops and found a nice HD 4770 at a low cost and bought it. It now crunches happily at Collatz (not MW).
Hope you find a CAL compliant video card too!

Ah, btw, take a look at these strange messages:

13.01.2010 22:51:29 Milkyway@home work fetch resumed by user
13.01.2010 22:51:31 Milkyway@home Sending scheduler request: To fetch work.
13.01.2010 22:51:31 Milkyway@home Requesting new tasks for CPU
13.01.2010 22:51:36 Milkyway@home Scheduler request completed: got 0 new tasks
13.01.2010 22:51:36 Milkyway@home Message from server: No work sent
13.01.2010 22:51:36 Milkyway@home Message from server: Your computer has no NVIDIA GPU
13.01.2010 22:51:36 Milkyway@home Message from server: Jobs for CPU are available, but your preferences are set to not accept them



What should I do to get GPU workunits?
ID: 35610 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile banditwolf
Avatar

Send message
Joined: 12 Nov 07
Posts: 2425
Credit: 524,164
RAC: 0
Message 35611 - Posted: 13 Jan 2010, 22:10:08 UTC - in response to Message 35610.  
Last modified: 13 Jan 2010, 22:10:21 UTC


What should I do to get GPU workunits?

I would say to go back to using the opti app.
Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.
ID: 35611 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile arkayn
Avatar

Send message
Joined: 14 Feb 09
Posts: 999
Credit: 74,932,619
RAC: 0
Message 35617 - Posted: 13 Jan 2010, 23:18:15 UTC
Last modified: 13 Jan 2010, 23:29:32 UTC

Now if they can only figure out why it will not send ATI work via the anonymous platform.

Without it having to request it as CPU work and mess up debts.

For the time being, I am going to leave it as a GPU only request and turn back on Collazt work fetch to keep the 4830 busy.
ID: 35617 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Slicker [TopGun]

Send message
Joined: 20 Mar 08
Posts: 46
Credit: 69,382,802
RAC: 0
Message 35618 - Posted: 13 Jan 2010, 23:26:22 UTC - in response to Message 35609.  

I suppose that we loose possibility to process 2 WU at a time without app_info.xml.
Am I right?


Not necessarily, but it requires a change on the MW server. The same XML that you would have in the app_info.xml is also sent by the server when using standard apps. In order to get it to run 2 WUs at a time, the server needs to set <count>0.5</count> instead of <count>1.0</count> when it sends the response. I don't remember if that code is in the sched_customize.cpp or not, but that's the first place I'd look.

To get the app_info.xml to work, you should be able to take a look at the response xml in the boinc data folder (not the project folder) without using the app_info. Then edit your app_info to look the same but with <count>0.5</count> instead of <count>1.0</count>.
ID: 35618 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 35632 - Posted: 14 Jan 2010, 3:52:25 UTC - in response to Message 35618.  

I suppose that we loose possibility to process 2 WU at a time without app_info.xml.
Am I right?


Not necessarily, but it requires a change on the MW server. The same XML that you would have in the app_info.xml is also sent by the server when using standard apps. In order to get it to run 2 WUs at a time, the server needs to set 0.5 instead of 1.0 when it sends the response. I don't remember if that code is in the sched_customize.cpp or not, but that's the first place I'd look.

To get the app_info.xml to work, you should be able to take a look at the response xml in the boinc data folder (not the project folder) without using the app_info. Then edit your app_info to look the same but with 0.5 instead of 1.0.


Will this break things for users not using a GPU app?

ID: 35632 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Raimund Barbeln

Send message
Joined: 7 Oct 07
Posts: 25
Credit: 35,424,347
RAC: 2,176
Message 35635 - Posted: 14 Jan 2010, 6:18:35 UTC

Maybe you could add this parameter and some others, like the w parameter, to be set in the user settings, so everyone can choose what fits for him?
When life gives you lemons, make lemonade!
ID: 35635 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile cenit

Send message
Joined: 16 Mar 09
Posts: 58
Credit: 1,129,612
RAC: 0
Message 35637 - Posted: 14 Jan 2010, 7:35:55 UTC - in response to Message 35632.  

Will this break things for users not using a GPU app?

I can't see how.

Maybe the big problem would arise whenever another project comes out with a GPU app and <count>0.5</count>. Then we could have conflicts on the GPU... or really big slow downs...
Am I wrong?
ID: 35637 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile kashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
Message 35638 - Posted: 14 Jan 2010, 8:55:27 UTC
Last modified: 14 Jan 2010, 8:58:32 UTC

I would not set the default configuration for the automatically downloaded ATI application to process 2 tasks concurrently. Cards such as HD 3850 with 256MB memory cannot run 2 of the longer ATI tasks concurrently without slowing down almost to a standstill or crashing.

I may have misunderstood but I think Slicker was talking about contributors using an app_info_xml based on the details of the automatically downloaded 0.21 version in the sched_request_milkyway.cs.rpi.edu_milkyway.xml file in the BOINC folder. I don't understand how this would be any different from using the manually installed ATI v0.20b however because the application and brook files are the same. Well almost, the automatically downloaded version downloads brook64_ati.dll and copies and uses it as brook64.dll:
<file_ref>
<file_name>brook64_ati.dll</file_name>
<open_name>brook64.dll</open_name>
<copy_file/>

I have tried it anyway just using brook.64.dll but with the current server configuration once an app_info.xml file is used the anonymous platform is still treated as a CPU for work fetch and BOINC debt issues. It is for me and arkayn and others. The only one who has posted that his app_info.xml is still being treated as a GPU is Beyond, but I don't know what is different in his configuration or hardware that allows this.

I don't know why the automatically downloaded ATI version uses 1.29 CPUs for my HD 5870. A contributor with a 2 HD 5870s and a dual core CPU could not run any CPU projects at all. It may speed up processing very slightly to allocate a higher amount of CPU and thus leave one or more CPU cores free of CPU projects but this is really unnecessary and not the most efficient use of resources to process BOINC tasks, especially for computers with a lower number of CPU cores. The ATI 0.20b version has a default avg_ncpus value of 0.05.

I can understand the project preferring that only the automatically downloaded version is used to avoid misconfiguration issues and the resulting requests for help. However the default MilkyWay ATI application uses the GPU very efficiently and can generate a lot of heat. Users, especially those in hot climates or with hot running cards, sometimes need the ability to be able to reduce the GPU load by adjusting the w parameter in an app_info.xml file without having all the BOINC debt trouble and necessary babysitting caused by the anonymous platform being treated as a CPU. Certainly I am not able to operate my 4890 with the default MilkyWay configuration in an Australian summer or the card temperature will exceed 95°C. On most summer days it requires w1.2 parameter and GPU memory speed of 500MHz to process MilkyWay ATI tasks at a more reasonable temperature of 85°C or below.
ID: 35638 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile kashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
Message 35640 - Posted: 14 Jan 2010, 9:34:13 UTC

Here is an app_info.xml prepared from the sched_request_milkyway.cs.rpi.edu_milkyway.xml file using the default values of the auto download 0.21 version for a single 5870 @ 965MHz using Catalyst 9.11 on Windows 7 64-bit. There is no point in anyone using this because except for the high ncpus value (which you could change anyhow) it works exactly the same as the manually downloaded 0.20b version and is also treated as a CPU for BOINC work fetch and debt. It is just for interest.

<app_info>
<app>
<name>milkyway</name>
<user_friendly_name>MilkyWay@Home</user_friendly_name>
</app>
<file_info>
<name>milkyway_0.21_windows_x86_64__ati13ati.exe</name>
<executable/>
</file_info>
<file_info>
<name>brook64.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>milkyway</app_name>
<version_num>21</version_num>
<platform>windows_x86_64</platform>
<plan_class>ati13ati</plan_class>
<flops>629299200000.000000</flops>
<avg_ncpus>1.288992</avg_ncpus>
<max_ncpus>1.288992</max_ncpus>
<coproc>
<type>ATI</type>
<count>1.000000</count>
</coproc>
<cmdline></cmdline>
<file_ref>
<file_name>milkyway_0.21_windows_x86_64__ati13ati.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>brook64.dll</file_name>
</file_ref>
</app_version>
</app_info>
ID: 35640 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile valterc

Send message
Joined: 28 Aug 09
Posts: 23
Credit: 1,268,608,254
RAC: 125,302
Message 35644 - Posted: 14 Jan 2010, 11:11:09 UTC - in response to Message 35640.  

Well, I'm not able anymore to crunch MW workunits on ATI 4850.... I did it fine since a couple of days ago (using anonymous platform).

Now. I don't think it's a good idea to select CPU only tasks (getting some units to crunch with the GPU) in my preferences and micromanage the whole thing...

I also tried to restart BOINC without app_info.xml (this way loosing some important cmdline parameters) and all I got was

1/14/2010 11:58:10 AM	Milkyway@home	Message from server: No work sent
1/14/2010 11:58:10 AM	Milkyway@home	Message from server: ATI Catalyst 9.2+ needed to use GPU
1/14/2010 11:58:10 AM	Milkyway@home	Message from server: Your computer has no NVIDIA GPU


I run XP x64 with Catalyst 8.11 (wich is the best one to have on this OS).

Hope this will be solved somehow (switching to Collatz for the moment)
ID: 35644 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile kashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
Message 35646 - Posted: 14 Jan 2010, 11:46:17 UTC - in response to Message 35644.  
Last modified: 14 Jan 2010, 12:08:50 UTC

Yes brook32_amd.dll and brook64_amd.dll are in the download directory in readiness, but 0.21_windows_intelx86__ati13amd.exe and 0.21_windows_x86_64__ati13amd.exe are not there yet.

Still in testing phase before they release the 0.21 (ati13amd) versions I suppose.

Edit: I remember reading someone saying that the latest Catalyst version 9.12 fixes the instability experienced with the WinXP driver versions between 9.2 and 9.10/9.11.

So perhaps if you wish you could try Catalyst 9.12 and then you may be able to use the automatically downloaded 0.21 (ati13ati) version. You won't be able to adjust any parameters but on the positive side you will not need to micromanage BOINC to enable ATI tasks to continue to be downloaded.
ID: 35646 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ableeker

Send message
Joined: 24 May 09
Posts: 8
Credit: 11,850,851
RAC: 0
Message 35659 - Posted: 14 Jan 2010, 19:55:55 UTC

I've been using MW 0.20 on a 4850 running XP32 with Catalyst 8.12 for quite a while now because anything else caused problems, but I've been running 0.21 on Catalyst 9.12 for the last couple of days, and it seems to be working great!
ID: 35659 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile cornel
Avatar

Send message
Joined: 28 Feb 09
Posts: 38
Credit: 10,200,014
RAC: 0
Message 35660 - Posted: 14 Jan 2010, 20:08:37 UTC
Last modified: 14 Jan 2010, 20:29:31 UTC

Today I started getting workunits for GPU:

14.01.2010 22:02:34 ATI GPU 0: ATI Radeon HD 4700/4800 (RV740/RV770) (CAL version 1.4.344, 512MB, 1101 GFLOPS peak)


The result:

14.01.2010 22:06:18 Milkyway@home Starting task de_11_2s_const_3_23501_1263499526_0 using milkyway version 21
14.01.2010 22:06:20 Milkyway@home Computation for task de_11_2s_const_3_23501_1263499526_0 finished
14.01.2010 22:06:20 Milkyway@home Output file de_11_2s_const_3_23501_1263499526_0_0 for task de_11_2s_const_3_23501_1263499526_0 absent


L.E.: seems a project reset solved the problem, there were some old files in milkyway directory which caused the errors... Sorry for trashing so many workunits :(
ID: 35660 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 35668 - Posted: 15 Jan 2010, 2:59:31 UTC - in response to Message 35660.  


L.E.: seems a project reset solved the problem, there were some old files in milkyway directory which caused the errors... Sorry for trashing so many workunits :(


Don't worry about trashing workunits and testing things out :) The worst that will happen is you won't get credit for them. The way our algorithms run and the way things are set up client errors and lost workunits really don't hurt us much, if at all.
ID: 35668 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile valterc

Send message
Joined: 28 Aug 09
Posts: 23
Credit: 1,268,608,254
RAC: 125,302
Message 35671 - Posted: 15 Jan 2010, 10:40:33 UTC - in response to Message 35668.  

Thank you all for the hints about installing CAT 9.12 on XP x64, seems working and I am currently crunching (and better at all gettin valid results) MW workunits.

However it would be fine to have app_info availabile to fine tune the thing. I'm thinking mainly at the use of the 'w' parameter to reduce screen lag while I use this computer for work.

ID: 35671 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile David Glogau*
Avatar

Send message
Joined: 12 Aug 09
Posts: 172
Credit: 645,240,165
RAC: 0
Message 35695 - Posted: 16 Jan 2010, 1:01:15 UTC

I installed another 5970 card into an i7 box, bring the total up to four GPU's. detatched and reattached got all four running on the new apps. BOINC told me I was using .81 CPU for each WU and THREE of my (eight) CPDN models were listed as waiting for memory.

Needless to say I reinstalled the .20b opti apps, and now back to a normal .05CPU per Wu, with all eight CPDN models running.

So a little more fine tuning needed on the new apps before they are ready for the prime time I think.

PS: my 4850 dual core is running .38CPU's per WU. I wonder if it is the GPU's being utilized, or the fact they are running on a quad core Phenom?
ID: 35695 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile kashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
Message 35699 - Posted: 16 Jan 2010, 2:52:22 UTC - in response to Message 35695.  
Last modified: 16 Jan 2010, 3:00:02 UTC

I think that it is using some kind of calculation based on the flops value of the GPUs. The faster/more shaders your GPUs the more CPU cores it will steal.

Here's my 5870 at default 870MHz core clock using MilkyWay v0.21(ati13ati). Info from sched_ request file:

<avg_ncpus>1.163901</avg_ncpus>
<max_ncpus>1.163901</max_ncpus>
<flops>567936000000.000000</flops>

and here is the same 5870 at usual overclock of 965MHz:

<avg_ncpus>1.288992</avg_ncpus>
<max_ncpus>1.288992</max_ncpus>
<flops>629299200000.000000</flops>

Here is manually downloaded MilkyWay v0.20b:

<avg_ncpus>0.050000</avg_ncpus>
<max_ncpus>1.000000</max_ncpus>
<flops>100000000000.000000</flops>

I don't know how the flops value of the GPU is calculated though, on one install of v0.21(ati13ati) I got avg_ncpus and max_gpus of 1.99. Perhaps I had forgotten to reduce the GPU memory from 1300MHZ to 500MHz for that install after switching from Collatz.
ID: 35699 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 35719 - Posted: 17 Jan 2010, 4:03:42 UTC - in response to Message 35699.  

I applied a patch to the server which should hopefully fix the CPU utilization of the GPU apps. Let me know how it works out.
ID: 35719 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile David Glogau*
Avatar

Send message
Joined: 12 Aug 09
Posts: 172
Credit: 645,240,165
RAC: 0
Message 35736 - Posted: 17 Jan 2010, 11:02:09 UTC

Travis, you da man!

All running at .05CPU now. CPU usage bouncing mainly between 1 and 4% with occasional spikes into double figures. Good enough for me to use full time now.

I will be bringing down a box next week to convert to double oil cooled 5970's, so will try the new apps out on that to see how we go.
ID: 35736 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 . . . 3 · 4 · 5 · 6 · 7 · Next

Message boards : Number crunching : Testing ATI Application Availability

©2024 Astroinformatics Group