Is anyone running 2 WUs anymore?

So, just have to ask - with the new WU series and all from MW. Is anyone still running an app_info file to get 2 WUs simultaneously, and is this increasing the credit like it used to?
I'm thinking I used to get more points/day out of my 5870...but maybe not, and I just crunched DNETC so long that I was used to better.:confused:

For the record, I'm "stock" application settings right now - no app file, and one WU at a time...you can see my output on my stats @ Free-DC.
Although - it seems I might not be getting an accurate measurement because so much Collatz is going on. I have MW set to 100% and Collatz set to 0%, so it's only grabbing Collatz when it can't get MW WUs. And judging by how much Collatz I got today, MW availability must be quite intermittent for me.

Dont know about everyone else but I am running 2 WU's at a time with no app.

Before ATI implemented full OpenCL support, I used to run 2 wu's on my single 4850 all the time. But after the newer drivers came out my card seemed to slow down as I noticed completion times were not quite as short as before. Also my GUI seemed more sluggish.

Has anyone found an answer to this? I'm going to update to the new .60 version to see if that helps, but my numbers are down on both my 4870 cards running 2 WU at a time.

Yes you should run 2 wu at a time. It is more important now than it used to be because with the new app there is a significant dead time in between wu where the GPU goes idle.

If you go to the Number crunching forum on MW and look for Optimized Apps and follow the link you will find the new app with an app_info file. You change "count" to .5 just like on the old ones and it runs 2 wu .... .33 and it runs 3.

DO NOT try to use the p3 b-1 setting in the command line or it will error all wu!

Quote:

---

Originally Posted by Bryan

DO NOT try to use the p3 b-1 setting in the command line or it will error all wu!

---

God Bless Gipsel...

Quote:

---

Originally Posted by Bryan

Yes you should run 2 wu at a time. It is more important now than it used to be because with the new app there is a significant dead time in between wu where the GPU goes idle.

If you go to the Number crunching forum on MW and look for Optimized Apps and follow the link you will find the new app with an app_info file. You change "count" to .5 just like on the old ones and it runs 2 wu .... .33 and it runs 3.

DO NOT try to use the p3 b-1 setting in the command line or it will error all wu!

---

I have an App_info file to run 2 at a time. Do they actually already have an optimized ap for MW?

Quote:

---

Originally Posted by Fire$torm

Before ATI implemented full OpenCL support, I used to run 2 wu's on my single 4850 all the time. But after the newer drivers came out my card seemed to slow down as I noticed completion times were not quite as short as before. Also my GUI seemed more sluggish.

---

I ran into the same problem running 2 GPU's (4 running at a time)

Quote:

---

Originally Posted by Crazybob

I have an App_info file to run 2 at a time. Do they actually already have an optimized ap for MW?

---

Optimized, no. But they have added in options to the code that you can change in the app_info file. So you can change settings to make the WU run faster...

I don't much care if the screen gets sluggish, because the DrPop rig runs headless now!:p What I am curious about, is whether or not it's worth it to mess with running 2 WUs at a time any more? The HD5870 in there is definitely not pegged out like it was with DNETC, so maybe 2WUs at a go would make sense on that rig?
Thanks for the help...just looking for more points for the team...we're headed in the right direction!

Depends on the cards. Sometimes there is less lag when running 2, 3 or even 4 WUs. 3-4 seems to be ok for 5970s, 4 seems best for my 5870, two is fine for a 4770. It does not seem to make sense, but Matt Arsenault stated there is a link to the estimated computing power of the card.

OK, I am going to have to try this. Can someone please copy their new app file and paste it here for us? Thanks to anyone who has one.
In MSI Afterburner, I can see HUGE gaps of time where the 5870 is spinning down, sitting for like 2 full seconds, and then ramping back up. That would definitely explain my low numbers.
In contrast, card 2 in the rig - a GTX 460 on PG is at 99% pegged almost continually.

Okay I'll put it on your tab :-" I run 2 wu on MW on all 5870s/5970s.

<app_info>
<app>
<name>milkyway</name>
</app>
<file_info>
<name>milkyway_0.62_windows_intelx86__ati14.exe</name>
<executable/>
</file_info>

<app_version>
<app_name>milkyway</app_name>
<version_num>57</version_num>
<plan_class>ati14</plan_class>
<flops>1.0e11</flops>
<avg_ncpus>.5</avg_ncpus>
<max_ncpus>1</max_ncpus>
<coproc>
<type>ATI</type>
<count>.5</count>
</coproc>
<file_ref>
<file_name>milkyway_0.62_windows_intelx86__ati14.e xe</file_name>
<main_program/>
</file_ref>
</app_version>

</app_info>

Just curious, has anyone tried this for other GPU projects? Running Collatz my GUI isn't really slowed down at all (even though the GPU is @ 100%). Wondering if that might help in the daily take.

Quote:

---

Originally Posted by Bryan

Okay I'll put it on your tab :-" I run 2 wu on MW on all 5870s/5970s.

<app_info>
<app>
<name>milkyway</name>
</app>
<file_info>
<name>milkyway_0.62_windows_intelx86__ati14.exe</name>
<executable/>
</file_info>

<app_version>
<app_name>milkyway</app_name>
<version_num>57</version_num>
<plan_class>ati14</plan_class>
<flops>1.0e11</flops>
<avg_ncpus>.5</avg_ncpus>
<max_ncpus>1</max_ncpus>
<coproc>
<type>ATI</type>
<count>.5</count>
</coproc>
<file_ref>
<file_name>milkyway_0.62_windows_intelx86__ati14.e xe</file_name>
<main_program/>
</file_ref>
</app_version>

</app_info>

---

Is there any reason that the version number in here is 57 and not 62, or does the system even pay attention to that?

Quote:

---

Originally Posted by Crazybob's Son

Just curious, has anyone tried this for other GPU projects? Running Collatz my GUI isn't really slowed down at all (even though the GPU is @ 100%). Wondering if that might help in the daily take.

---

It did not with DNETC (b/c it was a multi-threaded app) and is not needed for Collatz (for various reasons). It DOES help with DistrRTgen's CUDA app, and I've heard rumors it helps with GPUgrid. Haven't looked at anything else.

Quote:

---

Originally Posted by Crazybob

Is there any reason that the version number in here is 57 and not 62, or does the system even pay attention to that?

---

The version of wu is currently 57 like the number of the original "new" app. One person I saw post said he had changed that line to 62 and all wu bombed. I don't know this for a fact, but I do know 57 works.

Quote:

---

Originally Posted by Crazybob's Son

Just curious, has anyone tried this for other GPU projects? Running Collatz my GUI isn't really slowed down at all (even though the GPU is @ 100%). Wondering if that might help in the daily take.

---

Don't run 2 wu on Collatz because on that project BOINC won't ask for new tasks!!!!! You should elevate the priority though with the p3 b-1 cmdline option.

I'm running 2 wu on PG on the Cudas ... there is debate whether it actually helps. Since I'm new to the Cuda/PG game I haven't tried going w/ just 1 wu. I'm VERY content with how things are working currently.

GPUGrid definitely is improved running 2 wu.

What is the minimum CPU allocation you can use without impacting the speed of the GPU? I am currently using two tasks, each using .5 CPU.

Quote:

---

Originally Posted by Bryan

The version of wu is currently 57 like the number of the original "new" app. One person I saw post said he had changed that line to 62 and all wu bombed. I don't know this for a fact, but I do know 57 works.

---

I'm running mine with version 62, don't seem to have a problem. Wasn't sure if that was just a tag to show up in the BOINC manger.

Quote:

---

Originally Posted by zombie67

What is the minimum CPU allocation you can use without impacting the speed of the GPU? I am currently using two tasks, each using .5 CPU.

---

I don't have a real answer but like you I'm allocating .5 CPU per wu on the I7. The real reason there is to keep the memory hog Constellation from running more than 6 threads.

When they 1st came out with the new app it was almost a requirement to allocate that much(or more) but since the .62 app came out I imagine that is much higher than it needs to be.

Quote:

---

Originally Posted by Bryan

I don't have a real answer but like you I'm allocating .5 CPU per wu on the I7. The real reason there is to keep the memory hog Constellation from running more than 6 threads.

When they 1st came out with the new app it was almost a requirement to allocate that much(or more) but since the .62 app came out I imagine that is much higher than it needs to be.

---

I don't allocate any CPUs to feed the GPUs since 0.62. Wouldn't make sense running 4 WUs per GPU that way. I haven't cross checked my crunch times though. But they seem to be fine. BoincTasks says about 1-2 sec. CPU time per GPU minute.

Wanted to get in on the double points, but I think I screwed up when creating a new App_info.xml file. Can anyone spot my screw up?

- <app_info>
- <app>
<name>milkyway</name>
</app>
- <file_info>
<name>milkyway_seperation_0.82_windows_x86_64_ati1 4.exe</name>
<executable />
</file_info>
- <app_version>
<app_name>milkyway</app_name>
<version_num>82</version_num>
<plan_class>ati14</plan_class>
<flops>1.0e11</flops>
<avg_ncpus>.5</avg_ncpus>
<max_ncpus>1</max_ncpus>
- <coproc>
<type>ATI</type>
<count>.5</count>
</coproc>
- <file_ref>
<file_name>milkyway_seperation_0.82_windows_x86_64 _ati14.exe</file_name>
<main_program />
</file_ref>
</app_version>
</app_info

That code had a few typos all fixed.

Code:

---


<app_info>
 <app>
  <name>milkyway</name>
 </app>
 <file_info>
  <name>milkyway_seperation_0.82_windows_x86_64_ati14.exe</name>
  <executable/>
 </file_info>
 
 <app_version>
  <app_name>milkyway</app_name>
  <version_num>82</version_num>
  <plan_class>ati14ati</plan_class>
  <flops>1.0e11</flops>
  <avg_ncpus>0.05</avg_ncpus>
  <max_ncpus>1</max_ncpus>
  <coproc>
  <type>ATI</type>
  <count>0.5</count>
  </coproc>
 <file_ref>
  <file_name>milkyway_seperation_0.82_windows_x86_64 _ati14.exe</file_name>
 <main_program/>
 </file_ref>
 </app_version>
</app_info>

---

Edit: If you still have issues with the above code then you could always try arkyns code ---> http://www.arkayn.us/forum/index.php...a=view;down=32

Thanks for the info! Seems to be running now.:)

Anyone done this on a Linux system?

Quote:

---

Originally Posted by trigggl

Anyone done this on a Linux system?

---

Nope, sorry. Although arkayn does have an opti-app for 32bit Linux and 64bit FreeBSD here. So his app should allow for it and he usually has the app_info packaged with them.