It does matter on quads - there is a high CPU utilisation. Where no CPU WUs are run, then times are at or about 0.23. Where a CPU WU is run, times are 25% or so slower.

So for those who want to run CPU apps, its currently a no no, and for the moment GPU app is elsewhere. Zy

I'm seeing very high CPU usage here with .59 also. In fact .57 was better. .59 uses a CPU core for each GPU while at least .57 shared the core for both GPUs. As a comparison, .23 was 5-10% faster and used virtually no CPU at all. In fact it's hammering the CPU so badly with .59 that it's bringing the CPU project tasks to their knees. I've reluctantly had to suspend MW work until this is fixed :(

I am running two workunits and it does increase the GPU usage up to 100% and the CPU usage is still at 0%

How are you running two workunits concurrently ?

With an app_info.xml

<app_info>
 <app>
 <name>milkyway</name>
 </app>
 <file_info>
  <name>milkyway_0.59_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>0.05</avg_ncpus>
    <max_ncpus>1</max_ncpus>
    <coproc>
      <type>ATI</type>
      <count>0.5</count>
    </coproc>
    <cmdline></cmdline>
  <file_ref>
   <file_name>milkyway_0.59_windows_intelx86__ati14.exe</file_name>
   <main_program/>
  </file_ref>
 </app_version>
</app_info>

With an app_info.xml

<app_info>
 <app>
 <name>milkyway</name>
 </app>
 <file_info>
  <name>milkyway_0.59_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>0.05</avg_ncpus>
    <max_ncpus>1</max_ncpus>
    <coproc>
      <type>ATI</type>
      <count>0.5</count>
    </coproc>
    <cmdline></cmdline>
  <file_ref>
   <file_name>milkyway_0.59_windows_intelx86__ati14.exe</file_name>
   <main_program/>
  </file_ref>
 </app_version>
</app_info>

ahh, i see...for some reason i remember being told that i could only run 1 MW@H tasks at a time...i guess that isn't correct.

---

You can run more than one WU on a GPU, but if you do it, dont run more than two. It depends on how big the WU is as to whether or not its worth it. The gain is in the load/unload of Data CPU-GPU, so it also depends on how the individual application handles it as well. It takes a finite time to pass the data from CPU main memory to GPU memory before it can start crunching.

Is it worth the hassle .... horses for courses, there are so many variables. Collatz for example not worth it, the percentage gain compared to WU size is puny. MW is at the other end of the scale with a small WU, and certainly a marginal gain was to be had on 0.23 - again many factors not the least which card was being used and its GPU bandwidth.

Bottom line .... you might gain a second, two if you are lucky, chances are less than a second per WU. Broadly speaking running two on a GPU will result in both being crunched at twice the time period of one ie, no gain as such ... but back to the paragraphs above :)

If a 5870 flung out - say 40 WUs and hour, then you *might* save circa 40 seconds or so .... say .... 240 an hour or around 2 WUs, thats .... say 10,000 credits a day. 1 or 2 % kind of thing.

The other main factor is apparent using Collatz as an example - with Collatz it reserves a chunk of main memory equivalent to 30% of GPU memory per WU - for a 1Gb card thats around 330Mb, have - say - two gpus running and 660Mb+ main memory is eaten. Run 2 WUs per GPU, 1.3Gb main memory .... can be an ouch. MW is not that kind of problem as its data sets are small, but worth baring in mind elsewhere.

Pays your money takes your choice as they say :)

Regards
Zy

I am running two workunits and it does increase the GPU usage up to 100% and the CPU usage is still at 0%

cedricdd, thanks for posting your app_info. Running 2 WUs at once does increase the GPU usage to 99% but with either 1 or 2 GPUs the CPU usage here is still 1 full CPU core, sometimes grabbing 2 cores (no difference between the 1 and 2 GPU machines, they all vary between 1 and 2 full CPU core usage).

BTW, minor change in the app_info:

<version_num>57</version_num> changed to <version_num>59</version_num>

What OS version and drivers are you using that you're seeing low CPU usage (computers hidden)?
Win7-64, 10.12 here. Thanks again.

Zy, thanks for putting that in terms of real world gains. i was going to trying running 2 MW@H tasks simultaneously to see what the gains would be, but already it doesn't seem worth trying. if doubling the work load isn't going to double the efficiency, or at the very least result in 2 tasks taking significantly less time to crunch than a single task would, then i think i'll pass on the experimentation...

---

.... if doubling the work load isn't going to double the efficiency, or at the very least result in 2 tasks taking significantly less time to crunch than a single task would, then i think i'll pass on the experimentation...

It *might* be worth trying with the current stage of development of 0.59, as its GPU utilisation *can* at present be low depending on what else is happening on the PC, the card type and processor. Maybe get a short term gain circa 1 or 2%, maybe more, but dont hold your breath. Certainly any marginal gain will evaporate again as soon as the application gets tweeked and GPU utilisation goes up.

Back to the famous Horses again rofl :)

Regards
Zy

I am running two workunits and it does increase the GPU usage up to 100% and the CPU usage is still at 0%

cedricdd, thanks for posting your app_info. Running 2 WUs at once does increase the GPU usage to 99% but with either 1 or 2 GPUs the CPU usage here is still 1 full CPU core, sometimes grabbing 2 cores (no difference between the 1 and 2 GPU machines, they all vary between 1 and 2 full CPU core usage).

BTW, minor change in the app_info:

<version_num>57</version_num> changed to <version_num>59</version_num>

What OS version and drivers are you using that you're seeing low CPU usage (computers hidden)?
Win7-64, 10.12 here. Thanks again.

At the end of each tasks, when it reach 100% I have a 0% for the GPU and for 4-5 seconds the task is using one CPU at a 100% but it was the same when I was running only one task. Exept for the end the CPU usage is at 0%, my config is Win-7 64 bits and driver 11.4 (1.4.1353).

OK a little update from me.

HD5850, win 7 64 SP1, BM 6.10.58, Cat 11.2, Pent D 3.2

Using the aforementioned App_info file I could indeed get two MW units running together using count=0.5. However .......

(1) I have to access this machine remotely and it just continued to lock up on me.
(2) when I turned the CPU tasks on the GPU utilisation dropped to 35%!

I've left the App_info in place and changed the count to 1, so at least I can access it. But if I turn the CPU tasks on the GPU drops to 65% again. It may be working for others but not for me. All I want to do is to be where I was before with 100% GPU on MW, and two CPU tasks on Seti. But I'll wait for a solution.

But, many thanks to those that have posted thoughts and ideas, we all appreciate it :-)

LATE EDIT - I tried 85% CPU to no avail, but on a twin core processor, perhaps I should have tried 50%? I.E. 1 core for the CPU task and 1 core for MW ?? I could accept crunching 1 Seti instead of none.

---

Don't drink water, that's the stuff that rusts pipes

i now believe this somehow has to do with my running E@H on all 6 CPU cores. when i set BOINC to use only 85% of the processors (forcing BOINC to use only 5 of my 6 CPU cores), or when i suspend E@H altogether (freeing up my CPU entirely), GPU load jumps back up to a constant and steady 98-99%. accordingly, run times approach the lower end of the crunch-time ranges i've historically observed. so my revised "big picture" is this: when i run E@H on all 6 CPU cores, MW@H GPU crunching efficiency suffers a bit in the form of slightly increased run times. when i leave a single CPU core free (or if i suspend E@H altogether and leave all 6 cores free), MW@H GPU crunching efficiency doesn't suffer at all, and run times are slightly decreased.

This seems to confirm what I've been finding. It seems that the slow happens with a high CPU load (at least in Windows. It seems to not happen for me in Linux). The time is about what it should be with a low CPU load, and quite a bit higher if the CPU load is light. This is still a problem to fix though.

Matt, I believe the way to sort this would be to make sure that the CPU part of the MW app runs in normal priority, or at least higher priority than the CPU-only tasks. (After first having made certain that the GPU app releases the CPU for a sane amount of time while waiting for the GPU to report back, of course.)

---

I am also seeing the customary negligible CPU time being used for GPU WU's even for version 0.59.

However, what does appear new is an up to 2sec delay at the UI to show keystrokes echoed to the screen. This problem stops when the GPU WU ends and then starts up again when a new WU starts.

This is happening with an ATI HD5870 under Windows/XP running Catalyst 10.12. Oddly enough, Catalyst claims GPU usage is less than 90% when this is happening.

This is happening with an ATI HD5870 under Windows/XP running Catalyst 10.12. Oddly enough, Catalyst claims GPU usage is less than 90% when this is happening.

i was going to suggest trying the Catalyst 11.3 drivers b/c they have been working for me and we have some similarities to our setups. but then it dawned on me that, judging by your comments, you're running your display with the same 5870 that is crunching MW@H tasks. i on the other hand am using my mobo's integrated video for the display to that my 5870 is completely devoted to MW@H.

nevertheless, the Catalyst 11.3 drivers might still be worth a shot, as Windows XP 32-bit users have generally been having luck with these drivers.

---

Hi Sunny129,

I on the other hand am using my mobo's integrated video for the display to that my 5870 is completely devoted to MW@H.

How did you do that? If you force the BIOS to use the on-board video, will boinc still see a 3rd party graphics card and use it? I thought that a computer uses one or the other? I tell you, the learning curve around here is getting steeper every day!

---

Don't drink water, that's the stuff that rusts pipes

How did you do that? I thought that a computer uses one or the other? I tell you, the learning curve around here is getting steeper every day!

lol tell me about it. i was completely in the dark no more than a month ago myself. i couldn't possibly explain to you how i managed it without completely derailing the topic of this thread, so i'll link you to the thread in which several folks were able to help me troubleshoot and solve the issue. long story short, it all boiled down to having to place a cc_config.xml file in the BOINC data directory that contained a directive that forces BOINC to ignore either one GPU (my integrated HD 3300 GPU) or the other (my HD 5870 GPU). it looks like this:

<cc_config>
<options>
<ignore_ati_dev>0</ignore_ati_dev>
</options>
</cc_config>

in the above configuration, device 0 (the integrated HD 3300 GPU) is disabled, which allows MW@H to run flawlessly on the 5870 without getting confused and sending work to a GPU that can't handle MW@H tasks and causes them to error out. likewise, when i change the 0 to a 1, device 1 (the 5870 GPU) is ignored by BOINC, and it allows SETI@Home Multibeam and Astropulse tasks to run flawlessly without getting confused by the integrated video, which causes computation errors. i know the latter doesn't make much sense b/c i'm ignoring the GPU that actually crunches the S@H tasks. but its really just playing a trick on BOINC, while the S@H MB and AP apps themselves still seem to override BOINC and use the 5870 anyways.

obviously the situation isn't ideal b/c i have to edit the cc_config.xml file and restart BOINC every time i want to switch from MW@H to S@H or vice versa. what would be ideal is if both MW@H apps and S@H apps recognized the GPU's in the same way - then i wouldn't have to edit the cc_config.xml file and restart BOINC every time i switched between those projects/apps - i could simply suspend one project and restart the other. unfortunately there still exist bugs that don't allow me to do this, and ultimately cause tasks to error out immediately if i try to switch between GPU projects in this manner...hence why i have to edit the cc_config and restart BOINC when switching projects.

at any rate, here is the thread that should be able to get you moving in the right direction: http://setiathome.berkeley.edu/forum_thread.php?id=63294

...its somewhat long-winded, but worth the read

---

Q9450 Win 7 64bit. HD5970 w 11.4 early preview drivers. MW 0.59 still taking a whole CPU core per GPU wu.

Even though I watched a few wu's with each taking a whole core at the time of writing the above, I now am not seeing the same behaviour. Excellent!

I am running two workunits and it does increase the GPU usage up to 100% and the CPU usage is still at 0%

cedricdd, thanks for posting your app_info. Running 2 WUs at once does increase the GPU usage to 99% but with either 1 or 2 GPUs the CPU usage here is still 1 full CPU core, sometimes grabbing 2 cores (no difference between the 1 and 2 GPU machines, they all vary between 1 and 2 full CPU core usage).

BTW, minor change in the app_info:

<version_num>57</version_num> changed to <version_num>59</version_num>

What OS version and drivers are you using that you're seeing low CPU usage (computers hidden)?
Win7-64, 10.12 here. Thanks again.

At the end of each tasks, when it reach 100% I have a 0% for the GPU and for 4-5 seconds the task is using one CPU at a 100% but it was the same when I was running only one task. Exept for the end the CPU usage is at 0%, my config is Win-7 64 bits and driver 11.4 (1.4.1353).

I'm baffled. Have been monitoring these machines for a few hours. They're consistently using 1-3 whole CPU cores while running the above app_info. Same without the app info. I tried the 11.4 preview drivers that you're using on one machine and the results are still 1 to 3 core usage on a quad, usually 2 or 3. Tried using the -r# commandline, no improvement until the # gets so high that the GPU utilization drops throuh the floor and the WU times get ridiculously slow. At this point I sure would like to have .23 back. Never had a bit of trouble with it.

Q9450 Win 7 64bit. HD5970 w 11.4 early preview drivers. MW 0.59 still taking a whole CPU core per GPU wu.

Even though I watched a few wu's with each taking a whole core at the time of writing the above, I now am not seeing the same behaviour. Excellent!

I tell lies.

Just watched one get 80 sec of CPU time and note the previous one has 50 sec of CPU time. I also have some with only 5 seconds of CPU time and others with 15 seconds and some with 30 seconds. All over the shop. I wonder if there is a pattern?

... obviously the situation isn't ideal b/c i have to edit the cc_config.xml file and restart BOINC every time i want to switch from MW@H to S@H or vice versa ...

Did something related a while back switching cc_config back and forth for logging purposes. Can't find the files I used, but IIRC I wrote a couple of batch files that would suspend the projects, copy over the cc_config, force a re-read of cc_config, and resume the projects.

It went something like this: wrote two text files that were the cc_config files, called them something other than cc_config, wrote two batch files that would suspend, force copy the proper new cc_config, force a re-read, then resume. That way with a click of a shortcut on my desktop I could perform the switch.

I used, again IIRC, the boinccmd tool with the --project and the --read_cc_config operators. All the info is in the Boinc wiki. Good luck if you decide to try it.

http://boinc.berkeley.edu/wiki/Boinccmd_tool

Q9450 Win 7 64bit. HD5970 w 11.4 early preview drivers. MW 0.59 still taking a whole CPU core per GPU wu.

Even though I watched a few wu's with each taking a whole core at the time of writing the above, I now am not seeing the same behaviour. Excellent!

I tell lies.

Just watched one get 80 sec of CPU time and note the previous one has 50 sec of CPU time. I also have some with only 5 seconds of CPU time and others with 15 seconds and some with 30 seconds. All over the shop. I wonder if there is a pattern?

Many 117 second WUs with 81-82 seconds of CPU time, a few with much less. Somethings obviously not right with .59. Hope it's fixed soon.

@ Sunny129

Hmmmm,

I have an M4A78T-E which has the HD 3300 (RS780D), along with an HD 4850.

I'm running straight stock for MW but also crunch Collatz and I'm not having problems with trying to start MW tasks on the integrated since the update to 59.

On the earlier versions it would try and start one on the 3300 if I snoozed the GPUs and then re-enabled them though.