Multiple work units on one GPU

Petec888
Petec888
Joined: 3 Oct 06
Posts: 33
Credit: 1978975321
RAC: 0
Topic 197461

Can anyone direct me to a step by step post for getting an Einstein@Home client to run two work units on one GPU? I'm not sure whether to create a file called app_config.xml or app_info.xml, and I'm also not sure about how many Einstein work units will fit on a 2 GB VRAM card..

archae86
archae86
Joined: 6 Dec 05
Posts: 3161
Credit: 7267188417
RAC: 1666922

Multiple work units on one GPU

1. you don't need app_config.xml--nor app_info.xml here, instead...
2. go to your account page on the Einstein site
3. first click on the view link for "computers on this account" and review the location column to find whether the computer you wish to adjust is currently assigned to work, school, home, or "blank", which means default.
4. back up to the account page, and click on the Einstein@home preferences link
5. for the location (aka venue) to which your host was assigned:[pre] a. click the "edit preferences" link at the bottom of the list of current values
b. assure that the appropriate use xxxx GPU is ticked "yes" near the top
c. near the bottom adjust the numeric value in the GPU utilization factor of BRP apps box
d. here 1 means run one task, 0.5 means run 2, 0.33 means run 3, 0.25 means run 4
e. if you are running Gamma-ray pulsar work on your GPU, then make adjustments to the GPU utilization factor of FGRP apps[/pre]
Your 2GB card will hold more Perseus (BRP5) or Arecibo GPU (BRP4G) work than is remotely helpful, as they don't require much GPU RAM compared to what you have, but don't benefit very much from further increase beyond two, and in some cases give less production beyond some point. On the other hand, GRP work requires quite a lot of GPU RAM, and benefits considerably from running more jobs simultaneously. I think you can run three--not sure about four, probably not five, on your card.

Good luck.
If you did not have your computers hidden it would be easier for us to help you.

Holmis
Joined: 4 Jan 05
Posts: 1118
Credit: 1055935564
RAC: 0

Some additions to archae86's

Some additions to archae86's great guide:

The updated utilization factor will only be sent to and used by your computer when new work gets downloaded, an normal project update is not enough. I also recommend experimenting with the work cache set to some low values, if something goes wrong then not too many tasks will go down the drain and you have the opportunity to increase the cache to get the new work and thus new settings faster.

If you're using Windows then I highly recommend downloading GPU-Z or some equivalent program to be able to monitor GPU load, temp and memory usage.

Petec888
Petec888
Joined: 3 Oct 06
Posts: 33
Credit: 1978975321
RAC: 0

Thanks! I was mistaken about

Thanks! I was mistaken about the different projects, I realize now every project has a different method of attenuating work.

Next time I ask for help I will whisk the cloaks off of my stations.

David S
David S
Joined: 6 Dec 05
Posts: 2473
Credit: 22936222
RAC: 0

RE: e. if you are running

Quote:
e. if you are running Gamma-ray pulsar work on your GPU, then make adjustments to the GPU utilization factor of FGRP apps


I've had BRP set to 0.5 for some time, but never touched the GRP setting (until just now; I changed it to 0.5 for one venue). I also have no idea if either of my hosts does GRP on the GPU. I don't have the prefs set not to (as far as I can tell). One host currently has some GRPs on board, but they seem to all be assigned to the CPU. The other host doesn't crunch on the CPU beyond what is needed to feed the GPU. Is there any reason why I wouldn't want to change the GRP setting to 0.5 as well?

Also, I noticed it says the minimum setting is -1.0. What happens when you set a negative number?

David

Miserable old git
Patiently waiting for the asteroid with my name on it.

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2143
Credit: 2979214160
RAC: 791772

RE: RE: e. if you are

Quote:
Quote:
e. if you are running Gamma-ray pulsar work on your GPU, then make adjustments to the GPU utilization factor of FGRP apps

I've had BRP set to 0.5 for some time, but never touched the GRP setting (until just now; I changed it to 0.5 for one venue). I also have no idea if either of my hosts does GRP on the GPU. I don't have the prefs set not to (as far as I can tell). One host currently has some GRPs on board, but they seem to all be assigned to the CPU. The other host doesn't crunch on the CPU beyond what is needed to feed the GPU. Is there any reason why I wouldn't want to change the GRP setting to 0.5 as well?

Also, I noticed it says the minimum setting is -1.0. What happens when you set a negative number?


The usage of -1 was explained in Technical News, in message 129792

You can see which applications you are/have been running in either the 'Advanced' view in BOINC Manager (tasks tab, column 'Application'), or by filtering the task list linked from your account page on this website. You may find it helpful to refer to the applications page to help you fight your way through the acronym soup.

David S
David S
Joined: 6 Dec 05
Posts: 2473
Credit: 22936222
RAC: 0

RE: RE: RE: e. if you

Quote:
Quote:
Quote:
e. if you are running Gamma-ray pulsar work on your GPU, then make adjustments to the GPU utilization factor of FGRP apps

I've had BRP set to 0.5 for some time, but never touched the GRP setting (until just now; I changed it to 0.5 for one venue). I also have no idea if either of my hosts does GRP on the GPU. I don't have the prefs set not to (as far as I can tell). One host currently has some GRPs on board, but they seem to all be assigned to the CPU. The other host doesn't crunch on the CPU beyond what is needed to feed the GPU. Is there any reason why I wouldn't want to change the GRP setting to 0.5 as well?

Also, I noticed it says the minimum setting is -1.0. What happens when you set a negative number?


The usage of -1 was explained in Technical News, in message 129792


No wonder I didn't see it. I didn't read that thread, until now. And the mention was all the way down near the end.

Back to the more important question, is there any reason why I wouldn't want to set FGRP to 0.5 for either my GT 440 (1 CPU core reserved) or my GT 630 (no CPU work because it has more important things to do)?

David

Miserable old git
Patiently waiting for the asteroid with my name on it.

Holmis
Joined: 4 Jan 05
Posts: 1118
Credit: 1055935564
RAC: 0

RE: Back to the more

Quote:
Back to the more important question, is there any reason why I wouldn't want to set FGRP to 0.5 for either my GT 440 (1 CPU core reserved) or my GT 630 (no CPU work because it has more important things to do)?


Well, the last I heard the FGRP3 GPU-app only got sent to GPUs with at least 2GB of video-RAM. The post here by Bernd (msg#129330) announced it but it also says it's going to be lower subsequently. I haven't seen any info that it's actually has been lowered so you might not be able to get any at all.

Another problem is with the GT440 as the server contact log states:

Quote:
2014-03-18 18:49:26.9730 [PID=27913] [version] NVidia device (or driver) doesn't support OpenCL


That's definitely a show stopper as the FGRP3 GPU-app is OpenCL, probably needs a driver update.

As to the actual question of benefit you would probably need to test it your self with your running conditions. My guess though is that it would benefit as the actual GPU load is quite low with this app, it's still under development and a lot of the calculations is still done on the CPU. Because of this it really needs a full CPU core per instance as support. But this also means that the GPU has plenty of power left over for a second task to use, I don't have any numbers but I suspect that the run time per task running x1 or x2 would be quite close to one another.

David S
David S
Joined: 6 Dec 05
Posts: 2473
Credit: 22936222
RAC: 0

RE: RE: Back to the more

Quote:
Quote:
Back to the more important question, is there any reason why I wouldn't want to set FGRP to 0.5 for either my GT 440 (1 CPU core reserved) or my GT 630 (no CPU work because it has more important things to do)?

Well, the last I heard the FGRP3 GPU-app only got sent to GPUs with at least 2GB of video-RAM. The post here by Bernd (msg#129330) announced it but it also says it's going to be lower subsequently. I haven't seen any info that it's actually has been lowered so you might not be able to get any at all.

Another problem is with the GT440 as the server contact log states:

Quote:
2014-03-18 18:49:26.9730 [PID=27913] [version] NVidia device (or driver) doesn't support OpenCL

That's definitely a show stopper as the FGRP3 GPU-app is OpenCL, probably needs a driver update.

As to the actual question of benefit you would probably need to test it your self with your running conditions. My guess though is that it would benefit as the actual GPU load is quite low with this app, it's still under development and a lot of the calculations is still done on the CPU. Because of this it really needs a full CPU core per instance as support. But this also means that the GPU has plenty of power left over for a second task to use, I don't have any numbers but I suspect that the run time per task running x1 or x2 would be quite close to one another.


Okay. On the driver, I downloaded the newest one last week on the 440, and then they released another one this week, which I installed on the 630 but not the 440.

As to the CPU power needed, I think I'll just set the 630 to -1 and the 440 back to 1. It sounds like, if I wanted to specialize in FGRP, I could set up that host to maximize its performance on them, but I like to keep it general. In fact, maybe I'll just set them both to -1.

Thanks for the info.

David

Miserable old git
Patiently waiting for the asteroid with my name on it.

mikey
mikey
Joined: 22 Jan 05
Posts: 12776
Credit: 1858902874
RAC: 1313751

RE: Okay. On the driver, I

Quote:

Okay. On the driver, I downloaded the newest one last week on the 440, and then they released another one this week, which I installed on the 630 but not the 440.

Do you also game, if not you may not want to upgrade to the latest and greatest gpu software right away. Prior to the latest one the previous 3 were not helpful to most crunchers, just gamers. Alot of crunchers say a 10% or more drop in their output after upgrading. We crunchers are NOT their target market so often have to 'make do' with what works. Alot of crunchers still use some very old versions of the gpu software for this very reason.

David S
David S
Joined: 6 Dec 05
Posts: 2473
Credit: 22936222
RAC: 0

RE: RE: Okay. On the

Quote:
Quote:

Okay. On the driver, I downloaded the newest one last week on the 440, and then they released another one this week, which I installed on the 630 but not the 440.

Do you also game, if not you may not want to upgrade to the latest and greatest gpu software right away. Prior to the latest one the previous 3 were not helpful to most crunchers, just gamers. Alot of crunchers say a 10% or more drop in their output after upgrading. We crunchers are NOT their target market so often have to 'make do' with what works. Alot of crunchers still use some very old versions of the gpu software for this very reason.


No, I don't game (other than MS Internet Backgammon occasionally). I think the 630 still had the 310.something driver. I wouldn't be surprised if the 440 did too. Do you recommend a particular rollback?

David

Miserable old git
Patiently waiting for the asteroid with my name on it.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.