BU WUs seem to fall of the GPU at the end

[Maxwell]

I just want to see if I'm nuts or something...

I'm running Bitcoin Utopia on a couple crappy ATI cards I have at work. It had been working fine for a while, but lately, it seems like the WUs are falling off the GPU. They'll crunch up to 90-95% really quickly, then the last 5-10% goes at the pace of a snail wading through molasses in January.

Is this normal? Should I be doing something differently? My credit output on this has dropped by around 80%...

[myshortpencil]

It may be a coincidence, but a daily restart on my Win7 7790 and 6870 machines practically eliminates hangs.

[Maxwell]

It may be a coincidence, but a daily restart on my Win7 7790 and 6870 machines practically eliminates hangs.
One of the machines, I restarted late yesterday and have the same issue...

[Bryan]

Mine just plod along right to the end ... no fast and then slow up. I've had no errors except on Cuda cards.

[Maxwell]

I looked at a laptop today, and it had a WU hanging out at 100% after running for 19+ hours, and still had yet to finish. I shook my fist at it, then did a project reset. We'll see how that goes.

[Mumps]

Did you see the recent thread about disabling Benchmarks? (http://www.setiusa.us/showthread.php?6411-skip_cpu_benchmarks-in-cc_config-cml&highlight=benchmarks) I wonder if that might help at all...

[Maxwell]

Did you see the recent thread about disabling Benchmarks? (http://www.setiusa.us/showthread.php?6411-skip_cpu_benchmarks-in-cc_config-cml&highlight=benchmarks) I wonder if that might help at all...
Thanks for that - didn't see it, but it certainly couldn't hurt. I added it to both machines.

[myshortpencil]

Did you see the recent thread about disabling Benchmarks? (http://www.setiusa.us/showthread.php?6411-skip_cpu_benchmarks-in-cc_config-cml&highlight=benchmarks) I wonder if that might help at all...

I have BOINC benchmarks disabled in my cc_config file and BU still hangs occasionally, though less with daily computer reboots :)

[Maxwell]

Oy. Stupid Work Units.

After an afternoon of futzing, I've figured out that the portion of the CPU that's being reserved for the Bitcoin work unit is not being obeyed by some of the other projects I'm running. The other projects (Rioja, Beal, CAS) taking up all the CPU, and leaving nothing in reserve to feed the GPU.

So: is there an app_info file out there that will let me reserve an entire core to feed the GPU? I don't want to have to give up CPU crunching on this laptop if I can at all help it...

[zombie67]

You need app_config.xml, not app_info.xml. Much easier.

See the bottom of this page:

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

[Bryan]

Put a file (app_config.xml) in your BU project folder that contains this:

<app_config>
<app>
<name>lineB-sgminer-4.1.153-v2</name>
<max_concurrent>2</max_concurrent>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
</app_config>

Then go into BOINC and under the ADVANCED tab tell it to read config files.

The BOINC scheduler will reserve 1 thread for your GPU.

[Maxwell]

Damn.

Thank you, gentlemen, very much. This is what happens when you're away for a while - things change, and you don't realize it. This is so much easier than an app_info file. Yay!

[Bryan]

Damn.

Thank you, gentlemen, very much. This is what happens when you're away for a while - things change, and you don't realize it. This is so much easier than an app_info file. Yay!

Obviously you've been away for a long time and need retraining .... you referred to me as a gentleman =))

[Maxwell]

Obviously you've been away for a long time and need retraining .... you referred to me as a gentleman =))
My bad - I should have known better... :p

Speaking of which, is there any benefit to running two BU WUs at a time? Or is sticking with one the way to go?

[Bryan]

My bad - I should have known better... :p

Speaking of which, is there any benefit to running two BU WUs at a time? Or is sticking with one the way to go?

I tried it and it made no difference. That is why the app_config I gave you says max_concurrent of 2. If you are only going to run 1 WU you can change that to "1". If you decide to try running 2 you don't have to give it more CPU. To run 2 change the GPU AND CPU usage to .5.

[Mumps]

I was wondering why he quietly included that in his post. Keep in mind, if the WU's run fast enough, you waste more time waiting to get the next one started than it takes to run one in the first place. So for projects like this with sometimes short run-times, it can be a big improvement in RAC by always having one on the GPU... Best thing would be to monitor for a while and see if you get better RAC than just running a single one. My bet is, for BU, you will. That probably will also lead to less lost cycles if one of the WU's hangs, but the other one doesn't.

[Bryan]

They've raised the cache per machine up to 40 WU now. I guess I've been lucky since I haven't had any WU hangup yet ... knock on wood. I did have all 3 7970s go into 24 hour backoff mode sometime last night and they had run dry. I guess the project was down for a while.

[Maxwell]

Thanks, guys (that better, Bryan?)!

I've set it up to run two on the desktop, and one on laptop. We'll see how they do. And Mumps, that was my exact thinking, so I'm glad you said that.

[STE\/E]

Are you guy's leaving 1 Thread open for BU, I hardly ever have a Wu Hang on 3 7970's ...

[Maxwell]

Are you guy's leaving 1 Thread open for BU, I hardly ever have a Wu Hang on 3 7970's ...
I am, yes. I've noticed this becoming much more of a problem with some of the "newer" projects (Beal, Rioja), since they don't seem to want to let the CPU feed the GPU. Things have been running much more smoothly since I started leaving a thread open, though that's only been a few hours now...

[zombie67]

FWIW, I use .5 GPU and .25 CPU with my dual 7970 machines. That means running 4 at a time, and reserving a full CPU thread for them all.