Newish cruncher w/ Qs for milkyway

[MindCrime]

I just started crunching and joined the team almost a month ago and I'm enjoying the challenges to keep me more involved. I think I did pretty well on einstein@home for my hardware. I was able to manipulate the settings on the einstein page to get my gpus running good .33 BRP per gpu.

I'm on win 7 65 bit, My BOINC is not installed to the default path, it's installed to E:\Boinc

I see all these xml excerpts pasted in threads.... can someone please tell me the file name of these? I see account_, master_, sched_reply, statistics_, all of which don't look like the stuff i want

In boinc >data>projects>milkyway.cs... there are no xml files just the apps and .stars, .txt and .prmts

If i search for app_config i get localizations for non related apps.

Now that that is all said, maybe if i fix it up with better settings it'll resolve my issue of getting work. ATM i have the resource share set very high so if it gets work it'll run but i'm not getting much work from milkyway. I've issued a couple of updates and fiddled with what apps to run and not run to entice some traffic. Then I feel like i'll try my hand at bunkering, ready to firewall it but i figure i need work and setup right before i do that. Any and all info is appreciated.

Hardware:
GPU - gtx 560 currently stock (850mhz gpu, 2052mhz mem 1.025vddc) can go to 975mhz gpu core
GPU - intel hd4000 (not impressed with performance on other apps so far)
CPU - i5-3570k (4 logic cores) @ 4.2ghz

The apps in >boinc>data>projects>milkway.cs.rpi.edu_milkway are
milkway_separation_modified_fit_1.28_windows_x86_6 4_opencl_nvidia.exe
milkway_separation_1.00_windows_x86_64.exe
milkway_separation_1.02_windows_x86_64__opencl_nvi dia.exe

Currently the only milkway@home work I have is for 1.28 opencl_nvidia app and no more work after this.

[FourOh]

Unfortunately the GTX 560 is not going to be a great MilkyWay cruncher, but it's a good project and easy to test setting because of the short WU times. You need to create a text file with the name app_config.xml in your projects/milkyway folder with the following content:
<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.05</cpu_usage>
</gpu_versions>
</app>
</app_config>

From there, play with the gpu usage and cpu usage until you find your optimal setting. For testing, I would go to project settings and select MilkyWay@Home v1.02 only. I looked up your computer on the MW site, it looks like you have better production with that application anyway. As others have mentioned, you can drop GPU memory settings as low as they go on MW with no ill effects. Good luck and report back!

[MindCrime]

Alright this is easier than i thought. Now i think i have to wait for some work to finish up before i start seeing the .5 manifest into 2 concurrent gpu apps. Problem is, i only have 4 tasks. Is there a way to load up? Is milkyway not a good bunkering project for this reason and short task times?

[FourOh]

Save your app_config file, then close/restart your BOINC client. It will start crunching the way you have it set up immediately.

Anyone have thoughts on how to get a large cache of MilkyWay work units? It's something I haven't tried, and you are right - the project won't give you much to work with.

[MindCrime]

Alright that was easy. I figure i have to wait for some work to finish (or abort/suspend) to see the changes manifest to two concurrent GPU tasks. I was concerned I wasn't going to have enough work to keep the gpu saturated for a while on milkyway alone to get it dialed in but after understanding what you mean about my production:

Completed and validated 199.15 53.15 79.93 MilkyWay@Home v1.02 (opencl_nvidia) ~.4 credits per second
Completed and validated 1,008.58 589.04 213.76 Milkyway@Home Separation (Modified Fit) v1.28 (opencl_nvidia) ~.2 credits per second

I unchecked all the applications except "Run only the selected applications MilkyWay@Home" which i believe is the first of the two listed above. After which i got some new work but i still have a variety.

Now im getting flustered with all the different location settings and cpu settings. I have cpu work turned off for milkyway but i do have a couple cpu tasks for it.

What I want is boinc to use lower amounts of CPU globally so the GPU gets all the CPU it needs.

Right now i've got different settings in home,default its kind of pissing me off because I dont know if either are effective with my settings locally, other projects, other settings within this project....

[FourOh]

Once you deselect cpu tasks, MilkyWay should stop sending them after an update or two. However, MW CPU tasks pay decent credit and play nicely with the GPU app. Really, most CPU apps do ok with MW because MW uses so little CPU resources. Do you have a cc_config.xml file in your /BOINC folder? If you want to crunch Collatz on your HD4000 while your GPU works on MW, I think it would look something like this:
<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
<report_results_immediately>1</report_results_immediately>
<exclude_gpu>
<url>http://milkyway.cs.rpi.edu/</url>
<device_num>0</device_num>
<type>intel_gpu</type>
</exclude_gpu>
</options>
</cc_config>

Edit: I'm not even sure this is necessary, as long as you only have Intel GPU selected in Collatz.

[Al]

Milkyway uses very little cpu, so that is a non issue. There is no need to leave a core or multiple cores available for its use because it won't use them anyway. In your local preferences, set "on multi processor systems, use at most" to 99 percent. This will free up one core for MW gpu and that is all you will ever need, regardless of the amount of tasks you run concurrently. I'm currently testing nv cards on mw, so you can see and compare. To be blunt, nv really suck on MW, but it's our challenge, so I'll run them. The completion times you see are running 4 tasks concurrently, which is probably more than necessary, but I'm testing.

Look here....

[Mumps]

As far as bunkering, MilkyWay as a whole is designed in such a way that large numbers of outstanding WU's are the worst thing for the project. Each WU that's returned helps dictate what the next WU's to run need to do. So the project can't let you have a lot of WU's, because the more WU's out in the wild, the less good work they get accomplished. Sorry, but this is not a project you can (or should, even if you find a way) bunker on.

[zombie67]

Save your app_config file, then close/restart your BOINC client.

No need to close/restart your BOINC client. Just Advanced -> Read config files.

[MindCrime]

All of this is helpful, no i don't have a configure file but now that I understand the computing settings on each projects site are BOINC global settings I've fixed some of my issues. I think i was contradicting some settings between app_config and the settings for boinc on individual project sites. After defaulting all those I was just now able to restart boinc and get 3 concurrent tasks on the nvidia gpu. Computer is definitely more hesitatant, as i write this, definitely feels like more credit pouring out.

Good tip on the read config files.

And thanks for the info on the MW WUs, how they go in succession, glad i didn't block the IP.