Welcome to MilkyWay@home

Posts by Xenu

1) Message boards : Number crunching : Run Multiple WU's on Your GPU (Message 58510)
Posted 5 Jun 2013 by Xenu
Post:
Im running at 1100 Mhz Currently , takes 1min 20 to do 6 WU ( 3 on each card )



My results are similar to this, running a 7870xt (7950 rebranded by the marketing department). I can run one in 42 seconds, or two in 83 seconds, a speed increase of less than 2%. Adding a third gives no increase at all.

At first, I thought this might be a method that only works on Nvidias, since this has been how it's worked for me on both the old and new clients, with a couple of different mobos and a few different AMD CPUs. (Current CPU is an X6 1045t.) I do leave spare cores.

Does anyone have any suggestions as to why it seems to work amazingly for some people, but barely, or not at all, for others with similar hardware?
2) Message boards : Number crunching : Run Multiple WU's on Your GPU (Message 57755)
Posted 2 Apr 2013 by Xenu
Post:
I have been trying to run the app_config file shown above and whenever I do the Boinc event log shows an error message saying "App file not found". Has anyone else had this problem?


Yes, and haven't found the answer yet. Did try the ideas mentioned above.

edit: Mine wanted <use_all_gpus>1</use_all_gpus> in the cc_config, apparently.
3) Message boards : Number crunching : Run Multiple WU's on Your GPU (Message 57581)
Posted 20 Mar 2013 by Xenu
Post:
Just curious what card you're doing this with. I'm running single MW WUs on a Radeon HD 6950. WUs take about 90 seconds at 98% GPU load and approx 10% GPU memory. I'd like to try multiple WUs at a time, but it doesn't look like I have much extra room.


On a 7950, running two at once only netted me a gain of roughly 7% in speed, but similar methods can be used to make a MW job combine well with other jobs/projects. I'm running two fast AMD cards with an AMD hexacore CPU, and find that if I want to run POEM on all of it, my system becomes totally CPU bound, and half of my cards' capacity go unutilized.

On the other hand, using default settings for something like MW or donate@home will get close to keeping my cards busy, while my CPU twiddles its thumbs.

Something like 3 POEM jobs and 1 MW (or other GPU-intensive, CPU-light job) on each card uses up everything very neatly & efficiently, and neither project's jobs are slowed much if you find the right balance between them.
4) Message boards : Number crunching : Hoping for config help: disparate cards (Message 57554)
Posted 19 Mar 2013 by Xenu
Post:
That works, thanks very much!
5) Message boards : Number crunching : Hoping for config help: disparate cards (Message 57539)
Posted 18 Mar 2013 by Xenu
Post:
I'm having problems running MW because my AMD cards are unalike, and am hoping for some sort of fix or workaround.

Card 0 runs both single and double precision very fast. Card 1 runs great single precision, but does double precision at a glacial pace, less than 1/7 as fast as card 0. If card 1 is running MW, it's being wasted, I'd like to restrict it to one MW job on card 0.

I tried inserting code like this into my app_info.xml:
<max_concurrent>1</max_concurrent>
<coproc>
<type>ATI</type>
<count>0.4</count>
<device_nums>0</device_nums>
</coproc>

This failed to either limit it to one MW job at a time, or to keep MW restricted to card 0; it ran multiple copies on both.

Does anyone know how to get it to do what I need? It seems like it'd be a relatively common problem, but I'm not finding much.




©2022 Astroinformatics Group