[TBR] Viva La Roast! (Perpetual Thread V)

[url=http://arstechnica.com/civis/viewtopic.php?p=27936437#p27936437:3lsxeiu9 said:
TheJet[/url]":3lsxeiu9]w00t! Finally broke the 5M GNode mark, thought I might be able to catch up to aEric, but he's proving just too much for my existing computing power to handle. :)
Oh, it's not so much, i5 and an ancient thinkpad, and usually run climateprediction.net but got a bit tired of their server struggles :devious: :eng101: :confused:

Seems we lost 3 places this week :/

aEric
 
[url=http://arstechnica.com/civis/viewtopic.php?p=28261687#p28261687:tu0cw9kd said:
EG[/url]":tu0cw9kd]Question, what does it take to run this on an AMD box with my hardware.

Never having run this before, how do I get it to run on my flock of AMD GPU's?

Their website doesn't explain it very well and is, well, rather dated.

What do I need to know brother Arsians?

For AMD GPU's you have the option of running the stream client which is more native to AMD or the opencl client which is more generic. I have several different versions of AMD GPU's running and it has been a hit and miss on which client works. Some GPU's will run both and some will only run one or the other. The setup of the GPU client is exactly the same as the CPU client. Make sure you have the dnetc.ini file with your user info in it. Verify you are running the latest drivers for your GPU before trying either of the GPU clients. If it still doesn't work give us some details on the system configuration you are trying to get running.
 

EG

Ars Scholae Palatinae
626
[url=http://arstechnica.com/civis/viewtopic.php?p=28263183#p28263183:t8ncveig said:
guru[/url]":t8ncveig]Add the following two lines to your dnetc.ini file. The "=0" tells the client not to run that project.

[misc]
project-priority=RC5-72,OGR-NG=0

Thank you!

Think i've got it running now. Should I leave a CPU core open for each GPU? or does it even matter?
 
Well it was good but short run for me. My hope of getting rc5 to run on a phi has failed. Hopefully after the second generation is released there will be more support for DC projects on it. My mini hord of GPU's will stay on rc5 but the CPU clients will revert back to OGR with a few dedicated to SOB domination. The nail in my rc5 coffin came from EG. Those are some nice numbers! Keep it up and rc5 will be over soon.
 
[url=http://arstechnica.com/civis/viewtopic.php?p=28262683#p28262683:1mj6h57s said:
EG[/url]":1mj6h57s]I think I've figured it out.
Two different clients one for CPU's and one for GPU's. Wait for 24 hours for your first stats run to complete and request your DB password then you can join the team.

Just one question remains, how do you configure the CPU client to get just RC-72 work?
Now that you have your Distributed.net ID: arguably the best way to run RC5-72 is to run Moo on BOINC. Just run Moo like any other BOINC project, tell it to use AMD GPUs only and set your Distributed.net ID in the project preferences. That way you'll get Moo BOINC credits for Ars and Distributed.net credits for Ars :)
 

EG

Ars Scholae Palatinae
626
[url=http://arstechnica.com/civis/viewtopic.php?p=28300369#p28300369:21z2d6gu said:
Beyond[/url]":21z2d6gu]
Now that you have your Distributed.net ID: arguably the best way to run RC5-72 is to run Moo on BOINC. Just run Moo like any other BOINC project, tell it to use AMD GPUs only and set your Distributed.net ID in the project preferences. That way you'll get Moo BOINC credits for Ars and Distributed.net credits for Ars :)

In other words I get credit on both sets of pages?

Running Boinc is easier for me than the non boinc client.
 
In other words I get credit on both sets of pages?
Running Boinc is easier for me than the non boinc client.
Yep, Moo is just a BOINC wrapper for RC5-72. He's set it up so the credits also transfer to your distributed.net account. All you have to do is enter your dnet ID in the project preferences on your Moo account page. Also select ATI GPU only as it's by far the most efficient way to runch RC5-72.
 

EG

Ars Scholae Palatinae
626
[url=http://arstechnica.com/civis/viewtopic.php?p=28308105#p28308105:3qdt85v2 said:
Beyond[/url]":3qdt85v2]
In other words I get credit on both sets of pages?
Running Boinc is easier for me than the non boinc client.
Yep, Moo is just a BOINC wrapper for RC5-72. He's set it up so the credits also transfer to your distributed.net account. All you have to do is enter your dnet ID in the project preferences on your Moo account page. Also select ATI GPU only as it's by far the most efficient way to runch RC5-72.

Got them all switched over, still doing some tweaking for efficiency.

but it's beginning to shake out...

:)
 

EG

Ars Scholae Palatinae
626
[url=http://arstechnica.com/civis/viewtopic.php?p=28310669#p28310669:1kbkew93 said:
Beyond[/url]":1kbkew93]
What kind of tweaking are you doing?

Well I do have 9 7990's and 5 7970's and like on Collatz I have most of them running. The issue is what is fast. When I was running the original client I couldn't run them all cause it uses 100% of the GPU and without any way to manipulate the client you just can't run them all given the heat issues of running Tahiti's at 100%

Under Boinc I have much more control of the client and Overdrive gives me complete control of the GPU's to handle the heat issue.

Looking at the top producers boxen I saw that they were turning WU's at 1500 seconds (25 minutes per) After tuning with Overdrive, I've got mine turning 1200+- seconds per for the most part and I have 8 90's and 2 70's running. that's 18 Tahitis. All running under 85 degrees at 97-99% usage.

Makes for a strong impression.

After a bit more config I should have 20 running.
 
[url=http://arstechnica.com/civis/viewtopic.php?p=28327671#p28327671:2uapjz6g said:
guru[/url]":2uapjz6g]Are you trying to run them all in one system or one per system? I've found it is best to run one card per system if you can. Are you trying the opencl client or the stream? Each card should do ~2100Mk/s for a total of ~100K blocks per day(total of 3 cards).
Used the opencl client, and all in one system (trifire), i had a bit under 100K so not bad.