Ideal system


log in

Advanced search

Questions and Answers : Getting started : Ideal system

Author Message
thechurchofdave
Send message
Joined: 24 Jul 10
Posts: 2
Credit: 0
RAC: 0
United States
Message 1019363 - Posted: 25 Jul 2010, 2:10:42 UTC

I am ultra new to this and after looking around the site for a few hours and not finding the answers I wanted I thought I would just ask.

I opened an account on a nothing special computer but some friends and I have been talking about building a few fast computers and dedicating them to distributed computing projects.

My question is this. What are the most important factors to consider when building a computer which is ONLY going to sit there crunching 24/7 ?

If you had some spare cash layin around and wanted to build (or buy) a new computer, which would be used strictly for doing this sorta thing, what kind of setup would you run?

Questions which come to mind.
1. Rackmount server of Pc?
2. Multiple processors or multiple GPU's or both?
3. Can I use super fast CPUs and not bother with
fancy GPUs"
4. What O.S.? I like linux since it's free. Is that
a problem or a plus? Will windows save me hours of
admin though?
5. How much Ram?
6. Fast, small hard drives OK?
7. Best way to set up multiple machines?
8. Anything else I am missing?


I am, of course, interested in a setup which gets as much done as possible without wasting money or sacrificing reliability.

I am sorry if this has all been addressed before but I was unable to find it.

Thank you
DAve

thechurchofdave
Send message
Joined: 24 Jul 10
Posts: 2
Credit: 0
RAC: 0
United States
Message 1019489 - Posted: 25 Jul 2010, 15:49:00 UTC - in response to Message 1019363.

So after reading a bunch of stuff on the nvidea site, I see that using multiple core GPU's seems to do this type of work more effectively than fast CPU's and with less power consumption.

So I am guessing that I should be looking at a tower which I can fill with GPU's or maybe something like the nvidia rackmount computing clusters. Since I am expecting to have a few bucks at my disposal, I would rather do something which is rackmount so I can expand, keep things tidy and easily cooled.

If I am using a tower with GPU's does the processor do much more than tell the GPU's what to do? Am I better off to use economy CPU's and spend on the GPU's? Can I get used rackmount servers and pile GPU's into them?

What about O.S. and the amount of memory required? I am guessing a small hard drive with speeds over 7200 would do just fine.

Any input is greatly appreciated.

Meanwhile Boinc is running 24/7 on this system now (unless I am watching a video) and is about 50 some hours away from completion of the first project. Easy install, no problems thus far.

About ten year ago my now ex-wife an I had a bunch of computers which were all hooked to SETI for about 4 years without any problems but I didn't have much to do with it. She was/is an IT person and took care of all that stuff on her own. My background is mainly film. So I am just wanting to get back into this again but I would really rather avoid calling and asking her for help. I am sure some of you guys can appreciate the right?

Sincerely
DAve

____________
I decided to give up quitting since failure is not only an option but a wonderful learning tool.

John McLeod VII
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 15 Jul 99
Posts: 24394
Credit: 519,750
RAC: 26
United States
Message 1019833 - Posted: 26 Jul 2010, 22:02:14 UTC

The CPUs can also do work on their own as SETI uses < 0.01 CPU / GPU task. So the CPUs can get their own tasks to work on.
____________


BOINC WIKI

Profile ignorance is no excuse
Avatar
Send message
Joined: 4 Oct 00
Posts: 9529
Credit: 44,433,274
RAC: 0
Korea, North
Message 1020100 - Posted: 27 Jul 2010, 14:13:06 UTC - in response to Message 1019833.

only a couple Projects run GPU tasks. So you could run multiple task on those Projects and dedicate your CPU to other Projects like Seti.

Also note: ATI cards are limited on Seti to Hybrid Astropulse tasks using the lunatics app. Nvidia CUDA cards will run the multibeam app on Seti
Einstein also has a CUDA apps for their WU's

the other projects that use GPU in no particular order are DNETC, Milkyway, and Collatz. Milkyway requires double precision GPU ATI 5970,5870, 5850, 4770 and a few others. Check the manunacturers data to be certain. I am not certain what the minimal CUDA card is.
____________
In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

End terrorism by building a school

Profile Ageless
Avatar
Send message
Joined: 9 Jun 99
Posts: 12288
Credit: 2,576,849
RAC: 818
Netherlands
Message 1020134 - Posted: 27 Jul 2010, 15:27:20 UTC - in response to Message 1020100.

ATI cards are limited on Seti to Hybrid Astropulse tasks using the lunatics app.

If you're giving 'advice' like that, please explain what you mean with "Hybrid", and point out what a "lunatics app" is. New people aren't as knowledgeable as you or I are.

Now, I am not going to explain what a Lunatics application is, I'll leave that to the next person coming along.

Nvidia CUDA cards will run the multibeam app on Seti

But not all Nvidia GPUs. Minimum required is 256MB of VRAM.
If running stock supplied by Seti application, Windows XP or above is required.
Nvidia drivers 185.85 or higher are required for use with BOINC 6.10

Einstein also has a CUDA apps for their WU's

Not entirely true. The Arecibo Binary Pulsar Search tasks are the same "Hybrid" breed as the Astropulse tasks, where only the FFT (Fast Fourier Transform) algorithm runs on the GPU, all the rest of the calculations run on the CPU. When running Einstein ABP2 tasks on a GPU, they require an Nvidia GPU with 512MB VRAM or more. ABP2 can also be run on the CPU.

Milkyway requires double precision GPU ATI 5970,5870, 5850, 4770 and a few others.

What do you call 'a few'?
Radeon HD3690, HD3830, HD3850, HD3870, HD3870 X2, (HD4770), HD4830, HD4850, HD4850 X2, HD4870, HD 4870 X2, HD4890, HD5830, (HD 5850, HD5870, HD5970), Mobility X800, X800 XT, X1800, X1800 XT, X1900, FireGL V7700, FirePro V8700, Firestream 9170, 9250, 9270. Phew. I wouldn't call that 'a few'. Choice enough!

I am not certain what the minimal CUDA card is.

For Double Precision, a GT200 or GT200b chip is required. So that makes GTX260, GTX275, GTX285 and GTX295 for consumer cards, the Tesla C1060, S1070 and S1075. The GTX4xx's have DP but in disabled form. If you want the full version, get a Tesla, is what Nvidia says.
____________
Jord

Fighting for the correct use of the apostrophe, together with Weird Al Yankovic

Questions and Answers : Getting started : Ideal system

Copyright © 2014 University of California