climateprediction.net home page
Optimise PC build for CPDN
Optimise PC build for CPDN
log in

Advanced search

Questions and Answers : Windows : Optimise PC build for CPDN

1 · 2 · 3 · 4 · Next
Author Message
MartinNZ
Send message
Joined: 22 Mar 06
Posts: 144
Credit: 24,695,428
RAC: 0
Message 46271 - Posted: 23 May 2013, 9:39:24 UTC

Profile MikeMarsUK
Volunteer moderator
Avatar
Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,546,884
RAC: 10,274
Message 46274 - Posted: 23 May 2013, 11:55:39 UTC
Last modified: 23 May 2013, 12:02:11 UTC

...but whether or not they would last the CPDN read/write throughput is another question. The arrangement would cut down on the huge read/write operations to the OS SSD, but Page File is still likely to be hitting well over 1MB/s on the SSD. I could of course move that to the CPDN or other hard drive.


Yeah, in my opinion don't put the CPDN data drive onto a SSD disk (unless it is SLC / enterprise quality).


How much memory do you have? It's pretty cheap at the moment. If you make sure your upgraded PC has loads, then it shouldn't hit the page file. I don't think there will be any particular advantage in having the CPDN executables on SSD because they are only loaded once (although on my system, the Boinc executables area is indeed cached by SSD).


I was pondering the setup of my own new PC recently in this thread, could that have been the one you were thinking of?
http://climateapps2.oerc.ox.ac.uk/cpdnboinc/forum_thread.php?id=7598

Having struggled with setting up a 3TB HD recently, I would advise sticking to one or two 2TBs instead!! :-) I did get it going eventually but it was unnecessarily difficult.

Depending on your budget you might want to consider a Socket 2011 setup, that allows loads of memory and cores. But it is expensive.

The bottleneck on mine seems to be after 6 models, for 7 and 8 the throughput levels out. With a hetrogenous project mix it scales better.
____________
I'm a volunteer and my views are my own.
News and Announcements and FAQ

ojum-le
Send message
Joined: 5 May 07
Posts: 27
Credit: 6,369,307
RAC: 0
Message 46276 - Posted: 23 May 2013, 20:18:58 UTC

Profile MikeMarsUK
Volunteer moderator
Avatar
Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,546,884
RAC: 10,274
Message 46278 - Posted: 23 May 2013, 21:35:06 UTC - in response to Message 46276.
Last modified: 23 May 2013, 21:58:51 UTC

Les Bayliss
Volunteer moderator
Send message
Joined: 5 Sep 04
Posts: 6909
Credit: 20,843,205
RAC: 108
Message 46279 - Posted: 23 May 2013, 22:07:46 UTC

I'm not sure if it applies here, but BOINC 7.0.25 was still a test version. 7.0.28 is stable.
Might be worth an upgrade.

But I don't like the look of the latest release. Lots of problems being talked about.


____________
Backups: Here

Profile MikeMarsUK
Volunteer moderator
Avatar
Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,546,884
RAC: 10,274
Message 46281 - Posted: 23 May 2013, 22:27:10 UTC
Last modified: 23 May 2013, 22:29:17 UTC



I'm not sure if it applies here, but BOINC 7.0.25 was still a test version. 7.0.28 is stable.
Might be worth an upgrade.

But I don't like the look of the latest release. Lots of problems being talked about.



I'm using 7.0.64 now, it came out about a month ago. Haven't had any issues myself, but experience shows that new Boinc versions are often problematic...

http://boinc.berkeley.edu/download_all.php


The only example of that 'End-of-central-directory signature not found' error that I could find was back in 2005, I don't think it was solved then from the look of it.
____________
I'm a volunteer and my views are my own.
News and Announcements and FAQ

Profile Greg van Paassen
Send message
Joined: 17 Nov 07
Posts: 142
Credit: 4,271,370
RAC: 0
Message 46282 - Posted: 23 May 2013, 22:29:47 UTC

I'd be interested in the results, Martin, especially if you go Haswell - I believe they're being released in September?

An 8% improvement in FP performance (over Ivy Bridge) means finishing a HadCM3N a day sooner. The improved memory controller in Haswell should further speed up climate models, which do a lot of memory writes. You get better performance per watt -- always a consideration with NZ's electricity prices. And you may be able to put a Broadwell processor in the same board later, if they turn out any good. (Not that I'm trying to sell you on anything...) :)

Disks: as you say, CPDN isn't particularly demanding in terms of disk performance (if you have enough memory for OS disk buffers), but durability is a different story. I measured some HadCM3Ns at 700MB of disk writes over 10 minutes, each. I'd been considering buying a consumer grade SSD before that. Not any more. (According to Anandtech, consumer SSDs are designed for ten years of life assuming a mere 10 GB of writes per day.)

But, definitely put the Photoshop program files and light table (working) storage on SSD(s). Others have used phrases like "night and day" after making that change.

Les Bayliss
Volunteer moderator
Send message
Joined: 5 Sep 04
Posts: 6909
Credit: 20,843,205
RAC: 108
Message 46284 - Posted: 23 May 2013, 22:51:59 UTC - in response to Message 46282.

According to this, Haswell is out on the 4th of June.
Be a while before prices come down a bit.



____________
Backups: Here

ojum-le
Send message
Joined: 5 May 07
Posts: 27
Credit: 6,369,307
RAC: 0
Message 46287 - Posted: 25 May 2013, 10:44:08 UTC
Last modified: 25 May 2013, 10:45:23 UTC

http://climateapps2.oerc.ox.ac.uk/cpdnboinc/result.php?resultid=15736358

<core_client_version>7.0.25</core_client_version>
<![CDATA[
<message>
Das Laufwerk kann einen bestimmten Bereich oder eine bestimmte Spur nicht finden. (0x19) - exit code 25 (0x19)
</message>
<stderr_txt>
[hadcm3n_data_6.07_windows_intelx86.zip]
End-of-central-directory signature not found. Either this file is not
a zipfile, or it constitutes one disk of a multi-part archive. In the
latter case the central directory and zipfile comment will be found on
the last disk(s) of this archive.
unzip: cannot find zipfile directory in hadcm3n_data_6.07_windows_intelx86.zip,
and cannot find hadcm3n_data_6.07_windows_intelx86.zip.zip, period.
Called boinc_finish

</stderr_txt>
]]>


My RAID dropped on this day. One if the 15k.7 had caused many struggle and the system crashed. But now it works correct. I think there were lost some files who was in SAS-Controller-RAM.
But actually i dont get new work :-(

MartinNZ
Send message
Joined: 22 Mar 06
Posts: 144
Credit: 24,695,428
RAC: 0
Message 46288 - Posted: 26 May 2013, 0:40:00 UTC - in response to Message 46284.

Les Bayliss
Volunteer moderator
Send message
Joined: 5 Sep 04
Posts: 6909
Credit: 20,843,205
RAC: 108
Message 46291 - Posted: 26 May 2013, 2:29:42 UTC - in response to Message 46288.

My, what a lot of replies.
Life keeps getting in the way of my new machine, or I'd add a bit. :(

2a) I'd set NNT, and run down all work.
There's an "interesting" little BOINC "feature" whereby, if you transfer a model to a much faster computer, they fail due to "something-which-is-mentioned-on-the-old-PHP-board-but-I've-forgotten-what". (They run out of "something" time.)

As for BOINC version, the problems with the latest may just be people who like to complicate things by running lots of GPU apps on a variety of cards and computers and wonder why it won't work.
I've used 7.0.28 OK, but only on POGS SkyNet.

4) When an SSD fails, that's it for ALL of the data. At least with an HD there's a chance that some can be recovered. So it depends on what goes on to it.
Cache would be good. The OS perhaps, if it's not too customised, and time consuming to redo after a failure.

The old PHP forum.
I don't know how "political" that is, but unofficially, don't expect to see it ever again. Security risk for Oxford Uni.



____________
Backups: Here

Profile MikeMarsUK
Volunteer moderator
Avatar
Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,546,884
RAC: 10,274
Message 46292 - Posted: 26 May 2013, 11:32:40 UTC - in response to Message 46288.
Last modified: 26 May 2013, 11:37:09 UTC

MartinNZ
Send message
Joined: 22 Mar 06
Posts: 144
Credit: 24,695,428
RAC: 0
Message 46316 - Posted: 30 May 2013, 4:52:08 UTC - in response to Message 46292.

Les Bayliss
Volunteer moderator
Send message
Joined: 5 Sep 04
Posts: 6909
Credit: 20,843,205
RAC: 108
Message 46318 - Posted: 30 May 2013, 7:54:35 UTC - in response to Message 46316.

The News and Announcements thread here, is at the top of Number crunching.

Re: SSD
See this post at SkyNet POGs, last paragraph.

Hope that I didn't "put the mocker's on it".


____________
Backups: Here

miriquidi
Send message
Joined: 11 Aug 11
Posts: 1
Credit: 595,449
RAC: 0
Message 46402 - Posted: 11 Jun 2013, 19:32:31 UTC

Regarding the question, which is the right CPU for a CPDN-computer - read this one.
I'm still searching for some people having an AMD Phenom II (with L3-Cache), Bobcat, Jaguar and Bulldozer, this are the ones missing:
http://foveon.de/sonstiges/cpdn/

MartinNZ
Send message
Joined: 22 Mar 06
Posts: 144
Credit: 24,695,428
RAC: 0
Message 46403 - Posted: 11 Jun 2013, 23:54:58 UTC - in response to Message 46402.

Thanks for the replies everyone, I've decided to build a new PC! I will probably go Xeon LGA 2011, but a question on memory first.

To ECC or not ECC, that is the question? Speaking to my supplier, ECC will drop the memory speed by around 2 cycles, or around 10-15%. ECC is not needed for my work, but I assume would be beneficial for CPDN. What are your opinions on this?

Currently looking at a rebuild using Asus P9X79 WS motherboard, or HP Z420 or Dell T7600. Will either have the Xeon E5-1650 or E5-2667 and allow CPDN to run on half the cores.
____________

Profile Greg van Paassen
Send message
Joined: 17 Nov 07
Posts: 142
Credit: 4,271,370
RAC: 0
Message 46404 - Posted: 12 Jun 2013, 5:57:10 UTC - in response to Message 46403.

Tough call.

The delay arises from buffering rather than ECC itself. For sequential accesses (as I'd expect with both CPDN models and visual image processing), buffering only delays the first bytes in the read or write request. After that throughput is the same for the rest of that memory access.

If your (proposed) motherboard can take non-ECC memory, it may require unbuffered memory. Unbuffered ECC memory does exist but I'm not sure how easy it is to buy.

My gut feel, from reading Wikipedia's ECC page and other pages about RAM, is that ECC would be worth a price premium of 20% or so to me.

Eirik Redd
Send message
Joined: 31 Aug 04
Posts: 363
Credit: 111,928,424
RAC: 173,293
Message 46405 - Posted: 12 Jun 2013, 10:48:59 UTC
Last modified: 12 Jun 2013, 10:59:44 UTC

If/when I buy another CPDN box -- yes try the ECC - few are using it, good to have a comparison -- think I'll go with that.
About speed - my ancient Core2 zinnia 2-core at 3GHz (on Linux) (see my machines) is still the fastest per/core. Not all that fast - the Linux optimizations were never that good, and the Darwin compiles on Rapid-rapit run really really fast.
Any machine you build will be subject to the chosen optimizations that the site finally settles on -- like right now Darwin runs fastest (but maybe less accurately) Windows is slower, and Linux is slower still.
If the CPDN site ever rebuilds software for models, this may change (not liklely soon) --
As far as I know, nobody is building new executables at CPDN -- all about plugging new data into the old stable exe.

Also note that what the site reports as RAC is about half your actual contribution -- if your Recent Average Credit on Rapid-rabbit is 2000 per the site, you are actually contributing about 4111 credit/day - just an idiosyncrasy.
Go for high GHz -- seems to be a constant productivity enhancer- forget the caches, (unless l1 and l2 get bigger -- ha ha)
Forget disk speed altogether - the fastest SSSD won't help with CPDN.
____________

MartinNZ
Send message
Joined: 22 Mar 06
Posts: 144
Credit: 24,695,428
RAC: 0
Message 46406 - Posted: 12 Jun 2013, 11:04:24 UTC - in response to Message 46404.
Last modified: 12 Jun 2013, 11:19:40 UTC

Thanks Greg, interesting, but your second reference was waaaaay to technical for me. My main thought was, ignoring the speed issue for a moment, could the use of ECC memory be of beneficial effect to the completion of the CPDN tasks, or is it insignificant compared to all the other things that could go wrong.

There seem to be a lot of task failures in the last 12 months, with my existing PC (which has just finished all its takss) being no exception. Some are model failures, but from the posts I've read, a lot come up with failures that are either the fault of the PC somewhere, or undefined.

Yes there is a price premium on ECC ram, here about $110/8GB cf $90/8GB, but that is a relatively small cost in a rebuild. Both non-ECC and ECC seem to be available in unbuffered (which is what the motheboard requires.)

Just noticed Eirik' post came in so will edit this one.

Figured the high GHz and disc aspects, so thanks for the confirmation. I will be putting BOINC/CPDN on it's own HDD as I've a few spare around. Any thoughts on whether that should be both the program and data, or is the program OK on the OS disc with just the data on a separate HDD?

Interesting on the ECC. Hopefully some more contributions will come in as well.

The processor E5-1650 runs at 3.1GHz which is 'reasonable' value of money, but I've never really been concerned with RACs (even though it is interesting to see where things are at!)
____________

Eirik Redd
Send message
Joined: 31 Aug 04
Posts: 363
Credit: 111,928,424
RAC: 173,293
Message 46407 - Posted: 12 Jun 2013, 11:24:44 UTC

I would, and will, next rebuild, go for the ECC. But a UPC first -- me already got - it helps.

Also, as advised all over this board - no automatic updates. Exclude BOINC from antivirus.

Don't let your OS reboot your box on it's own -- get a clean shutdown before every reboot if you possibly can. Pulling the dll's out from under will definitely wreck your CPDN models.
____________

1 · 2 · 3 · 4 · Next

Questions and Answers : Windows : Optimise PC build for CPDN


Main page · Your account · Message boards


Copyright © 2019 climateprediction.net