climateprediction.net home page
Support for BOINC - x64 version

Support for BOINC - x64 version

Questions and Answers : Wish list : Support for BOINC - x64 version
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
old_user466513

Send message
Joined: 14 Aug 07
Posts: 1
Credit: 82,092
RAC: 0
Message 30475 - Posted: 9 Sep 2007, 22:20:33 UTC

I would like to see support for BOINC - x64 version cause i\'m using 64bit windows and a 32bit Boinc client. How long will this take to become available to us all?
ID: 30475 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 30476 - Posted: 9 Sep 2007, 22:31:02 UTC


Basically: Sometime in the future.

There\'s a thread about this here.
And another in Number crunching, I think.


Backups: Here
ID: 30476 · Report as offensive     Reply Quote
jrenals

Send message
Joined: 25 Feb 05
Posts: 2
Credit: 12,694,238
RAC: 0
Message 30769 - Posted: 30 Sep 2007, 9:50:49 UTC - in response to Message 30475.  

For what it\'s worth so would I. I\'ve been using x64 PCs for a couple of years now and it seems a waste not to use this facility.


ID: 30769 · Report as offensive     Reply Quote
Profile MikeMarsUK
Volunteer moderator
Avatar

Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,613,038
RAC: 0
Message 30820 - Posted: 5 Oct 2007, 17:28:06 UTC


The model won\'t actually be any quicker when run via 64-bit Boinc, so assuming you\'re only running this project it won\'t make any difference. There are very few projects which have faster code in 64-bit (APS and primegrid come to mind), but most don\'t.
I'm a volunteer and my views are my own.
News and Announcements and FAQ
ID: 30820 · Report as offensive     Reply Quote
old_user2139

Send message
Joined: 27 Aug 04
Posts: 2
Credit: 94,104
RAC: 0
Message 30860 - Posted: 7 Oct 2007, 18:38:38 UTC
Last modified: 7 Oct 2007, 18:40:16 UTC

The model won\'t actually be any quicker when run via 64-bit Boinc, so assuming you\'re only running this project it won\'t make any difference. There are very few projects which have faster code in 64-bit (APS and primegrid come to mind), but most don\'t.

I would hazard to say that those of us with 64-bit BOINC aren\'t too bothered about whether 64-bit will be faster or not - what we\'re bothered about is this project NOT supporting us, after we have supported this project for a long time when on 32-bit systems.

It\'s a sad state of affairs when a science-based project won\'t (for whatever reason) upgrade their back-end systems to support people willing to donate their CPU time.

I for one am suspending any work here, until the project shows that they do want my effort - other projects will get my time instead, those that have bothered to upgrade & can pass out the 32-bit client to 64-bit systems - or better, have a 64-bit client already (although that is more a nice to have for me).

Mike

<edit - typos>
ID: 30860 · Report as offensive     Reply Quote
Profile mo.v
Volunteer moderator
Avatar

Send message
Joined: 29 Sep 04
Posts: 2363
Credit: 14,611,758
RAC: 0
Message 30862 - Posted: 7 Oct 2007, 21:00:05 UTC
Last modified: 7 Oct 2007, 21:11:44 UTC

We\'re all aware that a 64-bit version is impatiently awaited. We\'re also aware that CPDN is not in the vanguard among boinc projects in this respect. ABC@Home, for example, are getting their version out of beta.

The moderators have had a thread about this for some time in their own forum; the question is in no way being sidelined. Everyone at CPDN thinks it\'s a perfectly understandable request.

I cannot, however, accept the suggestion that anyone at CPDN (which in this respect means the programmers) \'won\'t\' \'bother\' to \'upgrade\' or \'support\' all its loyal crunchers.

Until recently CPDN had three programmers in Oxford - Carl, Tolu and Milo. In the News thread (top of the Number crunching section) you\'ll see that Carl recently left. We\'ve had no indication that this third post will be advertised.

The distributed computing side of CPDN used to have some administrative support. This is no longer the case.

On a Skype boinc phone-in late last year which I was lucky to take part in, Dr David Anderson mentioned that CPDN was one of the less well-funded boinc projects. He could have added both in terms of server hardware and manpower.

Tolu and Milo are constantly working to deadlines, usually but not exclusively for the development and testing of new versions. They keep 3½ projects on the road (CPDN + beta, BBC and SAP). Their priorities are necessarily to keep the servers fully-functional, to meet model development deadlines required by the researchers, and to debug existing versions.

I have never doubted their dedication to both researchers and crunchers. I hope they will find time to implement a X64 version soon.
Cpdn news
ID: 30862 · Report as offensive     Reply Quote
Profile MikeMarsUK
Volunteer moderator
Avatar

Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,613,038
RAC: 0
Message 30865 - Posted: 7 Oct 2007, 21:41:32 UTC


As far as I know all the projects you subscribe to use 32-bit code, so there\'d be no problem in running the 32-bit client on your system?

I'm a volunteer and my views are my own.
News and Announcements and FAQ
ID: 30865 · Report as offensive     Reply Quote
ebahapo
Avatar

Send message
Joined: 28 Aug 04
Posts: 29
Credit: 1,061,281
RAC: 8,788
Message 30905 - Posted: 8 Oct 2007, 16:09:43 UTC - in response to Message 30865.  
Last modified: 8 Oct 2007, 16:19:25 UTC


As far as I know all the projects you subscribe to use 32-bit code, so there\'d be no problem in running the 32-bit client on your system?

This kind of reply is not helpful. This is a volunteer project, yet you\'re suggesting the volunteer to change his preferences if he wants to crunch for CPDN. Shouldn\'t CPDN cater to the volunteers instead, if for anything out of gratitude for their contribution?

Come on, how hard is it to create a copy of the 32-bit applications with names including the 64-bit platforms and adding support for them to the server? This has already been done in the beta project...

Thanks.
ID: 30905 · Report as offensive     Reply Quote
Profile MikeMarsUK
Volunteer moderator
Avatar

Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,613,038
RAC: 0
Message 30906 - Posted: 8 Oct 2007, 17:04:31 UTC
Last modified: 8 Oct 2007, 17:14:40 UTC

This kind of reply is not helpful. ...


I am offering a possible workaround if he wishes to use it. How is that not helpful?

However I notice from the list you kindly posted on the other thread that he is running SIMAP, which *does* have a native 64-bit app. I don\'t know what the relative performance of that SIMAP app is, compared to the 32-bit version.

The main advantage of 64-bit code is that it can easily handle process-memory space of 2GB and above, it is not necessarily quicker than 32-bit code (APS is an exception, 50% faster than the 32-bit version). At present there are no Boinc apps using more than 2GB of memory, although at one point Carl was working on a 4.7GB climate model.
I'm a volunteer and my views are my own.
News and Announcements and FAQ
ID: 30906 · Report as offensive     Reply Quote
ebahapo
Avatar

Send message
Joined: 28 Aug 04
Posts: 29
Credit: 1,061,281
RAC: 8,788
Message 30919 - Posted: 9 Oct 2007, 18:00:18 UTC - in response to Message 30906.  
Last modified: 9 Oct 2007, 18:04:22 UTC

However I notice from the list you kindly posted on the other thread that he is running SIMAP, which *does* have a native 64-bit app. I don\'t know what the relative performance of that SIMAP app is, compared to the 32-bit version.

As you can see here, it\'s about 7% faster. With vectorization enabled (which can be assumed in 64 bits), it improves by additional 8%. Therefore, a potential improvement of 15%. Even pessimistically, as CPDN WUs take weeks to complete, there might be a shortening of the run-time by up to 3 days.
The main advantage of 64-bit code is that it can easily handle process-memory space of 2GB and above, it is not necessarily quicker than 32-bit code (APS is an exception, 50% faster than the 32-bit version). At present there are no Boinc apps using more than 2GB of memory, although at one point Carl was working on a 4.7GB climate model.

No. The main advantage of AMD64 over x86 is that it provides more registers and uses SSE by default, allowing for greater performance due to reduced number of instructions and fewer memory loads and stores. Typically, applications display an increased performance by 10% when ported to AMD64 from x86.

This rule-of-thumb is confirmed by SETI, SIMAP, ABC, APS, Chess960 and other BOINC projects. Rather, no performance improvement is the exception, with performance degradation being even more exceptional.

But given the that effort behind a port may be substantial, for the moment upgrading the BOINC server software would be a step in the right direction.

HTH

ID: 30919 · Report as offensive     Reply Quote
old_user627570

Send message
Joined: 9 Jul 10
Posts: 3
Credit: 216
RAC: 0
Message 40097 - Posted: 9 Jul 2010, 21:07:11 UTC - in response to Message 30862.  

mo.v

Can you ask someone in the Met to speak to Sir Richard Branson or Sir Terry Matthews lol.

Those two guys seems very open to donations for scientific projects, I am sure they would be willing to enlist the help of some people they know in getting 64-bit and GPU support up and running for CPDN, considering Met is Britsh as well.

;)
ID: 40097 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 40099 - Posted: 9 Jul 2010, 21:37:30 UTC - in response to Message 40097.  

This is a very old thread, and since the last post before yours was made, 64 bit BOINC support has been added. But the science applications still use 32 bits, because there is no additional benefit from the processor's fpu in going to 64 bits.
Plus that would exclude a large part of the existing cruncher base.

As for talking to the Met Office, the moderators here aren't associated with Oxford Uni, or the Met Office in any way.


Backups: Here
ID: 40099 · Report as offensive     Reply Quote
old_user627570

Send message
Joined: 9 Jul 10
Posts: 3
Credit: 216
RAC: 0
Message 40102 - Posted: 10 Jul 2010, 14:09:56 UTC - in response to Message 40099.  

Ah right, ok thats cool.
ID: 40102 · Report as offensive     Reply Quote
old_user133667

Send message
Joined: 9 Dec 05
Posts: 1
Credit: 39,159
RAC: 0
Message 41921 - Posted: 6 Apr 2011, 22:23:35 UTC - in response to Message 40099.  

Les, I think you miss the point.

There is a desire from the community for a x64 bit client as well as the current x86 client. The project may not deem it necessary, but having a x64 client WILL have improvements in calculation speed for use in the community.
It is possible to service the desires of the user base and achieve the research as desired, TOGETHER. The suggestion was never to disable the 32bit client, just provide a better one for people with the capability.

On top of the x64 debate/request. Given that a large percentage of the world is now running Multi-Core cpus of some description, it would be time for CP to also get Multi-Threading capability like some of the other projects do. And again, this will have dramatic improvements in calculation time.

Remember, we are helping the Project by donating our computer cycles, maintenance, electricity bills and such, for the benefit of the Project. The project should support the community that supports it.

Keith
ID: 41921 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 41922 - Posted: 6 Apr 2011, 23:11:23 UTC - in response to Message 41921.  

I've already said that this project supports both 64 bit computer systems and 64 bit BOINC.
Lots of people have been using them for a long time since this old thread was started in 2007.

The only difficulty is that some Linux 64 bit systems don't automatically provide all of the 32 bit files needed. But that's a problem for the repositories. And they can still be added manually, at least for most systems.

As for what the community wants, perhaps these people should apply for an unpaid job with the University of Oxford, and write/convert the software, because, as you should know from posts in the news thread over the past 6 months, the projects original programmers have been gradually moving on to other work in other locations.
There's still only one part time person here now, although he should soon become full time. And then have a lot of work to do with all of the new models that are required, as well as continuing to sort out the server problems.

Work was done to build a multi-core threaded version of the FAMOUS model for the Millennium project, but the science results were too unstable to even release it to the beta test site.

As for speed, the best way to improve this, is with a fast computer running only this project, 24/7.
Which is what sufficient numbers of people are already doing to provide the various sub-project researchers with quick access to the data they need.


Backups: Here
ID: 41922 · Report as offensive     Reply Quote
ebahapo
Avatar

Send message
Joined: 28 Aug 04
Posts: 29
Credit: 1,061,281
RAC: 8,788
Message 41924 - Posted: 7 Apr 2011, 15:51:34 UTC - in response to Message 41922.  

I'm a seasoned software engineer who contributed to the definition of the x86-64 ISA during my tenure at AMD. I ported the classic SETI application to that architecture and converted many scientific applications to multiple threads. I'd be glad to work on CPDN applications and port them to 64 bits and MT.

ID: 41924 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 41925 - Posted: 7 Apr 2011, 19:24:29 UTC - in response to Message 41924.  

The code is propriety to the UK Met Office, and you'll need a license from them to work on it, or to be working for one of the research institutions that are licensed to do so.


Backups: Here
ID: 41925 · Report as offensive     Reply Quote
Profile JIM

Send message
Joined: 31 Dec 07
Posts: 1152
Credit: 22,053,321
RAC: 4,417
Message 41927 - Posted: 7 Apr 2011, 20:12:49 UTC

There is another thing that needs to be considered here, 64 bit is the future. It’s inevitable. 32 bit is quickly going the way to the brontosaurus. Only five years ago it was relatively rare to find someone running 64 bit Windows. This is no longer true. Most Windows computers are now sold with the 64 bit operating system already installed. This is because it is no longer rare to find computers with 4, 5, or 6 GB’s of RAM installed right out of the box. 32 bit is rapidly being relegated to low power, low coast desktop models and netbooks. This trend will only continue.

There are persistent rumors floating around that Windows 8 might be the last 32 bit version of Windows. When Windows 9 appears 5 or 6 years from now there may not be a 32 bit version. It could be available in 64 and just possibly 128 bit versions only! 32 bit Windows days are numbered.

ID: 41927 · Report as offensive     Reply Quote
ebahapo
Avatar

Send message
Joined: 28 Aug 04
Posts: 29
Credit: 1,061,281
RAC: 8,788
Message 41928 - Posted: 7 Apr 2011, 20:28:28 UTC - in response to Message 41927.  
Last modified: 7 Apr 2011, 20:30:50 UTC

Actually, Windows 7 is the last version of Windows to be available as a 32-bit OS. But, in all fairness, Windows 8 will continue to run 32-bit programs like CPDN just like Windows 7 does. And that without the library issues that some 64-bit Linux systems pose.

However, such a long-running application like CPDN can benefit greatly from some performance improvements. The 64-bit calling interface is much more efficient and floating-point operations are more efficient too, placing intermediate values in registers rather than on a stack as in IA32. The difficulty is that floating-point operations are performed on only 64 bits in a 64-bit application, whereas they are performed on 80 bits in a 32-bit application. This means that the results might be slightly different. Other projects, especially those that run on RISC systems, like SETI and Einstein, have some latitude when validating the results of different platforms.

But then there's the benefit of MT, which could decimate the run-time of CPDN. And this can be done even if the result differences above are a major hurdle in validating the correctness of a 64-bit port. An MT 32-bit application would still be several times faster than the current CPDN application.

If I can get the source code under an NDA, I'd be glad to be of any help.
ID: 41928 · Report as offensive     Reply Quote
Profile Iain Inglis
Volunteer moderator

Send message
Joined: 16 Jan 10
Posts: 1079
Credit: 6,904,878
RAC: 6,593
Message 41929 - Posted: 7 Apr 2011, 21:50:35 UTC
Last modified: 7 Apr 2011, 21:55:28 UTC

As I understand it, some of the CPDN applications were ported from 64-bit to 32-bit to run on PCs - so making a native 64-bit version is a practical question of resources, penetration of 64-bit PCs, numerical stability etc. No doubt they'll do it eventually.

The benefits of multi-threading are, I think, sometimes overstated in relation to distributed computing. If finishing a single-threaded application takes T and an N-threaded application T / N (optimistically) then a fully loaded N-core processor will produce N single-threaded models in T or (T / (T / N)) = N multi-threaded models. So, multi-threading does not produce a greater number of complete models in T - it just produces some of them earlier. It is an advantage only if for some reason T is 'too long'. But in CPDN the time taken to get a working ensemble of models is much greater than T because the failure rate is high and users are not expected to run 24/7.

There are situations where improving the turn-around time could be useful: it might contribute to the slow reduction in failure rate from bug fixes, education and a better BOINC, or it could allow a project team to respond more dynamically to those models that have completed, or actively manage work-unit completion to get quicker ensemble completion etc. etc. However, I suspect there's no free lunch.
ID: 41929 · Report as offensive     Reply Quote
1 · 2 · Next

Questions and Answers : Wish list : Support for BOINC - x64 version

©2024 climateprediction.net