Are there any sites providing optimized clients? -- PART II

Message boards : Number crunching : Are there any sites providing optimized clients? -- PART II
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 3 · 4 · 5 · 6 · 7 · 8 · 9 . . . 19 · Next

AuthorMessage
Profile KWSN - Chicken of Angnor
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 9 Jul 99
Posts: 1199
Credit: 6,615,780
RAC: 0
Austria
Message 349707 - Posted: 27 Jun 2006, 5:12:04 UTC
Last modified: 27 Jun 2006, 5:13:07 UTC

And thanks again Alex Kan!

He has released a version of his OS X G4/G5 optimized clients without graphics. This should make Macs even faster :o)

Kudos and hats off to him, as usual.

Regards,
Simon.
Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information
ID: 349707 · Report as offensive
AC
Avatar

Send message
Joined: 22 Jan 05
Posts: 3413
Credit: 119,579
RAC: 0
United States
Message 349709 - Posted: 27 Jun 2006, 5:15:12 UTC
Last modified: 27 Jun 2006, 6:03:00 UTC

Hmm, that does look good. 16627.984375 (3800) vs. 12023.583137 (3500). Although the CPU time was less, the granted credit result with your version was very interesting. [edit] Compare that to one of my results with a Pentium 4 532 3.06 GHz, 1 MB L2.

Yee Haw!


ID: 349709 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 66517
Credit: 55,293,173
RAC: 49
United States
Message 349750 - Posted: 27 Jun 2006, 6:47:55 UTC - in response to Message 349703.  

Well the two results I linked to above have been reported. I was the first to report so I'll have to wait a while to see them validated.

And in the meantime 2 more have finished:
346850557 and 346850627.

Looks like they took 4.3 hours or so. The initial "to completion" estimate was 7.5 hours.

Take a look at this WU (or all that host's WUs).

Compare the first result (Athlon64 3800+ using Crunch3r's 5.12 on Windows) and the 3rd, (my Athlon64 3500+, using my Linux SSE2 build in a Linux VM on Windows).
This is a high-credit WU, of which there are really not a lot. One host took almost 180.000 seconds to return it.

There haven't been a lot of other optimized clients among the results I've looked at, and this seems a pretty close match (with 200 MHz clock speed advantage on the 3800+'s side).

Tests are underway to see which one is quicker on which hardware at which AR - AMD and Intel both.
I'm also going to investigate the unlikely case that Linux is actually quicker running in a VM on Windows than on its own...cause it sure seems like it so far, believe it or not.

Regards,
Simon.

P.S. on an unrelated note, I noticed with satisfaction that my pending credit is over 3000 again (hardly went above 1000 with enhanced so far), so it seems I'm returning more results than before.


Mine was above 3000 too come to think about It, Seems strange, And now It's around 2950 or so, Of course the valiators are workin It looks like, There seems to be a mystery a foot Me thinks and No I'm not Sherlock Holmes. It just sounds cool to Me. ;) I'm getting near the 500,000 club too, Then all I have left is to slowly close on the Uncle B's, Hardly a fair fight, My 10GHz vs 144GHz, Even with all modern cpus at near 44GHz soon, I don't think I can match them, I feel like a Mouse attackin an Elephant. Where's an elephant swatter when I need one. Ho hum.
CA HSR built a foundation, is laying Track!
PRR T1 Class 4-4-4-4 #5550 Loco, US's 1st HST

ID: 349750 · Report as offensive
Profile Toby
Volunteer tester
Avatar

Send message
Joined: 26 Oct 00
Posts: 1005
Credit: 6,366,949
RAC: 0
United States
Message 349756 - Posted: 27 Jun 2006, 6:53:39 UTC
Last modified: 27 Jun 2006, 6:54:16 UTC

Just finished another. Looking at my results I see it claimed exactly the same credit as a work unit I finished with the standard client right before I switched which I assume means they are the same AR?

stock client = 11,615 seconds and optimized = 7,660 seconds. In both cases, credit claimed was 31.13
A member of The Knights Who Say NI!
For rankings, history graphs and more, check out:
My BOINC stats site
ID: 349756 · Report as offensive
Profile KWSN - Chicken of Angnor
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 9 Jul 99
Posts: 1199
Credit: 6,615,780
RAC: 0
Austria
Message 349758 - Posted: 27 Jun 2006, 6:57:02 UTC
Last modified: 27 Jun 2006, 6:57:19 UTC

The AR is exactly the same on those two results, yes.

You'll notice different sorts of speedups with different ARs. The ones that take longest seem to be the quickest compared to the default client.

I'll shortly be removing my optimized apps for a couple of WUs on my hosts so I can get an idea how it looks.

Regards,
Simon.
Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information
ID: 349758 · Report as offensive
EricVonDaniken

Send message
Joined: 17 Apr 04
Posts: 177
Credit: 67,881
RAC: 0
United States
Message 349803 - Posted: 27 Jun 2006, 8:01:38 UTC - in response to Message 349643.  


It also seems to jibe with what I read out of the Windows license stuff for the Intel packages - namely, that you can't really release stuff compiled with the trial version (and there is no non-com version on Windows according to the Intel page, only commercial trial versions which expire).

So that's why I decided to only release Linux for now, since my use and distribution classifies as personal and non-commercial according to the license distributed by Intel (and my understanding of it).

Barring further developments (I already posted about the as yet unsent snail mail letter), that won't change quickly. I have access to a commercial license but it ain't mine, so that's not an option. Having to ask people to download code, compile it for me then send it back each time just doesn't work, not to mention them maybe having issues and/or getting flak for their kind help.

Regards,
Simon.

How much do all the relevant licenses cost?

ID: 349803 · Report as offensive
Profile KWSN - Chicken of Angnor
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 9 Jul 99
Posts: 1199
Credit: 6,615,780
RAC: 0
Austria
Message 349806 - Posted: 27 Jun 2006, 8:10:32 UTC
Last modified: 27 Jun 2006, 8:17:57 UTC

Around 900 dollars AFAIK, but I haven't checked closely. Sadly, the licenses are time-limited, I believe that's for one year and includes ICC, IPP and MKL (the latter two being library packages).

Intel pricing page

Also don't forget both Linux and Windows (and maybe even OS X, for X86-based Macs) are required, so that doubles/triples normal license costs.

--edit
Intel Compiler for Linux - $399
Intel Compiler for Windows - $399
Intel Compiler for OS X - $399

IPP - $199
MKL - $399 (maybe not necessary, but probably faster than fftw).

From the pricing page, it seems that all supported OS flavours are included when you buy one license for the library packages. The compiler has to be licensed once per OS.

There's a promotional MacOS package available that includes Compiler, IPP and MKL and costs $549. Since the libraries seem to license cross-OS, that might be the cheapest path overall.

Regards,
Simon.
Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information
ID: 349806 · Report as offensive
Profile Diego -=Mav3rik=-
Avatar

Send message
Joined: 1 Jun 99
Posts: 333
Credit: 3,587,148
RAC: 0
Message 349872 - Posted: 27 Jun 2006, 10:21:38 UTC - in response to Message 349806.  

Intel Compiler for Linux - $399
Intel Compiler for Windows - $399
Intel Compiler for OS X - $399
IPP - $199
MKL - $399 (maybe not necessary, but probably faster than fftw).


Yes, give me 2, please. :>

I've been buying intel chips since like 1989. Intel owes ME a lot, so I'll ask them to give me a license as a present. And then I'll give it to you.

I'll let you know how my petition goes. ;)

/Mav

We have lingered long enough on the shores of the cosmic ocean.
We are ready at last to set sail for the stars.

(Carl Sagan)
ID: 349872 · Report as offensive
Pepperammi

Send message
Joined: 3 Apr 99
Posts: 200
Credit: 737,775
RAC: 0
United Kingdom
Message 350138 - Posted: 27 Jun 2006, 14:43:15 UTC - in response to Message 349806.  

Around 900 dollars AFAIK, but I haven't checked closely. Sadly, the licenses are time-limited, I believe that's for one year and includes ICC, IPP and MKL (the latter two being library packages).

Simon.

Strange question.

Does all this legal rubish stop you from realeasing your 'un-compiled' work? before you've compiled it with the intel pakage its your intelectual property/ under gpl ect though obviously but they cant say you've broke the intel or any compilers license agreement can they?

What i mean is have the program you've made ready - compile it to test on your system - if it works then release the pre-compile source or tarball or what ever its called. Then we download it and compile it with the free :) intel (or whatever) compiler that anyone can download. :) With instructions would be helpfull :)

Sorry if i'm missing something obvious but i've only just started to try to learn to code myself and i wont even bother to try and understand the legal cr*p.
ID: 350138 · Report as offensive
Alinator
Volunteer tester

Send message
Joined: 19 Apr 05
Posts: 4178
Credit: 4,647,982
RAC: 0
United States
Message 350146 - Posted: 27 Jun 2006, 14:56:53 UTC
Last modified: 27 Jun 2006, 15:07:14 UTC

No, the source code is your's to do with as you see fit.

The license only applies once you run it through the compiler and wish to release for public use.

I guess that was the point Maverick was making earlier, he can still work on optimizing with the Intel products, but he can't release them anymore because his license expired.

Alinator

<edit> On second thought, there might be a problem with providing the exact config files for the compiler, but I would think there's nothing wrong with including a text "crib" sheet to help folks along.
ID: 350146 · Report as offensive
Pepperammi

Send message
Joined: 3 Apr 99
Posts: 200
Credit: 737,775
RAC: 0
United Kingdom
Message 350149 - Posted: 27 Jun 2006, 15:01:13 UTC - in response to Message 350146.  
Last modified: 27 Jun 2006, 15:05:37 UTC

No, the source code is your's to do with as you see fit.

The license only applies once you run it through the compiler and wish to release for public use.

I guess that was the point Maverick was making earlier, he can still work on optimizing with the Intel products, but he can't release them anymore because his license expired.

Alinator

So someone could release there optimized source material with some instructions on compiling and when to get the compilers(free?) and they wouldn't be breaking any rules. :)
Now just wish i know how to do all that. I doubt compileing is as easy as all that.
<edit> On second thought, there might be a problem with providing the exact config files for the compiler, but I would think there's nothing wrong with include a text "crib" sheet to help folks along.

Oh. darn it. Thanks for the info anyway.
ID: 350149 · Report as offensive
Alinator
Volunteer tester

Send message
Joined: 19 Apr 05
Posts: 4178
Credit: 4,647,982
RAC: 0
United States
Message 350153 - Posted: 27 Jun 2006, 15:04:09 UTC - in response to Message 350149.  

So someone could release there optimized source material with some instructions on compiling and when to get the compilers(free?) and they wouldn't be breaking any rules. :)
Now just wish i know how to do all that. I doubt compileing is as easy as all that.


Note see edit above.

Yep, that much is certain, it's a lot harder than it sounds!

Alinator
ID: 350153 · Report as offensive
EricVonDaniken

Send message
Joined: 17 Apr 04
Posts: 177
Credit: 67,881
RAC: 0
United States
Message 350181 - Posted: 27 Jun 2006, 15:49:31 UTC - in response to Message 349806.  
Last modified: 27 Jun 2006, 15:49:55 UTC

Around $900 AFAIK, but I haven't checked closely. Sadly, the licenses are time-limited, I believe that's for one year and includes ICC, IPP and MKL (the latter two being library packages).

Intel pricing page

Also don't forget both Linux and Windows (and maybe even OS X, for X86-based Macs) are required, so that doubles/triples normal license costs.

--edit
Intel Compiler for Linux - $399
Intel Compiler for Windows - $399
Intel Compiler for OS X - $399

IPP - $199
MKL - $399 (maybe not necessary, but probably faster than fftw).

From the pricing page, it seems that all supported OS flavours are included when you buy one license for the library packages. The compiler has to be licensed once per OS.

There's a promotional MacOS package available that includes Compiler, IPP and MKL and costs $549. Since the libraries seem to license cross-OS, that might be the cheapest path overall.

Regards,
Simon.


I find it hard to believe that Crunch3r was paying $900 US every year just so he could make BOINC and s@h optimizations!

Something does not make sense here. How is Berkeley making WinXP SW using the Intel stuff? How is MIT (FFTW)?
There has to be some form of academic/non-commercial/non-profit site licensing scheme available; and most likely one that is not time-bombed.

There are far more WinXP hosts than Linux hosts running BOINC, so it seems fairly important that we find a way to make those hosts run a effeciently and effectively as possible.
ID: 350181 · Report as offensive
1mp0£173
Volunteer tester

Send message
Joined: 3 Apr 99
Posts: 8423
Credit: 356,897
RAC: 0
United States
Message 350190 - Posted: 27 Jun 2006, 16:08:14 UTC - in response to Message 349504.  

EricVonDaniken,

to my knowledge he bought a license, and that's how :o)

Regards,
Simon.

A= can you verify that so we are sure of what thw truth is?

B= how much does a license cost?

A= if you search the forums, I think you'll find where he asked for donations.

B= I see someone has already given pricing.
ID: 350190 · Report as offensive
1mp0£173
Volunteer tester

Send message
Joined: 3 Apr 99
Posts: 8423
Credit: 356,897
RAC: 0
United States
Message 350199 - Posted: 27 Jun 2006, 16:20:16 UTC - in response to Message 350181.  

Around $900 AFAIK, but I haven't checked closely. Sadly, the licenses are time-limited, I believe that's for one year and includes ICC, IPP and MKL (the latter two being library packages).

Intel pricing page

Also don't forget both Linux and Windows (and maybe even OS X, for X86-based Macs) are required, so that doubles/triples normal license costs.

--edit
Intel Compiler for Linux - $399
Intel Compiler for Windows - $399
Intel Compiler for OS X - $399

IPP - $199
MKL - $399 (maybe not necessary, but probably faster than fftw).

From the pricing page, it seems that all supported OS flavours are included when you buy one license for the library packages. The compiler has to be licensed once per OS.

There's a promotional MacOS package available that includes Compiler, IPP and MKL and costs $549. Since the libraries seem to license cross-OS, that might be the cheapest path overall.

Regards,
Simon.


I find it hard to believe that Crunch3r was paying $900 US every year just so he could make BOINC and s@h optimizations!

All of the licenses I saw are perpetual, the support contracts do run out.

Are the support contracts needed? Probably not.


Something does not make sense here. How is Berkeley making WinXP SW using the Intel stuff? How is MIT (FFTW)?
There are several other compilers that are free (GCC) or less expensive.

It is important to remember that Crunch3r's big claim to fame was that he was not using the same compiler as Berkeley, and he licensed some libraries that are not free (IPP), and then told the compiler to produce code tailored to specific processors.

I think TMR did the work to figure out ICC/IPP (someone please correct me if I'm wrong) and ICC may or may not be the best optimizing compier out there. Moving existing source from one compiler to another is work, and it's not the fun kind of work, it's just a slow grind to find all the differences between compilers.

There has to be some form of academic/non-commercial/non-profit site licensing scheme available; and most likely one that is not time-bombed.

There are far more WinXP hosts than Linux hosts running BOINC, so it seems fairly important that we find a way to make those hosts run a effeciently and effectively as possible.

One way to do this is to simply publish an "optimization cookbook" with links to available tools, installation instructions and etc.

Then anyone can download an eval. license for the compiler and libraries, and build their own. They just can't distribute the compiled code without violating the Intel license.

Another way is to find someone (surely out of the 460,000-odd users out there someone must have it) who already has a license and have them do the final compiles for us. There are ways to do that without violating the license, and without making the compiler owner do lots of work.
ID: 350199 · Report as offensive
EricVonDaniken

Send message
Joined: 17 Apr 04
Posts: 177
Credit: 67,881
RAC: 0
United States
Message 350206 - Posted: 27 Jun 2006, 16:34:15 UTC

Craziness.

The best way to solve the problem is for Berekeley and/or BOINC to get the right licensing.

Frankly, I'd bet Berkeley does have a site Intel license for these tools and we just do not know it.
ID: 350206 · Report as offensive
Profile KWSN - Chicken of Angnor
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 9 Jul 99
Posts: 1199
Credit: 6,615,780
RAC: 0
Austria
Message 350249 - Posted: 27 Jun 2006, 19:24:36 UTC
Last modified: 27 Jun 2006, 20:15:20 UTC

And I believe you'd be wrong Eric -

where do you see any inkling that Berkeley have even used any of the Intel packages themselves? The official client is compiled using GCC(Linux) and Microsoft Visual C++ (Windows) and NOT using Intel's package.

No, I do not believe that Berkeley or Seti@Home has a license for them.

And yes, it seems that Crunch3rs license was paid for using donations via paypal for at least a part of the cost. I haven't asked for anything similar because I don't think it's the right way.

TMR said his license ran out (he also found a sponsor) - so if only the support packages run out, he should actually still be able to release things with it. That also would alleviate some of the licensing cost (the recurring part), obviously.

I agree that there is only a problem releasing Apps on Windows right now BUT - that only goes for me. The project itself cannot use the personal free for non-commercial tools, namely because it's not a person as such.

So to use them, they would definitely need a License for Linux, too.

There are of course academic license programs available (just check their site), for which the project may qualify. In any case, that is something that I can neither influence nor do for them - all I can do, as promised, is write to Intel and see whether they would like to help us out.

Regards,
Simon.
Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information
ID: 350249 · Report as offensive
Profile Byron Leigh Hatch @ team Carl Sagan
Volunteer tester
Avatar

Send message
Joined: 5 Jul 99
Posts: 4548
Credit: 35,667,570
RAC: 4
Canada
Message 350251 - Posted: 27 Jun 2006, 19:32:51 UTC
Last modified: 27 Jun 2006, 19:33:46 UTC



On 6/16/06, Charles Elliott wrote:

> Why don't you use the Intel compiler and libraries? You
> would not have to do much, if any, re-writing as it can
> produce runtime versions that pick and choose what instructions
> to use based on the capabilities of the CPU.

I don't think the compiler actually does what you are suggesting here.
(I'm prepared to be convinced otherwise, though). The last version I
tried didn't (a couple years ago).

> Moreover, in the studies I have read the Intel tools are the
> best in the industry and produce the fastest executables.

Most of the time we have to develop to the most common tools rather
than the fastest. The cost in my time required to switch compilers is
far larger than the cost of the compiler. GCC, OTOH, works everywhere
on every platform.

Eric

the above was copied from here ............


http://www.ssl.berkeley.edu/pipermail/boinc_opt/2006-June/000764.html





ID: 350251 · Report as offensive
Profile KWSN - Chicken of Angnor
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 9 Jul 99
Posts: 1199
Credit: 6,615,780
RAC: 0
Austria
Message 350261 - Posted: 27 Jun 2006, 19:44:24 UTC
Last modified: 27 Jun 2006, 20:16:31 UTC

Thanks Byron,

I'm subscribed to that list already :o)

Actually, you can now do exactly what Eric Korpela said in that message you couldn't, which is have one executable that combines optimization for a couple of platforms (didn't use to be possible but now works).

The relevant compiler switches on Windows are /QaX and /QxX where X is either K, W, N or B (K-> P3/SSE, W ->Generic SSE2, N -> P4 SSE2, B -> P4 SSE3). So you can also combine more than one to make it look like /QaKWNB, which would make a monolithic executable that's optimized from P3 upwards and executes everywhere.

It'll be larger than the original, of course, and it has about a 3% performance penalty vs. a "pure" optimized binary. Compared to the default, though, that's still massively quicker.

A point to remember, to be sure.

Regards,
Simon.
Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information
ID: 350261 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 350292 - Posted: 27 Jun 2006, 20:45:56 UTC - in response to Message 350249.  

The official client is compiled using GCC(Linux) and Microsoft Visual C++ (Windows) and NOT using Intel's package.
...
Simon.

Eric Korpela switched to DevC++/MinGW for the Windows builds in Beta starting with the 5.10 version, and has used that combination for all Windows releases here. He also keeps the VC++ project files updated, of course.
                                                    Joe

ID: 350292 · Report as offensive
Previous · 1 . . . 3 · 4 · 5 · 6 · 7 · 8 · 9 . . . 19 · Next

Message boards : Number crunching : Are there any sites providing optimized clients? -- PART II


 
©2025 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.