I like your avatar Uli ;)

Sorry for OT:
Hehe, it's my original avatar on the german firefox board. At some point everyone wanted to have an avatar with something gecko'ish on it. (Gecko is the render engine used under the hood of firefox.) I got so used to it, i now use it everywhere ;)

You can get DevCPP here: http://www.bloodshed.net/
Personally i prefer MinGWStudio: http://www.parinyasoft.com/
Both are free!

[edit]
How about using Eclipse + CDT (C++ Developer Tools) + MinGW/GCC?
Then you'll have a platform independent development solution, completely for free! :D

Sounds like an interesting idea, actually. Independence from commercial tools isn't bad at all. Also, you can of course use IPP and ICC with Eclipse as well. In fact, the Linux ICC comes with an Eclipse plug-in. Not sure about Windows, but since it's all Java anyway, who knows.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

You can get DevCPP here: http://www.bloodshed.net/
Personally i prefer MinGWStudio: http://www.parinyasoft.com/
Both are free!

[edit]
How about using Eclipse + CDT (C++ Developer Tools) + MinGW/GCC?
Then you'll have a platform independent development solution, completely for free! :D

The main reason for using ICC is that it has been found to generate really fast code.

Of course, there could be even faster compilers out there that haven't been tried.

I've been fairly curious about the Visual C requirement -- if all it does is act like a "MAKE" utility, then it should be easy to replace it. Doesn't ICC have a "MAKE" utility as part of the (Windows) package?

---

(...) AFAIK Eric began to use devcpp (MinGW32) for WIndows official release, but he also updates VC++ project files. (...)

Really? :O
That would be the greatest thing, cause MinGW (GCC for Windows) has an extremely vast collection of optimization switches not only for different instruction sets but also for different cpu architectures, especially for all derivates of AMD processors. :D

Hi Simon,

Having spent MANY hours experimenting with various compiler optimizations for GCC/Linux, the following are some that we found to be beneficial in addition to the usual SIMD and CPU architecture-specific switches.

-malign-double -maccumulate-outgoing-args -fomit-frame-pointer -falign-functions=4 -finline-limit=1024 -fprefetch-loop-arrays -funroll-loops -fforce-addr -ftree-vectorize -ffast-math -ftracer

(all testing was performed on Athlon XP processors with non-enhanced seti, so YMMV ;))

Regards,

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

Any word yet on optimized clients for WinXP?

My host could desperately use SSE 2/3 support for some of these compute-heavy WU's ... ...and I do not have a slow host.

Largest so far my host has sucessfully crunched is ~100 credits. That's big!

{the hanging WU's that act like they need infinite CPU are another matter...}

---

Word from me is that so far, my applications are around 3-5% slower than Crunch3rs 5.12 on Windows (but quicker on Linux) so there ought to be more that can be gotten out of it.

No idea about others :o) People haven't been saying too much about their level of success.

Daniel Schaalma was working on a build for Windows AFAIK too, no idea on his progress.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Any word yet on optimized clients for WinXP?

My host could desperately use SSE 2/3 support for some of these compute-heavy WU's ... ...and I do not have a slow host.

Largest so far my host has sucessfully crunched is ~100 credits. That's big!

{the hanging WU's that act like they need infinite CPU are another matter...}

One of my optimized boxes just crunched a 94.21 credit WU too - they're few and far between, that's my 3rd that I have seen out of hundreds.

I'm about to give up on my Windows builds getting a lot faster - don't know whether Crunch3r used all the newest Intel products (I used the free non-com ones for now which are one version behind). Funnily enough though, on Linux they still seem faster (same versions used there).

Also, I tried devcpp (mingw etc) but builds failed horribly (even without ICC), so not sure whether I'm going to go that way. Eclipse etc. I have yet to try.

Visual Studio 2005 compiles the BOINC client itself fine, but the seti_boinc crunch3r gives a couple of errors, probably fixable though. Just looks like a stricter syntax that it adheres to. The solution files import fine - but of course, you can't open them with 2003 anymore once you import them, so phooey on MS for that...been that way with EVERY Visual Studio release.

Does anyone think it's worth it getting apps that are slower than previous optimized ones? Not me, really...I'll still be writing up how to do it though. Maybe someone has a suggestion to improve speed :o)

VTune hasn't given me anything, by the way.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Greetings Simon,

"Does anyone think it's worth it getting apps that are slower than previous optimized ones"

I for one am awaiting your "How To" for Windoze with anticipation. Unfortunately I was on vacation for a month when Crunch3rs opt application was released and missed out on downloading it before it was removed. I am presently stuck crunching at a snails pace. Your efforts and time are greatly appreciated and it should be fun to build my own opt application using your guide when it's posted.

Keep up the great work.

Hi Simon can u give some numbers u got on that win version
Here is my stats on an AMD Barton 2800+ no sse2

standard 795.906250
sse 575.375000
Crunch3r 4.11 sse 537.406250 (wrong credit claim one)

Thanx
Conrad

---

Simon, have you experimented with "turning off" or "commenting out" the graphics functionality from the Win build?

Crunch3r seemed to think that graphics capability had to be removed from the Win binary(s) in order to get best speed.

Also, what are you doing with VTune to profile the code and find hot spots?

As to the other question:
I frankly never want to run a "not officially blessed" client ever again. Regardless of its speed.

So for me the perfect world is for the best compile options etc to be found and "pushed up" to the s@h folks so that BOINC and s@h will do the compiles using all the Good Stuff you, Alax, etc have found.

OTOH, if that can't or won't happen fast enough then I want to crunch as many results as I can before BOINC and/or s@h dies.

So post a link to some binaries and in parallel let's all try and get this stuff supported officially.

---

So for me the perfect world is for the best compile options etc to be found and "pushed up" to the s@h folks so that BOINC and s@h will do the compiles using all the Good Stuff you, Alax, etc have found.

OTOH, if that can't or won't happen fast enough then I want to crunch as many results as I can before BOINC and/or s@h dies.

So post a link to some binaries and in parallel let's all try and get this stuff supported officially.

A big part of this is that BOINC sees Windows pretty much as Windows. It does not know an Athlon X2 from a VIA C3.

The newest beta BOINC client is reporting capabilities, but I would expect it to be a while before a 5.6 official version knows enough to ask the server for a specific science app. tuned to the specific processor.

For the short term, if you want a faster app. you'll have to use the anonymous platform.

---

...
Does anyone think it's worth it getting apps that are slower than previous optimized ones?
...
Simon.

There are new participants joining the project, some participants have complied with Crunch3r's request to stop using his apps, and some simply missed downloading when they were available. Builds which are faster than stock would be useful to all those, and of course benefit the project by getting more work done in a given time.

Even without using the Intel tools, builds for specific SIMD capabilities should provide a valuable improvement. That would avoid the issue of Intel license terms for Windows. I'd like to see a set of those made available for download, very few participants running Windows will be able to build their own fully optimized versions.

                                                                  Joe

I agree Josef, which is why I was playing with devcpp earlier.

However, it died horribly on building no matter what I did (maybe I just forgot something, but I'm more familiar with Visual Studio). There are other ways to go on Windows - using cygwin and gcc and Makefiles, for example. Get rid of all that nasty graphical IDE and go back to good old configure. I'll be trying that, too :o)

Today I bought a system to dedicate to client development - mainly because I need the system I've been doing testing on for something else. So I installed 4 operating systems on it - WinXP32/64 and Debian Sarge 32/64. The chip can do SSE3 as well, so it should be able to compile whatever.

I'll have to study the free for personal non-commercial use license again and see whether it prohibits me from releasing my builds or not.

I have access to a commercial license, but it's not mine. I'd have to get permission from someone else to do that and I don't want to go that way.

I tested GCC vs. ICC on Linux - though I did not spend a lot of time on experimenting with GCC flags, it was about 12-15% quicker than the default 5.12 build whereas ICC is 30-150% quicker depending on AR. So that's really a power of magnitude separating them.

Nevertheless, worth a try.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Simon,

I want to also thank you for your hard work on this. I sure hope you are able to come up with a Windows build, as there are quite a lot of people that run it, and would love to have an optimized app. Good luck, and thank you for keeping us informed.

---

OTOH?

Simon, have you experimented with "turning off" or "commenting out" the graphics functionality from the Win build?

Crunch3r seemed to think that graphics capability had to be removed from the Win binary(s) in order to get best speed.

Also, what are you doing with VTune to profile the code and find hot spots?

As to the other question:
I frankly never want to run a "not officially blessed" client ever again. Regardless of its speed.

So for me the perfect world is for the best compile options etc to be found and "pushed up" to the s@h folks so that BOINC and s@h will do the compiles using all the Good Stuff you, Alax, etc have found.

OTOH, if that can't or won't happen fast enough then I want to crunch as many results as I can before BOINC and/or s@h dies.

So post a link to some binaries and in parallel let's all try and get this stuff supported officially.

---

Why not go to an academic library and look up Intel in Dun & Bradstreet? Find out who the new CEO is and write him a personal letter (by snail mail, maybe registered) asking him to donate a compiler, IPP library, and tech support for a year to the project. Explain how so many individual are contributing thousands of dollars in electricity to the effort. It is worth a try.

Thanks Gecko!

Let me quote a part of the email I got back from Eric Korpela after asking him about specifics on optimizing and distributing S@H.

My question:
> In addition, I'd like to ask how you feel about someone paying for an
> educational-type license for ICC and IPP and offering X86 builds (and maybe OS X, as
> they now support it) and using this compiler and library package to release the
> official BOINC-distributed Apps. This License would be made out to the Seti project,
> of course. Build scripts to automate this already exist.

Eric's response:
That's a good idea, but I'm not sure whether SETI@home falls into the
academic use license terms. Before spending any money the buyers
should check the license terms.

So that should explain where I personally would like this to go. Level playing field, anyone? Preaching to the choir, I know...

In related matters, I'm trying to find contacts at Intel - since they make the compiler and tools we're using to optimize, it seems logical to ask them whether they would like to sponsor the project
(or rather, provide us with the necessary licenses). Hardware would be neat too, but that's probably a pipe dream :o)

If anyone has contacts at la Intella, please don't hesitate to pitch this idea to them (and/or get in touch with me).

Regards,
Simon.

---

Thanks for the idea C Elliot -
you aren' the first to think of that, however, I have yet to find out who to write to exactly.

It doesn't help that I'm european - so just swinging into your local library and grabbing that data might prove more difficult than anticipated.

If anyone wants to take the time to do that and hit me with the details, I'd be more than glad to write a snail-mail letter, because that's what I think should be done, too.

Email just doesn't cut it, doesn't have any personal touch. These days, important things arrive on paper ;o)

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

OTOH?

On The Other Hand.

---

Oh, and I do know who the current CEO is, but finding contact details...

Paul Otellini (it used to be Craig Barrett, who is now chair of the board). Anyone? :o)

Also, I'm not sure whether Anand Chandrasekher or Eric Kim wouldn't be the more likely persons to contact - they're senior VPs and GMs of Sales and Marketing, which I think this would fall under.

Link: http://www.intel.com/pressroom/ExecBios.htm

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Simon: You see this yet?

http://www.pixelglow.com/macstl/

New version 0.3.1 support for x86 Linux

There's a lot to look at.
This caught my attention:

http://www.pixelglow.com/stories/macstl-intel-autovectorization/

Thought it was worth the mention.
Regards.

---

Looks promising - but the stats look a bit unbelievable.

Still, worth checking out as well - if I can figure out how to set up a dev environment to use it, I'll give it a whirl and do straight comparisons (as I know have a dedicated machine to).

Much easier on all the other stuff I do and zero other tasks allotted to it, so results should have less variance (they did fluctuate up to around +-3% before due to different load).

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information