sse3 optimized linux clients ?

[Soulkeeper]

anyone seen some sse3 optimized Linux clients out there yet ?
any experiences ?
what kinda performance gains can sse3 offer over sse and sse2 ?

feel free to paste any benchmark comparisons you might have


thanks

 

[BlackMountainCow]

I remeber having read about SSE3 vs. SSE2 at Einstein@Home and if I remember correctly, it showed only a 5% to 7% increase, which is still a good thing in my opinion. But I haven't seen any released applications nor clients yet.

 

[Soulkeeper]

where do you get einstein optimized clients ?
or is the default one already sse2 optimized ?

 

[BlackMountainCow]

You need to differentiate between client and application.

Client is the BOINC thing that basically does nothing but the benchamrks.
Application is what does the real scientific work.

There are optimized clients - mostly to gain more credits which some folks consider cheating as the ammount of work done stays the same, only the claimed creds are higher

And there are optimized apps - to do more scientific work in the same ammount of time.


Einstein used to have an extra optimized app with the old data, but the current original Einstein app is already optimized.

But that's a question for NetWorkMan, he's the Einstein guru in here.

:beer:

 

[Soulkeeper]

i refered to everything proper
the projects run a client because it interfaces with there network as in client/server relationship
boinc is nothing more than a frontend and stats displayer
the client being optimized isn't gonna give anyone more credit for the same work unit, it will just mean better use of the person's computer resources
saying this is unfair is like saying someone who overclocks there system is cheating
it's a win win situation for both the stats and the humanitarian efforts of the project
if you are sore about someone else having an sse2/sse3 optimized client then just get one yourself

 

[BlackMountainCow]

To my knowledge you're wrong.

If you ever downloaded an optimized CLIENT it will replace the BOINC.EXE file (and a few DLLs).
If you ever downloaded an optimized APP it will replace files like "sulphur_se_4.22_windows_intelx86.exe" or "rosetta_5.25_windows_intelx86.exe".

All the client does is networking, up and download and benchmarks. If you don't believe me, just head over to the SETI forums and take a look at the recent developements there involving Crunch3r's optimized clients and apps.

And as a side mark: Please don't read things into posts which aren't there. I wrote "which some folks consider cheating". That doesn't mean that I do and it doesn't mean that I accused you of cheating. An optimized client does claim more credit for the same ammount of work than a standard client. If that's cheating or just a better representation of what you PC is able to do I leave up to you. And an point like "get a sse2/3 client yourself" is hardly valid in my eyes, as not everybody is enthusiastic about DC like us who visit boards on a daily basis. You should also be able to get fair credits and the best performance out of your PC with the stock and standard client and app.

But I don't want to let this go into a flame war soulkeeper, we've had that before here on the forums and it made people leave for a long time (BadThad and Crazee) but luckily they returned. In that sense, do what you want and use what app or client you want, it doesn't matter to me as long as your results are still scientificaly valid.

 

[Soulkeeper]

i'm not wrong
hahaha

yeah flamewars are bad...

i just don't consider boinc a client in itself
also outta curiousity do you have any proof that an optimized app/client(whatever you wanna call it) claims more credit for the same WU as a non-optimized client/app (i mean sse/sse2/etc. not optimized as in optimized to cheat or skip things or inflate stats artificially) ?

thanks

 

[BlackMountainCow]

Yes I have - for an optimized client.

I run Rosetta with an optimized client which yields me about 40% more credits per WU than with the standard client. But there is no optimized application for Rosetta. That means that I still need about 3 ours per WU, but with my optimized app I receive rather 80 creds than the "normal" 50 creds.

Just take a look here and try for yourself. You'll see that the client gives you much more claimed credit (if it is granted is a different story) than the regular client. But your WU times won't change at all.




If you were to run an optimized application, you would claim LESS credit per WU as the PC needs less time to crunch a WU, but as you process more WUs a day with an optimized app than without, you also get more credit for more WUs. Thus in the end both phenomena level out.

 

[networkman]

BOINC is a client in itself. Previously, you could download both optimized BOINC clients as well as seperate optimized Science apps for the project - referring specifically to Einstein@Home.

A wonderful gentleman by handle of akosf took to compiling optimized science apps to take advantage of the various instruction sets available in various processors; this was wholly seperate from the optimized BOINC clients that you could find on the Web, such as the Truxsoft ones as well as others. Because of the phenomenal improvements that akosf was able to deliver to his science apps, many of us urged the E@H project admins to hire this guy, get him on their staff and delivering the science apps "officially", so that everyone downloading the client/app from the E@H site could benefit, not just those who happened to read the forums. The admins did exactly that, and now the client/app being downloaded from the E@H site is already optimized to take advantage of the various instructions sets of the processors out there.

As for optimized BOINC clients, that depends greatly on how each project is administered -- as some projects like E@H use a quorum system to grant credit, meaning that even if you happen to find an "optimized" client that reports higher numbers for "claimed credit", the project still issues "granted credit" based on what is submitted, typically, dropping the high number and low number of 3 units submitted and taking the middle number, which is very different than an average. For example, if you use an "optimized" client and submit a unit with a value of 80, and two others are put in at 10 & 15, the "granted credit will be 15, where as an average would be 35, but averaging isn't what's being done. The only way to really reap the "benefits" of an optimized client is to get as many people using them as possible.

There are some projects however that do NOT use a quorum type system, nor any other of comparing submitted results by one user against those of another, meaning that the "granted credit" is equal to the "claimed credit"! In a project using this scheme, an optimized client would be a HUGE advantage over those people not running such a client. My personal belief is that this is blatant cheating, but that's really a whole other subject.

Edit: I found a little write-up on the whole "optimized" thing in general in a message thread at the BoincSynergy's team site HERE

 

[BlackMountainCow]

Thx for elaborating on the point NWM!

:beer:

 

[Soulkeeper]

interesting stuff
things have changed a lot since seti classic as far as wu credit goes and the other more obvious stuff


thanks for the input guys