Pages: [1] 2
UBT - Halifax--lad
 
BAM!ID: 25
Joined: 2006-02-27
Posts: 366
Credits: 49,272
World-rank: 919,245

2006-03-11 10:49:56

I have been using the optimised Einstein App for the last couple of days and am impressed with what I have seen.



[color=]My WU's are 50% quicker than before and others are reporting faster times as well. The results I have sent back have all validated & got credit.




Thread about Optimisation



<!--coloro:#009900--><span style="color:#009900"><!--/coloro-->To install the optimised app all you need to do is download the relevant Zip file from below, it contains a .exe file for Einstein, just extract the file to the Einstein folder in BOINC and it should overwrite the exsisting .exe file in there.[/color]

<!--coloro:#009900--><span style="color:#009900"><!--/coloro-->[/color]

<!--coloro:#009900--><span style="color:#009900"><!--/coloro-->I would make sure your cache is empty of Einstein WU's 1st before you do this, then let the WU's flow again once installed, the WU processing will speed up massively.[/color]





<!--coloro:#ff0000--><span style="color:#ff0000"><!--/coloro-->Caution: Use at your own risk[/color]



<!--coloro:#0000ff--><span style="color:#0000ff"><!--/coloro-->387[/color] -> 386, Family 4 (486,5x86,MediaGX), Family 5 (Pentium,PentiumMMX,K5,K6(all),6x86,MediaGX-MMX) and some others oldies.

<!--coloro:blue--><span style="color:blue"><!--/coloro-->A36[/color] -> Family 6 (AMD K7 core processors,PentiumPro,PentiumII,PentiumIII,6x86MX,M2), Family 7 (Itanium), Family 15 (Athlon64,Pentium4 and so...)

[/color]

Join us in Chat (see the forum) Click the Sig
bartsob5
Translator
BAM!ID: 811
Joined: 2006-05-26
Posts: 151
Credits: 49,567,751
World-rank: 18,154

2006-03-11 18:40:11

but how claimed credit look like? is it the same situation like in SETI, and you should have optimized boinc client too?
UBT - Halifax--lad
 
BAM!ID: 25
Joined: 2006-02-27
Posts: 366
Credits: 49,272
World-rank: 919,245

2006-03-11 19:27:22

but how claimed credit look like? is it the same situation like in SETI, and you should have optimized boinc client too?




I'm not here for credits am here for the science so anything that speeds up the scientific process is good for me

Join us in Chat (see the forum) Click the Sig
Trog Dog
 
Volunteer tester
BAM!ID: 206
Joined: 2006-05-11
Posts: 119
Credits: 66,172,318
World-rank: 14,985

2006-03-12 02:51:42
last modified: 2006-03-12 22:03:03

Thanks for the heads up Halifax Lad.
Lee Carre
 
BAM!ID: 41
Joined: 2006-04-19
Posts: 262
Credits: 299,581
World-rank: 397,496

2006-03-13 23:04:54
last modified: 2006-03-13 23:50:40

Thanks Halifax
Always looking for something to plough thru WUs faster

[edit]hmmm, just finished reading that whole thread (phew, quite a long one) and it's a tad worrying that it's not officially supported, that is, bruce hasn't given the OK, but then there's been no objection either[/edit]
Want to search the BOINC Wiki, BOINCstats, or various BOINC forums from within firefox? Try the BOINC related Firefox Search Plugins
UBT - Halifax--lad
 
BAM!ID: 25
Joined: 2006-02-27
Posts: 366
Credits: 49,272
World-rank: 919,245

2006-03-14 08:19:02

Thanks Halifax
Always looking for something to plough thru WUs faster

[edit]hmmm, just finished reading that whole thread (phew, quite a long one) and it's a tad worrying that it's not officially supported, that is, bruce hasn't given the OK, but then there's been no objection either[/edit]




I wouldn't worry about it, its producing results that validate and if there was a problem with using it someone would have said no by now

Join us in Chat (see the forum) Click the Sig
Lee Carre
 
BAM!ID: 41
Joined: 2006-04-19
Posts: 262
Credits: 299,581
World-rank: 397,496

2006-03-14 08:47:53
last modified: 2006-03-14 08:48:08

I wouldn't worry about it, its producing results that validate and if there was a problem with using it someone would have said no by now

true, i'll run my queue down and give it a go
50+% increase sure does sound good
Want to search the BOINC Wiki, BOINCstats, or various BOINC forums from within firefox? Try the BOINC related Firefox Search Plugins
Guest

2006-03-17 12:42:16

According to akosf, he's now recommending everyone use c37

http://einstein.phys.uwm.edu/forum_thread.php?id=3895

I know I am not akosf, but in order to keep things in a central location, I think the following is beneificial. Delete if you need to.

387 -> 386, Family 4 (486,5x86,MediaGX), Family 5 (Pentium,PentiumMMX,K5,K6(all),6x86,MediaGX-MMX) and some others oldies.

A36 -> Family 6 (AMD K7 core processors,PentiumPro,PentiumII,PentiumIII,6x86MX,M2), Family 7 (Itanium), Family 15 (Athlon64,Pentium4 and so...)


C37 is better for all.


More details, it apperently is using 386 specific optomizations, but gives 20% performance improvements over either 387 or a36... That's in other threads on this...

By and large the crunch times are decent, completing in about 30-40 or so minutes on my Athlon 64... I do have 1 WU here for E@H that's been crunching for 1 hour 26 mins, and only 92% done... So there are some that will still take long. Looking at credits, others were claiming between 38 or so and 43 credits, so I'm guessing long unit... Most, the claimed credit is no more then half of what gets claimed on that unit, so there are a few which even with this app will take about that long...
Guest

2006-03-19 18:10:42


According to akosf, he's now recommending everyone use c37


I stopped BOINC when I had no Einstein units, switched to this binary, re-started... but couldn't download a WU. I reset the project, which re-downloaded the official engine, but I got a WU which wasn't running yet. Stop BOINC, replace engine, restart, wait....

This is the first optomized engine I've loaded. I must say, I'm impressed.

It would be nice if future versions of BOINC could run checks for specific functions (MMX, 3DNow, SSE 1 through 3, and so on), and then report back to projects which optomized version to download and run. I know the programming for this is intimidating, but perhaps in a few years.
Guest

2006-03-24 19:14:11
last modified: 2006-03-24 19:15:33

For SSE capable cpus there is a new version: S39L

For even credits, take a look at truxofts calibrating core client

AndyK
Guest

2006-04-07 12:40:15

He's added D40 for those of us using an AMD CPU. The main thing it adds is 3Dnow optimizations. Over C37, this has brought another significant increase...

After the short WUs were done, I started getting one's that took about 1.5 hours under c37... Under d40, they're taking about 55-60 minutes...

There was a newer app to d40, but on a subsequent post yesterday, he suggested people not get it till he has a time to nail down a problem with it. Which leaves d40 as the lattest d/l akosf recommended people get (at least as of yesterday). Probably s39l (I would gather) for those that don't have 3Dnow capable CPUs...
Trog Dog
 
Volunteer tester
BAM!ID: 206
Joined: 2006-05-11
Posts: 119
Credits: 66,172,318
World-rank: 14,985

2006-04-08 10:12:57
last modified: 2006-04-08 10:18:57


He's added D40 for those of us using an AMD CPU. The main thing it adds is 3Dnow optimizations. Over C37, this has brought another significant increase...

After the short WUs were done, I started getting one's that took about 1.5 hours under c37... Under d40, they're taking about 55-60 minutes...

There was a newer app to d40, but on a subsequent post yesterday, he suggested people not get it till he has a time to nail down a problem with it. Which leaves d40 as the lattest d/l akosf recommended people get (at least as of yesterday). Probably s39l (I would gather) for those that don't have 3Dnow capable CPUs...



Now official optimised apps have been released too, although only as beta's.

If you're using one of Akosf's apps then stick with it, but if you are one of the many that wasn't game to try anything unofficial then grab a copy of the optimised beta.

For Linux
For Windows
Guest

2006-04-10 03:39:41


He's added D40 for those of us using an AMD CPU. The main thing it adds is 3Dnow optimizations. Over C37, this has brought another significant increase...

After the short WUs were done, I started getting one's that took about 1.5 hours under c37... Under d40, they're taking about 55-60 minutes...


I loaded S39L, as mentioned by AndyK. Your results look similar to mine. I have an Athalon XP 3200+. The results from C37 were taking about 1h20m - 1h30m. With S39L, the results are taking about an hour. I don't use the screen saver, so I don't know if I would benefit from the 3Dnow optimization. Otherwise, I didn't see a download for the D40 executable.
Guest

2006-04-10 08:46:48

Oh, I browsed in one of the app threads (think it was the read only one), and nicked it... There was a newer one like S40 or something, but Akosf was suggesting people don't grab it then. It might be OK now however, as he was working on a problem...

From the Windows thread, the new beta 4.50 is taking longer to execute then akosf's executable, so I guess there are a few optimizations he included they have yet to port over. That they're getting some optimizations into the Linux app is good though, as akosf's app is naturally Windows only...
Lee Carre
 
BAM!ID: 41
Joined: 2006-04-19
Posts: 262
Credits: 299,581
World-rank: 397,496

2006-04-13 15:29:25
last modified: 2006-04-13 15:30:33

the new beta 4.50 is taking longer to execute then akosf's executable, so I guess there are a few optimizations he included they have yet to port over.
it would be good if he was allowed access to the source code, all his optimisations have been on the compiled binary itself, and he seems really enthusiastic and determined about the whole thing, i think he could make a valuable contribution to the science code and increase the speed even further


A word of warning, S40.03 (and possibly S40 as well) seems to return invalid results, and thus you'll get no credits for them (in my experience, and yes; my systems are stable)

others may have better luck, but for anyone just wanting a hastle free install for windows on intel CPUs: S37L is still the best app (again only in my experience, other apps may be better on other processors)
Want to search the BOINC Wiki, BOINCstats, or various BOINC forums from within firefox? Try the BOINC related Firefox Search Plugins
Guest

2006-04-14 20:19:18


A word of warning, S40.03 (and possibly S40 as well) seems to return invalid results, and thus you'll get no credits for them (in my experience, and yes; my systems are stable)

others may have better luck, but for anyone just wanting a hastle free install for windows on intel CPUs: S37L is still the best app (again only in my experience, other apps may be better on other processors)


I thought akosf said the problems were with his Duron processor. I've been running S40.03 (a small bugfix of 40) for a few days now, and so far all my WU's have validated OK. I only run Einstein on two computers (occasionally a WU or two on a third, slow PC), so bad WU's would start appearing in my profile rather quickly.

Of course, I could just be lucky.

Is there a writeup of which of his apps are for which processors? I run Windows on an AMD Athlon XP 3200+, so I don't know how well the "optomized for Intel" will help. I was running a D-series optomization, which just amazed me how much faster it ran. I went to the latest S-series, because it seemed more AMD specific.

Marky-UK
BAM!ID: 100
Joined: 2006-05-10
Posts: 77
Credits: 15,186,686
World-rank: 41,050

2006-04-15 09:45:57
last modified: 2006-04-15 09:48:01


Is there a writeup of which of his apps are for which processors?

There's a list somewhere, but basically: D = 3DNow, S = SSE, C = no SSE. So AMD can usually run all 3 (with D being better than S), and Intel can run S or C.

Edit: Found a list, on one of the mirror pages: http://einstein.terrorfront.info/en/index.htm
Guest

2006-04-15 14:10:39


There's a list somewhere, but basically: D = 3DNow, S = SSE, C = no SSE. So AMD can usually run all 3 (with D being better than S), and Intel can run S or C.
Edit: Found a list, on one of the mirror pages: http://einstein.terrorfront.info/en/index.htm


Thank you very much! I've bookmarked it.

I'm curious, though, if anyone might know the benefit of 3DNow for anything but the screen saver. Does it assist in processing the data, or is it only to run the screen saver's celestial globe with fewer clock cycles?
Marky-UK
BAM!ID: 100
Joined: 2006-05-10
Posts: 77
Credits: 15,186,686
World-rank: 41,050

2006-04-15 21:22:29
last modified: 2006-04-15 21:25:03


I'm curious, though, if anyone might know the benefit of 3DNow for anything but the screen saver. Does it assist in processing the data, or is it only to run the screen saver's celestial globe with fewer clock cycles?

3DNow provides a bunch of fast floating-point instructions that the optimised app uses for the science bit (not the screen saver). Although 3DNow was marketed as a way to improve 3D graphics, it can be used for other things too - especially if they need lots of floating-point arithmetic.
Lee Carre
 
BAM!ID: 41
Joined: 2006-04-19
Posts: 262
Credits: 299,581
World-rank: 397,496

2006-04-15 23:38:06
last modified: 2006-04-15 23:39:15

3DNow provides a bunch of fast floating-point instructions that the optimised app uses for the science bit (not the screen saver). Although 3DNow was marketed as a way to improve 3D graphics, it can be used for other things too - especially if they need lots of floating-point arithmetic.
could 3DNow! be summarised as the AMD equivelent of SSE?
Want to search the BOINC Wiki, BOINCstats, or various BOINC forums from within firefox? Try the BOINC related Firefox Search Plugins
Guest

2006-04-16 01:52:03


could 3DNow! be summarised as the AMD equivelent of SSE?


Actually, I believe most modern and recent AMD processors have SSE. That is different.

SSE has to do with SIMD instructions, or "Single Instruction Multiple Data". The Mac's PowerPC chip is build around SIMD processing, as are video cards. SIMD kicks in when you have a large data set, and the same instruction will be applied to all data. For example, in pixel shading in a video card, you want to decrease a pixel's luminosity based on how far it is from a light source. It is a simple equation to determine this. By using SIMD data (or SSE instruction sets), you can load the points into the processor, and process the points with the single instruction quickly. Instead of loading the formula and point and processing for each point, you can think of it as loading the pixel luminosity formula once, loading a portion of a plane to be lit up, and the processor starts going through applying the formula to the points. The formula is applied to as many points as (1) can be loaded, and (2) the process pipe allows. The processor then spits out a group of results, rather than using a process pipeline to just get a single result. It's sort of a "quick and dirty" way of approximating parallel computing on a data set.

SIMD data is also very common in medical and bio-informatics research. In these fields, you tend to have large data sets, but only want to apply a simple equation across all the data points. That is why Apple tried to court many research institutions when the G5 processor was released.

I *think* SSE stands for "Simple SIMD Execution".
Lee Carre
 
BAM!ID: 41
Joined: 2006-04-19
Posts: 262
Credits: 299,581
World-rank: 397,496

2006-04-16 05:04:21
last modified: 2006-04-16 05:04:50

Actually, I believe most modern and recent AMD processors have SSE. That is different.

... ... ...

I *think* SSE stands for "Simple SIMD Execution".
interesting read
Want to search the BOINC Wiki, BOINCstats, or various BOINC forums from within firefox? Try the BOINC related Firefox Search Plugins
Marky-UK
BAM!ID: 100
Joined: 2006-05-10
Posts: 77
Credits: 15,186,686
World-rank: 41,050

2006-04-16 18:33:54
last modified: 2006-04-16 18:40:46


could 3DNow! be summarised as the AMD equivelent of SSE?

Yeah, except 3DNow! came before SSE. 3DNow! was AMD's answer to Intel's MMX, except AMD spotted the weakness of MMX (it only did integer arithmetic) and introduced 3DNow! which did floating-point too (it's also compatible with MMX). Intel then introduced SSE which was their answer to 3DNow! Most modern AMD processors seem to do MMX, 3DNow! and SSE.

Edited to add: SSE = Streaming SIMD Extensions
Lee Carre
 
BAM!ID: 41
Joined: 2006-04-19
Posts: 262
Credits: 299,581
World-rank: 397,496

2006-05-12 12:22:22
last modified: 2006-05-23 17:40:09

General Update:

As new apps are being released all the time, serveral new, and faster, ones are available, the latest versions can be found on the einstain site, in the forums

to find out which is appropriate for your platform; search the einstein forums, the answers are there
Want to search the BOINC Wiki, BOINCstats, or various BOINC forums from within firefox? Try the BOINC related Firefox Search Plugins
Guest

2006-05-23 06:29:01

Yeah, there have been more recent updates. I'm currently running u41-04 on my Athlon 64, and crunch times have dropped back down to about 34-35 minutes on these (now not so new) longer Albert WUs. This was from over 50 minutes on D40.

One word of note however... U41 uses SIMD3 instructions. If the A64 has an E3 core or latter, you'll be fine using this. Else one should use D40 on their AMD. My A64 uses a Venice core (which is E3) so in my case U41-04 is better. As to other A64 or X2 users, it'd be best to check CPUz to see what your core stepping is first, and confirm it's something like E3, E6, perhaps newer. If it's an older core stepping, grab the 3Dnow executable. Needless to say, anyone with an AXP won't have SIMD3 supported.

Actually, come to think of it, I think CPUz will say if SIMD3 is supported or not, so can go by that...
Pages: [1] 2

Index :: Optimized clients :: Einstein Optimisation
Reason: