If everything was balanced properly, I'd imagine all apps (optimized or not) would be claiming 30-40 credits for a "normal" WU regardless of whether or not it took 5 minutes or 5 hours to finish.
That's exactly what S@h Enhanced is attempting to achieve, by reckoning in Flops instead of CPU-seconds. Unfortunately, older BOINC clients don't know what to do with the app's reported number of FFTs performed, and continue to make claims based on benchmarks-&-time.
If everything was balanced properly, I'd imagine all apps (optimized or not) would be claiming 30-40 credits for a "normal" WU regardless of whether or not it took 5 minutes or 5 hours to finish.
That's exactly what S@h Enhanced is attempting to achieve, by reckoning in Flops instead of CPU-seconds. Unfortunately, older BOINC clients don't know what to do with the app's reported number of FFTs performed, and continue to make claims based on benchmarks-&-time.
That's still not ideal - if an optimised app uses a more efficient algorithm than the official app (and thus needs fewer FLOPs to finish a work unit) then optimised apps will claim fewer credits for the same amount of science.
The only way round this would seem to be to have separate credit totals for each project, and to award some credits for every WU completed (varying the amount of credits for a WU according to its size). Unfortunately then you couldn't easily combine credits across projects because projects have different WU sizes. So this solution still has problems. (Although it might encourage projects to use smaller WUs and more optimised apps, which would be good for science and for owners of older / less powerful computers.)
I don't think there is a solution which everyone will agree is the best solution. :-(
I'm not to sure which file(s) would be missing. If you go to the directory where you extracted "albert_4.37_windows_intelx86.exe" what files are there?
Like I said...I've got about 3 of those d/ls in here somewhere. I have no idea how to find it. Perhaps I should detach from einstein and reattch? Would that point me in the right direction and leave me with just one app?
Founder of BOINC group, Objectivists, a group of philosophically minded rational data crunchers.
Like I said...I've got about 3 of those d/ls in here somewhere. I have no idea how to find it. Perhaps I should detach from einstein and reattch? Would that point me in the right direction and leave me with just one app?
That might be easiest. When you re-attach your host will download the official app and some new work, suspend all work and exit, then try the steps previously posted.
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
Like I said...I've got about 3 of those d/ls in here somewhere. I have no idea how to find it. Perhaps I should detach from einstein and reattch? Would that point me in the right direction and leave me with just one app?
Why not do a search for "alb*" this will show you the location of each file starting with alb its then just a matter of tidying things up.
You should have a file "albert_4.37_windows_intelx86.exe" in a directory like "C:\\Program Files\\BOINC\\projects\\einstein.phys.uwm.edu" (this is your project directory) and then you should have one or more .zip files named "alb_?????.zip" where ????? will be a letter followed by four numbers(this is the zip file containing the optimised app).
Once you have the zip file extract its contents to your project directory. When you're asked file already exists, overwrite? choose yes.
That's exactly what S@h Enhanced is attempting to achieve, by reckoning in Flops instead of CPU-seconds. Unfortunately, older BOINC clients don't know what to do with the app's reported number of FFTs performed, and continue to make claims based on benchmarks-&-time.
That's still not ideal - if an optimised app uses a more efficient algorithm than the official app (and thus needs fewer FLOPs to finish a work unit) then optimised apps will claim fewer credits for the same amount of science.
The only way round this would seem to be to have separate credit totals for each project, and to award some credits for every WU completed (varying the amount of credits for a WU according to its size). Unfortunately then you couldn't easily combine credits across projects because projects have different WU sizes. So this solution still has problems. (Although it might encourage projects to use smaller WUs and more optimised apps, which would be good for science and for owners of older / less powerful computers.)
I don't think there is a solution which everyone will agree is the best solution. :-(
Optimized app make be more efficient (isn't it the goal?) but while you will be able to finish more WUs, credit per hour/day will be very similar. So, that' not the point.
Idea of making WUs size (resp. crunch time) similar among BOINC project is very unrealistic. Some project *need* large WUs than can't be sliced - CPDN for exaple. Some projects can't determinated prior Wus is crunched how long it will take - BURP for exmaple. And trird argument - one can't expect that all project will remake their design, application and everything to meet some credit demands :-)
But idea of *fair* interchangeble credit (I don't count benchmark*CPU-time in this category) has been proposed many times with realistic how-to. Just one example: have small calibration WUs that will measure real performance (vs. synthetic benchmark easily manipuled...ehh optimalized) which will server also as a stability/validity test...
I think - according to SETI Enhanced Beta experience - FLOPs count will serve quite well; for sure better than nowadays benchmark. Claimed credit gets comparable within project (unlike with benchmark) and it also solve "problem" of where quorum is not used (Rosetta) and only one machine benchmark takes place.
When other projects will incorporate FLOPs count in their app is another story. I believe UCB/SETI did a good step...
If everything was balanced properly, I'd imagine all apps (optimized or not) would be claiming 30-40 credits for a "normal" WU regardless of whether or not it took 5 minutes or 5 hours to finish.
That's exactly what S@h Enhanced is attempting to achieve, by reckoning in Flops instead of CPU-seconds. Unfortunately, older BOINC clients don't know what to do with the app's reported number of FFTs performed, and continue to make claims based on benchmarks-&-time.
That's still not ideal - if an optimised app uses a more efficient algorithm than the official app (and thus needs fewer FLOPs to finish a work unit) then optimised apps will claim fewer credits for the same amount of science.
The only way round this would seem to be to have separate credit totals for each project, and to award some credits for every WU completed (varying the amount of credits for a WU according to its size). Unfortunately then you couldn't easily combine credits across projects because projects have different WU sizes. So this solution still has problems. (Although it might encourage projects to use smaller WUs and more optimised apps, which would be good for science and for owners of older / less powerful computers.)
I don't think there is a solution which everyone will agree is the best solution. :-(
An optimized app does not do less work, it does the same amount of work more efficiently! The new optimized apps that are being worked on for seti are using Fpops and still claim the same amount of credit as the official app. Why, because the same amount of work is still done.
Alright, I need big time help with a step by step for DUMMIES style instructions on optimizing 3 machines for einstein. Please give me the BIG TIME 'for dummies' instructions. :-)
I've tried d/l some of this for this AMD but got the wrong one I think. Here are my machines. If I can get this figured out I'll put it on the 3rd machine and run einstein there also.
AuthenticAMD
AMD Athlon(tm) XP Processor 3000+ Microsoft Windows XP
Home Edition, Service Pack 2, (05.01.2600.00) 5 5 May 2006 10:13:31 UTC
479026 3.94 850.11 GenuineIntel
Intel(r) Celeron(tm) Processor Microsoft Windows Millennium
, (04.90.3000.00)
Also have a 1.7 gig Celeron w/512ram running XP. (this is the 3rd machine)
Thank you in advance. If I can double or triple my crunch time it would be the equivalent of adding 6-12 machines running non optemized Akos apps.
I think you are meaning which of akosf's apps ( versions ) are best for each machine as above? I'm looking here, here and here.
#1 Authentic AMD Athlon XP 3000+ with MS Win XP Home - looks like you have 3DNow and SSE capability thus S41.07 is the go.
#2 Intel(r) Celeron(tm) Processor with MS Windows Millennium - ?
#3 1.7 gig Celeron w/512ram with XP - ?
For the Celeron's I don't think it's clear from your description ( ie. are they 'newer' ? ), so go get CPU-Zhere ( upper left under 'download latest version' : CPU-Z 1.33.1 ). This checks out a system for you to determine what's what. Unzip and run 'cpuz.exe' directly from where-ever you downloaded it too ( you don't need to install ). The first property sheet/tab on the left ( 'CPU' ), in the upper box 'Processor' has 'Instructions' at the lower boundary listing the modes ( on mine MMX, SSE, SSE2, SSE3, EM64T ). Note these down for each of your three systems and post back here, please.
Cheers, Mike.
I've been crunchin for Einstein and of course Seti for a while and My 1st question is:
1. Are optimized Einstein Apps/Clients officially allowed here and encouraged?
In Seti there are those that call It cheating to use an Optimized app/client.
2. And what about the credit system and crunching times, Will either be changed as in longer crunching times?
Seti just added an enhanced 5.12 app that counts in fpop instead and the latest Boinc Client 5.4.9 doesn't do calibration like Trux's 5.3.12.tx36 and Yeah I use Trux's client and the U104 app, Works great on Joker2 which runs at 2.65GHz I've seen results in the 19's and on anothe PC Joker3 which runs at 2.36GHz I saw about 13 or so in Einstein and Joker3 is the Dual cpu PC, Both rigs use Opteron Dual cores of course.
And Yes I have 3 Rigs, 1 is complete, 1 is half complete(It needs a 2nd cpu and the last one needs 2 cpus and video cards).
My PCs specs are Here.
Ignore the 4th one as I don't have any plans for that as My finances changed some. Going to have to fix that part of that page one day to something like Joker3.
1. Are optimized Einstein Apps/Clients officially allowed here and encouraged?
Yep. They have just hired the guy, Akos Fekete ( all hail!! ), who produced the optimised apps, in order to begin generalising the benefits to non-Windows boxes. It is preferred that calibrating clients ( eg. Truxoft ) are used to prevent quora being downgraded in credit ( where there is a mix between optimised/non-optimised apps ).
Quote:
In Seti there are those that call It cheating to use an Optimized app/client.
Don't recall that being said here, but I could be wrong. If anything the fourfold-plus improvement in productivity from optimisations has put pressure on the server, resulting in daily download limits and that has been a contentious issue particularly for the higher end crunchers..... look here for an instance of 'robust' discussion on that....... :-)
Quote:
2. And what about the credit system and crunching times, Will either be changed as in longer crunching times?
Can't be precise here, as there is quite a variation in the types of WU's presented to the apps ( generically called p's, z's, j's etc according to the initial letter of the file ), hence times/credits etc. These serve different science aims ( true vs. fake signal detection say ) and is a bit complex to cross compare ( the j's have 6 per quorum I think ) but it still comes down to the quora/validation process ....... :-)
Cheers, Mike.
( edit ) Oh, and you have some nice rigs too ..... :-)
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
I am now running an optimized app for an AMD sse processer. It now crunches a work unit in about 45min which previously took about 2 hours and 30 minutes. WOW
Great work Akos
RE: If everything was
)
That's exactly what S@h Enhanced is attempting to achieve, by reckoning in Flops instead of CPU-seconds. Unfortunately, older BOINC clients don't know what to do with the app's reported number of FFTs performed, and continue to make claims based on benchmarks-&-time.
RE: RE: If everything was
)
That's still not ideal - if an optimised app uses a more efficient algorithm than the official app (and thus needs fewer FLOPs to finish a work unit) then optimised apps will claim fewer credits for the same amount of science.
The only way round this would seem to be to have separate credit totals for each project, and to award some credits for every WU completed (varying the amount of credits for a WU according to its size). Unfortunately then you couldn't easily combine credits across projects because projects have different WU sizes. So this solution still has problems. (Although it might encourage projects to use smaller WUs and more optimised apps, which would be good for science and for owners of older / less powerful computers.)
I don't think there is a solution which everyone will agree is the best solution. :-(
RE: RE: > One or more
)
Like I said...I've got about 3 of those d/ls in here somewhere. I have no idea how to find it. Perhaps I should detach from einstein and reattch? Would that point me in the right direction and leave me with just one app?
Founder of BOINC group, Objectivists, a group of philosophically minded rational data crunchers.
RE: Like I said...I've got
)
That might be easiest. When you re-attach your host will download the official app and some new work, suspend all work and exit, then try the steps previously posted.
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
RE: Like I said...I've
)
Why not do a search for "alb*" this will show you the location of each file starting with alb its then just a matter of tidying things up.
You should have a file "albert_4.37_windows_intelx86.exe" in a directory like "C:\\Program Files\\BOINC\\projects\\einstein.phys.uwm.edu" (this is your project directory) and then you should have one or more .zip files named "alb_?????.zip" where ????? will be a letter followed by four numbers(this is the zip file containing the optimised app).
Once you have the zip file extract its contents to your project directory. When you're asked file already exists, overwrite? choose yes.
RE: RE: That's exactly
)
Optimized app make be more efficient (isn't it the goal?) but while you will be able to finish more WUs, credit per hour/day will be very similar. So, that' not the point.
Idea of making WUs size (resp. crunch time) similar among BOINC project is very unrealistic. Some project *need* large WUs than can't be sliced - CPDN for exaple. Some projects can't determinated prior Wus is crunched how long it will take - BURP for exmaple. And trird argument - one can't expect that all project will remake their design, application and everything to meet some credit demands :-)
But idea of *fair* interchangeble credit (I don't count benchmark*CPU-time in this category) has been proposed many times with realistic how-to. Just one example: have small calibration WUs that will measure real performance (vs. synthetic benchmark easily manipuled...ehh optimalized) which will server also as a stability/validity test...
I think - according to SETI Enhanced Beta experience - FLOPs count will serve quite well; for sure better than nowadays benchmark. Claimed credit gets comparable within project (unlike with benchmark) and it also solve "problem" of where quorum is not used (Rosetta) and only one machine benchmark takes place.
When other projects will incorporate FLOPs count in their app is another story. I believe UCB/SETI did a good step...
RE: RE: RE: If
)
An optimized app does not do less work, it does the same amount of work more efficiently! The new optimized apps that are being worked on for seti are using Fpops and still claim the same amount of credit as the official app. Why, because the same amount of work is still done.
98SE XP2500+ @ 2.1 GHz Boinc v5.8.8
RE: RE: Alright, I need
)
I've been crunchin for Einstein and of course Seti for a while and My 1st question is:
1. Are optimized Einstein Apps/Clients officially allowed here and encouraged?
In Seti there are those that call It cheating to use an Optimized app/client.
2. And what about the credit system and crunching times, Will either be changed as in longer crunching times?
Seti just added an enhanced 5.12 app that counts in fpop instead and the latest Boinc Client 5.4.9 doesn't do calibration like Trux's 5.3.12.tx36 and Yeah I use Trux's client and the U104 app, Works great on Joker2 which runs at 2.65GHz I've seen results in the 19's and on anothe PC Joker3 which runs at 2.36GHz I saw about 13 or so in Einstein and Joker3 is the Dual cpu PC, Both rigs use Opteron Dual cores of course.
And Yes I have 3 Rigs, 1 is complete, 1 is half complete(It needs a 2nd cpu and the last one needs 2 cpus and video cards).
My PCs specs are Here.
Ignore the 4th one as I don't have any plans for that as My finances changed some. Going to have to fix that part of that page one day to something like Joker3.
http://contract.rebuildthedream.com/
http://tinyurl.com/Pass-HJ47-NO-Time-Limit-On-ERA
RE: 1. Are optimized
)
Yep. They have just hired the guy, Akos Fekete ( all hail!! ), who produced the optimised apps, in order to begin generalising the benefits to non-Windows boxes. It is preferred that calibrating clients ( eg. Truxoft ) are used to prevent quora being downgraded in credit ( where there is a mix between optimised/non-optimised apps ).
Don't recall that being said here, but I could be wrong. If anything the fourfold-plus improvement in productivity from optimisations has put pressure on the server, resulting in daily download limits and that has been a contentious issue particularly for the higher end crunchers..... look here for an instance of 'robust' discussion on that....... :-)
Can't be precise here, as there is quite a variation in the types of WU's presented to the apps ( generically called p's, z's, j's etc according to the initial letter of the file ), hence times/credits etc. These serve different science aims ( true vs. fake signal detection say ) and is a bit complex to cross compare ( the j's have 6 per quorum I think ) but it still comes down to the quora/validation process ....... :-)
Cheers, Mike.
( edit ) Oh, and you have some nice rigs too ..... :-)
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
I am now running an optimized
)
I am now running an optimized app for an AMD sse processer. It now crunches a work unit in about 45min which previously took about 2 hours and 30 minutes. WOW
Great work Akos
Picture is of Saturn