This problem seems to be unavoidable with that whole category of code, regardless of the vector unit used - may it be SSE, AltiVec or VIZ - everything that at a certain point is limited to single precision arithmetic.
I'm currently trying to find out how important the error is scientifically. If it turns out to be neglectable, we would still have to change the validator quite a bit - it's not just tuning of parameters, but adding new features to it - to make it pass such results.
For the x86-based platforms I intend to publish an App rather soon that has been partly coded in assembler and shouldn't have that problem - hopefully.
BM
Thanks. It's a relief to know someone's working on it.
Nothing travels faster than the speed of light with the possible exception of bad news, which obeys its own special laws.
Douglas Adams (1952 - 2001)
I have had 31 successful validations, with none failing validation. This client took 10,000 seconds off of the run time of my Athlon 1200+ Linux 2.4.31 system with no issues other than the graphics thread terminating, but that also occurred on the production client as well.
I installed the 4.55 app on my notebook (Intel Centrino M 1.70GHz), using Debian
testing and kernel 2.6.15-1-686.
Got 8 successfull validations, none failing. Also no proplems with graphics.
This new app runs 40% faster than 4.40!
Great work!
Apart from the big hick up of the validater, hundreds of results have been calculated on an smp machine correctly (2.3.15 SuSE OSS 10). However, there are about 5% without accredited value even though they are correct. Speed up is from about 40,000s to 24,000s. Do the math ;-)
Thanks, Bernd, for the speed gains. I am sure you guys are doing the best you can to speed up things while making sure the project is able to take it. Yet, over four weeks after in to this beta, what is the current status of the findings? Can we expect some more speed improvement or when will the beta become official? Just curious where you wanna go with this one which we are supposed to test.
Thanks
bb
:
your thoughts - the ways :: the knowledge - your space
:
I'm not sure either. Because of the problems you're seeing as invalid, though successful results I think this App will never become the official one, at least not with other people outside Einstein@Home having looked at the code as well as the results.
I'm currently coding a assembler version that is even faster while still keeping enough precision and which is meant to become the basis for all x86 based platforms. I got hold up on this by various problems and other tasks (and a bit of vacation, too), but that's the way the official Apps will probably go.
Thank you, Bernd, for getting back at us. Reading your profile, I thought that you are responsible for the windows port. Yet, from your postings and sitting together with Akos it seems you are becoming (have become?) the chief master coder ;-)
When you talk about the "basis for all x86 based platforms" do you mean that there will be also some modified apps for say SSE, 3DNow! and SSE3 extensions, which can be expected in the future (e.g. S5)?
Just one last question: the server page states that the "last week valid results ~700,000 units" - but there are only ~1,400,000 units left. Does that mean S4 is over in a weeks time or am I completely left field here?
Thanks a lot Bernd for your hard effort. Keep up the good work!
Cheers
bb
:
your thoughts - the ways :: the knowledge - your space
:
valid results per week is about 720k
workunits left about 1.4 million
one workunit needs at least three results
--> about 4.2 million results left
--> about 5.8 weeks to crunch
additionally:
last day about 50k workunits were crunched
--> last avaiable workunit to crunch in about 28 days (15.June)
valid results per week is about 720k
workunits left about 1.4 million
one workunit needs at least three results
--> about 4.2 million results left
--> about 5.8 weeks to crunch
additionally:
last day about 50k workunits were crunched
--> last avaiable workunit to crunch in about 28 days (15.June)
greets
Ooopsi, I left my brain at home when I went out this morning. Thanks for fetching it! :-)
:
your thoughts - the ways :: the knowledge - your space
:
Thank you, Bernd, for getting back at us. Reading your profile, I thought that you are responsible for the windows port. Yet, from your postings and sitting together with Akos it seems you are becoming (have become?) the chief master coder ;-)
Well, I've been responsible for the Apps (i.e. the client-side code) since about the time the project went public last year (and before I last edited my profile...), but by far the most effort was necessary to port our program - that was originally designed to run on *nix clusters - to Windows.
Quote:
When you talk about the "basis for all x86 based platforms" do you mean that there will be also some modified apps for say SSE, 3DNow! and SSE3 extensions, which can be expected in the future (e.g. S5)?
We will keep the principle of one single executable that detects the CPU it is running on and internally calls code that is optimized for this type of CPU. At least regarding the "official", automatically downloaded Apps there should be no need for the user to find out which CPU he has in his "computer" and chose between different Apps himself. In the current (Linux and PPC Mac) Apps we use a scheme that can easily be extended to more than the two types of CPU we currently distinguish.
The next Linux Beta App (that's currently in internal testing) will still use SSE (when available). We don't gain anything from SSE2 (over SSE) and MMX doesn't serve our purpose at all, but I'm right now looking at what we may get from 3DNow!. If it should turn out that SSE3 also gives a noticable improvement I'd be happy to include some code for that as well.
RE: RE: Any idea when a
)
Thanks. It's a relief to know someone's working on it.
Nothing travels faster than the speed of light with the possible exception of bad news, which obeys its own special laws.
Douglas Adams (1952 - 2001)
And another invalid result of
)
And another invalid result of my Clawhammer running 4.55
27362108
SeeYa
Rudi
I have had 31 successful
)
I have had 31 successful validations, with none failing validation. This client took 10,000 seconds off of the run time of my Athlon 1200+ Linux 2.4.31 system with no issues other than the graphics thread terminating, but that also occurred on the production client as well.
I installed the 4.55 app on
)
I installed the 4.55 app on my notebook (Intel Centrino M 1.70GHz), using Debian
testing and kernel 2.6.15-1-686.
Got 8 successfull validations, none failing. Also no proplems with graphics.
This new app runs 40% faster than 4.40!
Great work!
Apart from the big hick up of
)
Apart from the big hick up of the validater, hundreds of results have been calculated on an smp machine correctly (2.3.15 SuSE OSS 10). However, there are about 5% without accredited value even though they are correct. Speed up is from about 40,000s to 24,000s. Do the math ;-)
Thanks, Bernd, for the speed gains. I am sure you guys are doing the best you can to speed up things while making sure the project is able to take it. Yet, over four weeks after in to this beta, what is the current status of the findings? Can we expect some more speed improvement or when will the beta become official? Just curious where you wanna go with this one which we are supposed to test.
Thanks
bb
:
your thoughts - the ways :: the knowledge - your space
:
I'm not sure either. Because
)
I'm not sure either. Because of the problems you're seeing as invalid, though successful results I think this App will never become the official one, at least not with other people outside Einstein@Home having looked at the code as well as the results.
I'm currently coding a assembler version that is even faster while still keeping enough precision and which is meant to become the basis for all x86 based platforms. I got hold up on this by various problems and other tasks (and a bit of vacation, too), but that's the way the official Apps will probably go.
BM
BM
Thank you, Bernd, for getting
)
Thank you, Bernd, for getting back at us. Reading your profile, I thought that you are responsible for the windows port. Yet, from your postings and sitting together with Akos it seems you are becoming (have become?) the chief master coder ;-)
When you talk about the "basis for all x86 based platforms" do you mean that there will be also some modified apps for say SSE, 3DNow! and SSE3 extensions, which can be expected in the future (e.g. S5)?
Just one last question: the server page states that the "last week valid results ~700,000 units" - but there are only ~1,400,000 units left. Does that mean S4 is over in a weeks time or am I completely left field here?
Thanks a lot Bernd for your hard effort. Keep up the good work!
Cheers
bb
:
your thoughts - the ways :: the knowledge - your space
:
valid results per week is
)
valid results per week is about 720k
workunits left about 1.4 million
one workunit needs at least three results
--> about 4.2 million results left
--> about 5.8 weeks to crunch
additionally:
last day about 50k workunits were crunched
--> last avaiable workunit to crunch in about 28 days (15.June)
greets
RE: valid results per week
)
Ooopsi, I left my brain at home when I went out this morning. Thanks for fetching it! :-)
:
your thoughts - the ways :: the knowledge - your space
:
RE: Thank you, Bernd, for
)
Well, I've been responsible for the Apps (i.e. the client-side code) since about the time the project went public last year (and before I last edited my profile...), but by far the most effort was necessary to port our program - that was originally designed to run on *nix clusters - to Windows.
We will keep the principle of one single executable that detects the CPU it is running on and internally calls code that is optimized for this type of CPU. At least regarding the "official", automatically downloaded Apps there should be no need for the user to find out which CPU he has in his "computer" and chose between different Apps himself. In the current (Linux and PPC Mac) Apps we use a scheme that can easily be extended to more than the two types of CPU we currently distinguish.
The next Linux Beta App (that's currently in internal testing) will still use SSE (when available). We don't gain anything from SSE2 (over SSE) and MMX doesn't serve our purpose at all, but I'm right now looking at what we may get from 3DNow!. If it should turn out that SSE3 also gives a noticable improvement I'd be happy to include some code for that as well.
BM
BM