Welcome to MilkyWay@home

New faster application?

Message boards : Number crunching : New faster application?
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · 10 · 11 . . . 12 · Next

AuthorMessage
Beau

Send message
Joined: 3 Jan 09
Posts: 270
Credit: 124,346
RAC: 0
Message 8460 - Posted: 16 Jan 2009, 13:13:38 UTC - in response to Message 8458.  

Heres what I got for the SSEv2 zip

Optimized
fitness: -3.046272445362275

Stock
fitness: -3.046272445361065


So are these results too different?
ID: 8460 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile banditwolf
Avatar

Send message
Joined: 12 Nov 07
Posts: 2425
Credit: 524,164
RAC: 0
Message 8461 - Posted: 16 Jan 2009, 13:22:56 UTC - in response to Message 8460.  

Heres what I got for the SSEv2 zip

Optimized
fitness: -3.046272445362275

Stock
fitness: -3.046272445361065


So are these results too different?


It is to the 11th place, so I believe they are fine. But let Travis say for sure.
Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.
ID: 8461 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 8462 - Posted: 16 Jan 2009, 13:23:14 UTC - in response to Message 8454.  

@ Travis

Why linux, OSX, and Win stock apps gave different results with the same wu?

Best regards.


At what level of accuracy?


Didn't you tested it...? Bad boy...:)

Try it, and give us your results.

Best regards.


With what we tested them on they were basically identical. I was wondering if someone was getting different results.

ID: 8462 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile banditwolf
Avatar

Send message
Joined: 12 Nov 07
Posts: 2425
Credit: 524,164
RAC: 0
Message 8463 - Posted: 16 Jan 2009, 13:24:59 UTC - in response to Message 8458.  

So, we have a new version of the SSE3 app that meets the 10 digit requirement. Does anyone have an updated SSE or SSE2 version?


Here we go...

http://rapidshare.com/files/184217251/astronomy_sse2_app.rar.html

Original app:
fitness: -3.046272445362277

this one:
fitness: -3.046272445362278

have fun...

LB


Thanks that helps me out.
Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.
ID: 8463 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[BOINCstats] LostBoy

Send message
Joined: 8 Mar 08
Posts: 4
Credit: 774,815
RAC: 0
Message 8464 - Posted: 16 Jan 2009, 13:25:06 UTC - in response to Message 8460.  
Last modified: 16 Jan 2009, 13:27:49 UTC

too slow....

I don t think so.
This is what Travis wrote:
Yeah that's fine. I think the cutoff where it would start to matter would be differences greater than 10^-10ish

So the last five digits are not necessary,i think...
There are only the last four incorrect...

Correct me,if i am wrong...

LB
ID: 8464 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 8465 - Posted: 16 Jan 2009, 13:25:50 UTC - in response to Message 8461.  

Heres what I got for the SSEv2 zip

Optimized
fitness: -3.046272445362275

Stock
fitness: -3.046272445361065


So are these results too different?


It is to the 11th place, so I believe they are fine. But let Travis say for sure.


Yeah this is fine. It would be nice to see how it does with a couple different inputs, just to make sure. Is this optimized app done with compiler flags or has there been significant code changes? If the 2nd is the case it would be nice if they were shared so we could update the stock app.
ID: 8465 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 8466 - Posted: 16 Jan 2009, 13:25:52 UTC - in response to Message 8461.  

Heres what I got for the SSEv2 zip

Optimized
fitness: -3.046272445362275

Stock
fitness: -3.046272445361065


So are these results too different?


It is to the 11th place, so I believe they are fine. But let Travis say for sure.


Yeah this is fine. It would be nice to see how it does with a couple different inputs, just to make sure. Is this optimized app done with compiler flags or has there been significant code changes? If the 2nd is the case it would be nice if they were shared so we could update the stock app.
ID: 8466 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
David @ TPS

Send message
Joined: 20 Nov 08
Posts: 24
Credit: 2,561,361
RAC: 0
Message 8467 - Posted: 16 Jan 2009, 13:48:35 UTC

Gonna try re-attaching with one SSE3 box. If I end up losing points I will shut it back down.

ID: 8467 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile GalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
Message 8468 - Posted: 16 Jan 2009, 14:00:54 UTC - in response to Message 8467.  
Last modified: 16 Jan 2009, 14:02:11 UTC

Gonna try re-attaching with one SSE3 box. If I end up losing points I will shut it back down.


I've been running with the latest SSE3 app all day without problems. The WUs are taking slightly (minimally) longer, but it's still turning WUs around far faster than the stock app.

ID: 8468 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
sandro

Send message
Joined: 17 Oct 08
Posts: 16
Credit: 16,783
RAC: 0
Message 8469 - Posted: 16 Jan 2009, 14:04:44 UTC - in response to Message 8466.  
Last modified: 16 Jan 2009, 14:05:46 UTC


Yeah this is fine. It would be nice to see how it does with a couple different inputs, just to make sure. Is this optimized app done with compiler flags or has there been significant code changes? If the 2nd is the case it would be nice if they were shared so we could update the stock app.

Im not shure but i think they "hand-optimzed" the code rather using compiler flags, dont know whtat is significant for you ;) . but better to ask them directly.
ID: 8469 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile [AF>HFR>RR] Sp0wn

Send message
Joined: 16 Mar 08
Posts: 10
Credit: 59,990,626
RAC: 0
Message 8470 - Posted: 16 Jan 2009, 14:09:12 UTC - in response to Message 8449.  
Last modified: 16 Jan 2009, 14:12:20 UTC

So I guess this SSE3-Version saves the party for us.

Result of the Test-WU:
fitness: -3.046272445362275


Yes, I also got
fitness: -3.046272445362275

But that still does not match the stock app output of
fitness: -3.046272445362277


Are they close enough it won't change the outcome (of the search) or should they be exact?


Yeah that's fine. I think the cutoff where it would start to matter would be differences greater than 10^-10ish


The app linked above is the only one I know of that is good. The last post above (the "Yeah that's fine" one) is from Travis.


I just install this new app on one of my hosts , and the finish complete time go to 195s from 22mn ....

That a a HUGE improvement !!!
ID: 8470 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
David @ TPS

Send message
Joined: 20 Nov 08
Posts: 24
Credit: 2,561,361
RAC: 0
Message 8471 - Posted: 16 Jan 2009, 14:23:13 UTC - in response to Message 8468.  

Gonna try re-attaching with one SSE3 box. If I end up losing points I will shut it back down.


I've been running with the latest SSE3 app all day without problems. The WUs are taking slightly (minimally) longer, but it's still turning WUs around far faster than the stock app.



Pentium D working fine, just attached P-4 HT, waiting for first results to come in.

YIPPIE SKIPPIE!!!!!

Cant wait to see what one of the Quad-cores will do........
ID: 8471 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Crunch3r
Volunteer developer
Avatar

Send message
Joined: 17 Feb 08
Posts: 363
Credit: 258,227,990
RAC: 0
Message 8472 - Posted: 16 Jan 2009, 14:35:08 UTC - in response to Message 8460.  

Heres what I got for the SSEv2 zip

Optimized
fitness: -3.046272445362275

Stock
fitness: -3.046272445361065


So are these results too different?


Would be interesting to know if the app still suffers from the memleak that was 'build in' to the stock app and what's more important, if it still cuts off significant numbers when restated form a checkpoint like the stock app does (which of course makes the result invalid )



Join Support science! Joinc Team BOINC United now!
ID: 8472 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
David @ TPS

Send message
Joined: 20 Nov 08
Posts: 24
Credit: 2,561,361
RAC: 0
Message 8473 - Posted: 16 Jan 2009, 14:49:11 UTC

Pentium D & P-4 HT are validating. Quad core AMD Phenom now attached.

ID: 8473 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 8475 - Posted: 16 Jan 2009, 15:05:02 UTC - in response to Message 8472.  
Last modified: 16 Jan 2009, 15:08:17 UTC

Heres what I got for the SSEv2 zip

Optimized
fitness: -3.046272445362275

Stock
fitness: -3.046272445361065


So are these results too different?


Would be interesting to know if the app still suffers from the memleak that was 'build in' to the stock app and what's more important, if it still cuts off significant numbers when restated form a checkpoint like the stock app does (which of course makes the result invalid )



I'm pretty sure the newest versions of the stock app checkpoint to 20 digits past the decimal point so that isn't happening anymore.

*edit* And honestly, if there are bugs in the application that you know about, please post them to the code discussion forum. If you make a post about them there (instead of random snarky comments in random threads), it would actually be helpful to us so we can update the application and get rid of them.
ID: 8475 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 8476 - Posted: 16 Jan 2009, 16:40:14 UTC - in response to Message 8472.  

Heres what I got for the SSEv2 zip

Optimized
fitness: -3.046272445362275

Stock
fitness: -3.046272445361065


So are these results too different?


Would be interesting to know if the app still suffers from the memleak that was 'build in' to the stock app and what's more important, if it still cuts off significant numbers when restated form a checkpoint like the stock app does (which of course makes the result invalid )


At least my SSE3-App I linked to in this thread does not suffer from any of these shortcomings ;)
ID: 8476 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile GalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
Message 8477 - Posted: 16 Jan 2009, 16:51:52 UTC - in response to Message 8476.  

Heres what I got for the SSEv2 zip

Optimized
fitness: -3.046272445362275

Stock
fitness: -3.046272445361065


So are these results too different?


Would be interesting to know if the app still suffers from the memleak that was 'build in' to the stock app and what's more important, if it still cuts off significant numbers when restated form a checkpoint like the stock app does (which of course makes the result invalid )


At least my SSE3-App I linked to in this thread does not suffer from any of these shortcomings ;)

I for one am grateful for you making this available Cluster Physik.

I've already added it to www.arizmoon.com which I will be updating now with what I think are the 'correct' SSE2 and SSE apps.


ID: 8477 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 8478 - Posted: 16 Jan 2009, 16:58:11 UTC - in response to Message 8475.  
Last modified: 16 Jan 2009, 17:05:29 UTC

I'm pretty sure the newest versions of the stock app checkpoint to 20 digits past the decimal point so that isn't happening anymore.

*edit* And honestly, if there are bugs in the application that you know about, please post them to the code discussion forum. If you make a post about them there (instead of random snarky comments in random threads), it would actually be helpful to us so we can update the application and get rid of them.

A silent update? I just checked it and the code version on the server is still from December 3rd which saves only 6 decimal places (not even the significant ones!) in the checkpoints and exhibits the memory leak, Crunch3r mentioned (some whopping 170MB!).

Concerning the memory leak, it shows up when calling calculate_probabilities from calculate_likelyhood, if I remember correctly.
Btw., you should output the values not with "%.nf" (with n the number of decimal places after the point), but better as "%.ng", as this preserves the significant digits in the shortest printable form.
ID: 8478 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile GalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
Message 8479 - Posted: 16 Jan 2009, 17:28:40 UTC - in response to Message 8477.  
Last modified: 16 Jan 2009, 17:29:25 UTC

At least my SSE3-App I linked to in this thread does not suffer from any of these shortcomings ;)

I for one am grateful for you making this available Cluster Physik.

I've already added it to www.arizmoon.com which I will be updating now with what I think are the 'correct' SSE2 and SSE apps.

OK, I've updated www.arizmoon.com with what has been agreed in this thread as usable optimized apps for SSE, SSE2 and SSE3.

Please let me know if I'm wrong, thanks.

ID: 8479 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 8480 - Posted: 16 Jan 2009, 17:32:35 UTC - in response to Message 8478.  

I'm pretty sure the newest versions of the stock app checkpoint to 20 digits past the decimal point so that isn't happening anymore.

*edit* And honestly, if there are bugs in the application that you know about, please post them to the code discussion forum. If you make a post about them there (instead of random snarky comments in random threads), it would actually be helpful to us so we can update the application and get rid of them.

A silent update? I just checked it and the code version on the server is still from December 3rd which saves only 6 decimal places (not even the significant ones!) in the checkpoints and exhibits the memory leak, Crunch3r mentioned (some whopping 170MB!).

Concerning the memory leak, it shows up when calling calculate_probabilities from calculate_likelyhood, if I remember correctly.
Btw., you should output the values not with "%.nf" (with n the number of decimal places after the point), but better as "%.ng", as this preserves the significant digits in the shortest printable form.


I just looked at the code and i see the problem. The stock app should be updated by the end of the day.
ID: 8480 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · 10 · 11 . . . 12 · Next

Message boards : Number crunching : New faster application?

©2024 Astroinformatics Group