32-bit Linux a tiny bit slower than before
64-bit Linux 20% or 10 minutes slower than v0.07 (?) but much better than 0.10

Hard to tell at the moment until there is new work to process....

Let me know if the performance issues have been fixed.

thanks!

Hi there ... Im probs on the wrong thread.
I detached from the project to get rid of the sse3 app and load your standard new app.
Im running xp64 on my e4500 and now the WU's are taking about 59 mins for 39 credits compaired to 41 mins for 40 credits a week ago.

Anyone else having the same problems ?

____________

@ Travis: Why don't you just put out test runs instead of screwing with everyones computers? You're the one who stresses the need for testing the new apps. 6 version in a few days is too many.
____________
Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.

Me likes testing new versions :-)

@ Travis: Why don't you just put out test runs instead of screwing with everyones computers? You're the one who stresses the need for testing the new apps. 6 version in a few days is too many.

@ Travis: Why don't you just put out test runs instead of screwing with everyones computers? You're the one who stresses the need for testing the new apps. 6 version in a few days is too many.

this project is still alpha, which means the whole thing is a test - so, in my opinion, no need for tests inside a test...

On 1 sample, I have seen a 5% faster time with .12. Not sure if that is in the noise or not.

Jim

EDIT: Running Mac OS 10.5.6 using BOINC 6.2.18 on a 2 dual core MacPro.

Hy,

Ubuntu 64 bits version 0.12 around 5400 s 56 minutes much better than 0.10.
Still 0.07 the best in time around 42 minutes for my computer.

On vista 32 bits no change around same time 0.07, 0.10 and 0.12.

On my Mac Pro I have not seen that v0.12 is different from 0.10 ...

Roughly 23-24 minutes ... because there is some variation in the run times it is hard to pin down ... along with the fact that you don't leave much history in the database (I know, other problems, but it makes it hard to go back and calculate a real time average) ...

All in all, I would say that there has been no change ...
____________

0.12 is running at about 57 minutes per woek unit, 0.07 was running at about 55 minutes per work unit. That's on an Athlon XP 3000+ clocked at 2.17 GHZ running Kubuntu 8.04.1

HTH,

Rob.

On AMD 285 Linux Fedora Core 6, times for v0.07 and v0.12 are the same, no change at all.
____________

Travis, do you know why, this host :

http://milkyway.cs.rpi.edu/milkyway/results.php?hostid=41246 ( Linux 64 )

has granted credit = 0 ???

I am using the offcial app V0.12 !

I try to detach and reattach but same problem. I try with Speedimic v0.12 optimized application but still no point ?

I put this machine in NMW until you can find what happens !

sp0wn

edit :

All those Wus are 0 Credit granted !

Those ones are using official V0.12 app :


http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63033709

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63033710

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63033711

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63033712

[url]http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63033713


Those one are using Speedimic SSE4.1_X86_64 optimized app :

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63040284

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63042064

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63042087

only that one receive credit !

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63042065

Before asking, this machine is not O/C....

Hi,

Opteron 170, 2,0 GHz, XP 32 : 3300 sec average, v10 and V12
Core2Duo 4300, 2,0 GHz, Ubuntu 64 : 4600 sec average V12 (7300 sec average V10)
Core2Quad 6600, 2,4 HZz, Ubuntu 64 : 3900 sec average V12 (6200 sec average V10)

So, V12 is undoubtly faster.
From memories, I would say say V12 is faster than V7 under Windows XP 32.
For sure, and completion times show it, Linux 64 V12 is undoubtly slower than V7 (about 30% slower).

Other hint for compiling : I doubt that my old Opteron (server version of the socket 939 Athlon 64 x2) could beat my Quad 6600 (both @ stock clock). I know, architecture are not the same, but the gap is not normal. The Linux 64 version lacks optimizations.

Good luck

V.7 and v.12 are running the same on OS X for mini core duo and MacBook core 2 duo. This is back to twice as fast as the intervening versions.

Travis, do you know why, this host :

http://milkyway.cs.rpi.edu/milkyway/results.php?hostid=41246

has granted credit = 0 ???

I am using the offcial app V0.12 !

I try to detach and reattach but same problem. I try with Speedimic v0.12 optimized application but still no point ?

I put this machine in NMW until you can find what happens !

sp0wn

Don't know , it's since this morning !

I did not change anything on the host !

Only when I started to get 0 pts , I tried differents app but look like a kind of bug.

Anf it happens only on that computer ...

Travis, do you know why, this host :

http://milkyway.cs.rpi.edu/milkyway/results.php?hostid=41246

has granted credit = 0 ???

I am using the offcial app V0.12 !

I try to detach and reattach but same problem. I try with Speedimic v0.12 optimized application but still no point ?

I put this machine in NMW until you can find what happens !

sp0wn

Hi Sp0wn

Check the result ;)

Validate state is Invalid

As I saw that some compuer are using optimized V0.12 app without any problem.

I reinstall the Speedimic SSE4.1 and crunch some wus and they get granted ....

I don t know what happen ??!!

Just come in after I read about the news, that you finally start granting no credits for bogus results, to test the new stock apps, first one was 0.10, than 0.12, both on my C2Q9450 @3.2GHz, ubuntu64, BOINC 6.4.5.
I did 3 0.10 that lasted 4200 seconds, the 0.12 lasted 2600 - 2700 seconds, seems like a good improvement.
I have no idea how fast they would have been with 0.7.
http://milkyway.cs.rpi.edu/milkyway/results.php?hostid=24253
____________
Grüße vom Sänger

As I saw that some compuer are using optimized V0.12 app without any problem.

I reinstall the Speedimic SSE4.1 and crunch some wus and they get granted ....

I don t know what happen ??!!

As I saw that some compuer are using optimized V0.12 app without any problem.

I reinstall the Speedimic SSE4.1 and crunch some wus and they get granted ....

I don t know what happen ??!!

If you're talking about the Linux apps: yes, he recompiled them with the appropriate flags!
See here

Yep , but you see here , I still have some bad wus ... without changing anything....


I think that the new assimilator/validator has some problem ... for validating wus...

Using 4 samples, I am seeing a consistent 5% improvement in performance on my 2,66Mhz, 2 dual core MacPro. That equate to about a minute plus for me.

Jim

Yep , but you see here , I still have some bad wus ... without changing anything....

I think that the new assimilator/validator has some problem ... for validating wus...

Other hint for compiling : I doubt that my old Opteron (server version of the socket 939 Athlon 64 x2) could beat my Quad 6600 (both @ stock clock). I know, architecture are not the same, but the gap is not normal. The Linux 64 version lacks optimizations.

Travis, do you know why, this host :

http://milkyway.cs.rpi.edu/milkyway/results.php?hostid=41246 ( Linux 64 )

has granted credit = 0 ???

I am using the offcial app V0.12 !

I try to detach and reattach but same problem. I try with Speedimic v0.12 optimized application but still no point ?

I put this machine in NMW until you can find what happens !

sp0wn

edit :

All those Wus are 0 Credit granted !

Those ones are using official V0.12 app :


http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63033709

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63033710

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63033711

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63033712

[url]http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63033713


Those one are using Speedimic SSE4.1_X86_64 optimized app :

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63040284

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63042064

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63042087

only that one receive credit !

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=63042065

Before asking, this machine is not O/C....

It looks like you might have some kind of problem with that machine if you're consistently getting bad results from BOTH speedimic's and the stock application. The new validator will mark WUs invalid if they return a likelihood that's not possible to be calculated by the application.

i TOO JUST HAD 2 WU's out of the latest batch of 32 come back as invalid - 0 credit granted. Those were only the first 2 I have come across. I will keep an eye on things and see if it was just a hiccup or something else.

along with the fact that you don't leave much history in the database (I know, other problems, but it makes it hard to go back and calculate a real time average) ...

no, only realized where I was after I clicked "post reply" All is well :-)

I don t know, I have sometime some invalid Wus ... in the middle of a bunch of good ones ....

I don t understand what happens !!

I get today only stock app 0.10 for windows 32bit.
They validate, but the application page tells me I should get 0.13.
Any explanation why this happens?

edit:
sample result
resultid=63393451
____________
Matthias

Nice, a release or 2 per day! Awesome, reminds me of Windows!!!

Based on my hosts:

Opteron 170 (XP 32) RAC estimation: 2085 (v0.7) / 2070 (v12) => 1% drop
Core2Duo 4300 (Ubuntu 64) RAC estimation: 1994 (v0.7) / 1475 (v12) => 26% drop
Core2Duo 6600 (Ubuntu 64) RAC estimation: 2385 (v0.7) / 1787 (v12) => 25% drop
Core2Quad 6600 (Ubuntu 64) RAC estimation: 4742 (v0.7) / 3564 (v12) => 25% drop

Houston, I think we have a problem with the Linux64 version...

Nice, a release or 2 per day! Awesome, reminds me of Windows!!!

0.13 is giving me times of 36 - 37 minutes on X86-64/XP Pro 64-bit, which is what 0.7 was doing.

Nice one. :-)

My MacBookPro C2D T7700 is finally back down to 3300-3400 cpusec/wu, like it was back on .07 versioon, and my iBook PPC G4 1.33 MHz is still hanging in there around a bit over 9000 cpusec/wu.

Having said that, My MBP should be down around 1000 cpusec/wu, based on the comparison between the same two computers running the optimized v8 for SETI. Is the SSSE3, and SSE2 in place on the MW 0.12 app? I'm not a programmer, but it seems to me the optimization for the MBP should give me the same results as it does for SETI, and approach the results the MS XP folks are enjoying.

C
____________
Team MacNN

My MacBookPro C2D T7700 is finally back down to 3300-3400 cpusec/wu, like it was back on .07 versioon, and my iBook PPC G4 1.33 MHz is still hanging in there around a bit over 9000 cpusec/wu.

Having said that, My MBP should be down around 1000 cpusec/wu, based on the comparison between the same two computers running the optimized v8 for SETI. Is the SSSE3, and SSE2 in place on the MW 0.12 app? I'm not a programmer, but it seems to me the optimization for the MBP should give me the same results as it does for SETI, and approach the results the MS XP folks are enjoying.

C

My MacBookPro C2D T7700 is finally back down to 3300-3400 cpusec/wu, like it was back on .07 versioon, and my iBook PPC G4 1.33 MHz is still hanging in there around a bit over 9000 cpusec/wu.

Having said that, My MBP should be down around 1000 cpusec/wu, based on the comparison between the same two computers running the optimized v8 for SETI. Is the SSSE3, and SSE2 in place on the MW 0.12 app? I'm not a programmer, but it seems to me the optimization for the MBP should give me the same results as it does for SETI, and approach the results the MS XP folks are enjoying.

C

Ah! A Team MacNN mate. I gave 10M credits to that team before moving on.

But I will say this about the intel mac apps: They should all have at least SSE3 optimization. Even the very first Yonah (Core Solo) that Macs ever had, are SSE3. So theoretically, the STOCK OSX Intel app should be at least as fast as an optimized SSE3 app for Windows. The idea is that there is no need to make it compatible with any lesser chip, like you have to with the Win apps. So while stock Win apps are handicapped by having to support Pentium I, stock OSX Intel apps are not.

Thanks for this. Someone had said that it was possible for an Intel mac to have SSE2. In the next release I'll bump all the x86_64 macs to SSE3 (assuming it performs better than SSE2).

You can bump i686-apple-darwin too. The earliest Intel processor in Macs was the Core Duo/Solo, which supported SSSE3 (notice the three S's, not two). So you can enabled all the (S)SSEx optimizations for all Intel Macs.

ok scratch compiling the i686 with sse3 :)

You can bump i686-apple-darwin too. The earliest Intel processor in Macs was the Core Duo/Solo, which supported SSSE3 (notice the three S's, not two). So you can enabled all the (S)SSEx optimizations for all Intel Macs.

I've just tried an SSSE3 compile and get the following message
Fatal Error: This program was not built to run on the processor in your system.
The allowed processors are: Intel(R) Core(TM) Duo processors and compatible Intel processors with supplemental Streaming SIMD Extensions 3 (SSSE3) instruction support.

So it looks like my T2500 CPU doesn't support SSSE3 :(

So i should be able to move the x86_64 apple versions to SSSE3? (thats 3 s). How would I go about doing that?

The first line of Intel Macs from 2006 used the Core Solo/Duo with:
All models support: MMX, SSE, SSE2, SSE3, EIST, XD bit

Only the following Macs with Core 2 Solo/Duo has SSSE3 and Intel 64.
Thanks to the stupid labeling by intel, its easy to confuse.

The first line of Intel Macs from 2006 used the Core Solo/Duo with:
All models support: MMX, SSE, SSE2, SSE3, EIST, XD bit

Only the following Macs with Core 2 Solo/Duo has SSSE3 and Intel 64.
Thanks to the stupid labeling by intel, its easy to confuse.

I explored the differences among the several SSE flavors starting here.

SSSE3 is probably useless for scientific applications.

HTH

____________

I explored the differences among the several SSE flavors starting here.

SSSE3 is probably useless for scientific applications.

HTH

yes you are right with one small exception, scientific applications with supported code and structure for SSE3 and SSSE3 can get really very much and good performance...