TAS Note: Quad Challenge

[salvorhardin]

119497 - Q9400: 425X8=3.4GHz; 4GB DDR2-800 @ 850 1.8V; Gigabyte EP45-UD3P; Radeon 4830 @ 700/910 and Windows 7.

 

[petrusbroder]

Originally posted by: Rudy Toody
TeAm CPUs on Ibercivis.

Yeah, I have these numbers, but they do not tell the real CPU frequency and FSB ... and that may be of some importance when I'll analyse the data ...

 

[petrusbroder]

Originally posted by: dajeepster

Originally posted by: petrusbroder
Yeah I had forgotten to do that on that computer --- so it had only WUs for 0.5 days - and it crunched through them faster than expected.
Won't happen again ...

grea...er.. i mean... too bad

LoL

Learning something new is really like winning!

 

[biodoc]

119225: Q6600@3.24 GHz; FSB=360 x 9; 4GB DDR2 800; OS= 64 bit ubuntu 9.04

Stock is 2.4 GHz; FSB=266 x 9

Is that third column supposed to add up to 111,000 points?

 

[dajeepster]

Originally posted by: biodoc
119225: Q6600@3.24 GHz; FSB=360 x 9; 4GB DDR2 800; OS= 64 bit ubuntu 9.04

Stock is 2.4 GHz; FSB=266 x 9

Is that third column supposed to add up to 111,000 points?

what graphics card are you running?.. i see that you have joined collatz conjecture with a first day ppd of 16,152... not bad

 

[biodoc]

Originally posted by: dajeepster

Originally posted by: biodoc
119225: Q6600@3.24 GHz; FSB=360 x 9; 4GB DDR2 800; OS= 64 bit ubuntu 9.04

Stock is 2.4 GHz; FSB=266 x 9

Is that third column supposed to add up to 111,000 points?

what graphics card are you running?.. i see that you have joined collatz conjecture with a first day ppd of 16,152... not bad

The above rig has a 9800GTX+ and is running gpugrid.

My other 2 rigs are now running collatz conjecture (another 9800GTX+ and the GTX295).

The validator on collatz has stopped running so I have a ton of WUs pending (50K+ points in 24 h). By my calculations, both cards together should do around 95K ppd.

Apparently the admin for the project is out hunting Bambi this weekend!:shocked:

 

[petrusbroder]

Originally posted by: biodoc
119225: Q6600@3.24 GHz; FSB=360 x 9; 4GB DDR2 800; OS= 64 bit ubuntu 9.04

Stock is 2.4 GHz; FSB=266 x 9

Is that third column supposed to add up to 111,000 points?

No, that was an addition error; I forgot to empty the "memory" of the calculator ...
Sorry. Corrected the post.

 

[petrusbroder]

Stats after 4½ days:

Computer ___ Credits ___ Credits ____ Diff ___ TAS-mate _______ CPU _______ GHz
_________ __ Oct 20 ____ Oct 25 ___________________________________________
_118825 ____ 52 976 ____ 60 256 ___ 7 280 __ Rudy Toody ___ Ph II 940BE __ 3.30
_118989 _____ 4 210 _____ 9 046 ___ 4 836 __ Philippart _____ C2Q9450 _____ ??
_119045 ____ 12 304 ____ 26 508 __ 14 204 __ dajeepster _____ i7 860 ______ 2.80
_119121 _____ 4 289 ____ 18 866 __ 14 577 __ Petrusbroder ___ i7 920 ______ 3.30
_119141 _____ 4 676 ____ 11 306 ___ 6 630 __ Neurodog _____ Ph II 940 _____ ??
_119225 _____ 4 055 ____ 11 759 ___ 7 704 __ biodoc _______ C2Q6600 _____ 3.24
_119497 ____ 17 401 ____ 22 064 ___ 4 663 __ Salvorhardin __ C2Q9400 _____ 3.40
Total _______ 99 911 ___ 159 805 __ 59 894 __ TAS-total

Notes:
Ph = AMD Phenom, C2Q= Intel Core2 Quad, i7 = Intel i7, ?? = CPU frequency not stated
I have taken the data from Ibercivis' site and from posts in this thread.

Interesting comparison!
It seems to me that Philippart and salvorhardin are crunching an other project at the same time as Ibercivis - if you compare to biodoc's numbers.
Also it seems that Neurodog's CPU is not OC'ed, while we know that Rudy Toody has OC'ed his CPU from 3.0 GHZ to 3.3 GHZ, which fits in quite well.
It is also very clear whar a good cache of WUs can do: Petrus' cache was too small (his computer was running empty for 10 - 11 hours during a maintenence outage) while dajeepster's was crunching full ahead: dajeepster came up ahead in spite of going 500 MHz slower.

Crunch on, TAS-mates!

 

[Rudy Toody]

I find it interesting that the 8-thread, 4-core quads produce double what the 4-core quads produce.

Rudy Toody 1820 credits/core
Philippart 1209c/c
dajeepster 1775.5c/thread
Petrusbroder 1822.125c/t
Neurodog 1657.5c/c
biodoc 1926c/c
Salvorhardin 1165.75c/c

biodoc has the best performance so far.

Edit: to call it threads instead of cores for the I's.

 

[biodoc]

Originally posted by: Rudy Toody
I find it interesting that the 8-core quads produce double what the 4-core quads produce.

Rudy Toody 1820 credits/core
Philippart 1209c/c
dajeepster 1775.5c/c
Petrusbroder 1822.125c/c
Neurodog 1657.5c/c
biodoc 1926c/c
Salvorhardin 1165.75c/c

biodoc has the best performance so far.

Now I'm really confused because I have the oldest chip!

I'm still amazed that the "i" series chips are producing about 2x more than the rest of us. The "i" series chips only have 4 physical cores. They do have hyperthreading, however so they run 2 threads (apps) per core. So the list RT posted should be per thread rather than per core? If it's per core, then the "i" guys numbers should be doubled?

I wouldn't be suprised by a 20-30% increase in performance due to caching and other factors I really don't understand, but a 2-fold increase in performance I don't understand.

Would the "i" guys be interested in testing ppd with hyperthreading turned off?

 

[Rudy Toody]

Originally posted by: biodoc

Originally posted by: Rudy Toody
I find it interesting that the 8-core quads produce double what the 4-core quads produce.

Rudy Toody 1820 credits/core
Philippart 1209c/c
dajeepster 1775.5c/c
Petrusbroder 1822.125c/c
Neurodog 1657.5c/c
biodoc 1926c/c
Salvorhardin 1165.75c/c

biodoc has the best performance so far.

Now I'm really confused because I have the oldest chip!

I'm still amazed that the "i" series chips are producing about 2x more than the rest of us. The "i" series chips only have 4 physical cores. They do have hyperthreading, however so they run 2 threads (apps) per core. So the list RT posted should be per thread rather than per core? If it's per core, then the "i" guys numbers should be doubled?

I wouldn't be suprised by a 20-30% increase in performance due to caching and other factors I really don't understand, but a 2-fold increase in performance I don't understand.

Would the "i" guys be interested in testing ppd with hyperthreading turned off?

I'm all for a special test with HT turned off. My quad also had some PrimeGrid and Cosmology@Home WUs that I picked up when I ran out of Ibercivis WUs, so my numbers are a tad low.

 

[Philippart]

my quad doesn't run 24/7...

 

[petrusbroder]

Originally posted by: biodoc

Now I'm really confused because I have the oldest chip!
...
Would the "i" guys be interested in testing ppd with hyperthreading turned off?

Well, the Q6600 is a very good chip, IMHO one of the best Intel ever has produced. It was "king of the hill" for quite a long time probably due to its efficiency. If it is the cache, the pipeline or some very well optimized instruction set I can't even guess...

Testing without hyperthreading? Yes, at least one of them would ... but after the race ...
The outcome is not given - it could be the same efficiency, it could be half ...

Edit: Comment on Q6600.

 

[Neurodog]

I don't get it...

My X4 940 is OC to 3.4Ghz

Disappointing


EDIT: Very Disappointing!

 

[petrusbroder]

Originally posted by: Neurodog
I don't get it...

My X4 940 is OC to 3.4Ghz

Disappointing


EDIT: Very Disappointing!

Hmmm, <petrus thinks ...>

BIOS: Power saving turned off?
BIOS: Smartfan turned off?
BIOS: other settings?
System: Temperatures - CPU, Mobo ...?
System: All cores on line crunching?
In BOINC preferences: 100% of processors?
In BOINC preferences: 100% CPU-time
In BOINC preferences: "Computing allowed while computer in use [box checked]?

I hope this could help ...

 

[Philippart]

also every app has other intel/amd preferences

 

[petrusbroder]

Stats after 6 days:

Computer ___ Credits ___ Credits ____ Diff ___ TAS-mate
_________ __ Oct 20 ____ Oct 26 _______________________
_118825 ____ 52 976 ____ 62 277 ___ 9 310 __ Rudy Toody
_118989 _____ 4 210 ____ 10 883 ___ 6 673 __ Philippart
_119045 ____ 12 304 ____ 29 380 __ 17 076 __ dajeepster
_119121 _____ 4 289 ____ 23 671 __ 19 382 __ Petrusbroder
_119141 _____ 4 676 ____ 14 081 ___ 9 405 __ Neurodog
_119225 _____ 4 055 ____ 14 636 __ 10 581 __ biodoc
_119497 ____ 17 401 ____ 23 046 ___ 5 645 __ Salvorhardin
Total _______ 99 911 ___ 177 974 __ 78 063 __ TAS-total


Looks very good! :sun:
Crunch on, TAS-mates!

 

[petrusbroder]

BTW: When should the race end? Halloween has different dates - I think - in Europe and the USA.
In Europe it Halloween ends Saturday, Oct. 31, 24:00.
Should that be the time I get the stats?
It does not matter for me att all, as long as we all know when the dead line for the stats is ...

 

[Rudy Toody]

Originally posted by: petrusbroder
BTW: When should the race end? Halloween has different dates - I think - in Europe and the USA.
In Europe it Halloween ends Saturday, Oct. 31, 24:00.
Should that be the time I get the stats?
It does not matter for me att all, as long as we all know when the dead line for the stats is ...

`That would be perfect! That is when the Great Pumpkin rises.

 

[petrusbroder]

So be it: the race ends as close as it can be to 24:00 October 31, 2009.

 

[GLeeM]

Originally posted by: biodoc
Would the "i" guys be interested in testing ppd with hyperthreading turned off?

OK!

ID: 120617

i7 920 D0
HT off
200x21 = 4.2 @ 1.296v
temps C 71, 70, 69, 66 (nice! I never had HT off before )
Win7 64

The memory (6GB) is at 1200. I had increased Vcore several times (BSODs) and the last time also decreased ram and uncore and QPI. It passed 5 tests of LinX finally. I might increase those after a bit, maybe it was just Vcore that needed increasing that last time.

Now if I can only get enough work! I'll have to read through this thread again to see what the trick is

EDIT: Is there a way to see if my WUs are finishing correctly and not bombing out on this OC?
Also, F@H GPU client is running on GTX275 using about 4% of one core.
Peter let me know if you can't see this computer.

 

[biodoc]

Originally posted by: GLeeM

Originally posted by: biodoc
Would the "i" guys be interested in testing ppd with hyperthreading turned off?

OK!

ID: 120617

i7 920 D0
HT off
200x21 = 4.2 @ 1.296v

WOW!!:shocked:

temps C 71, 70, 69, 66 (nice! I never had HT off before )
Win7 64

The memory (6GB) is at 1200. I had increased Vcore several times (BSODs) and the last time also decreased ram and uncore and QPI. It passed 5 tests of LinX finally. I might increase those after a bit, maybe it was just Vcore that needed increasing that last time.

Now if I can only get enough work! I'll have to read through this thread again to see what the trick is

EDIT: Is there a way to see if my WUs are finishing correctly and not bombing out on this OC?

I don't think so. There's no message board either.

Also, F@H GPU client is running on GTX275 using about 4% of one core.
Peter let me know if you can't see this computer.

I can see it!!:shocked:

Measured floating point speed 4291.61 million ops/sec
Measured integer speed 12614.25 million ops/sec


I tried the Q6600 version of "hyperthreading". I call it "hyperapping". I told boinc that I have 8 cores by editing the cc_config.xml file.


It looked cool with 8 apps running.....but no more ppd.

EDIT: formatting (3 times)

 

[Rudy Toody]

I am hyperapping 12 threads as we speak. Ibercivis can't keep up!

Edit: biodoc--your credits yesterday are the second highest you have had. Maybe hyperapping does make a difference. My credits per day are going up.

Edit2: Because my hyperapping has raised the cpu temp a bit, I have implemented a program of occasional hyperventilating.

 

[dajeepster]

Originally posted by: Rudy Toody
I am hyperapping 12 threads as we speak. Ibercivis can't keep up!

Edit: biodoc--your credits yesterday are the second highest you have had. Maybe hyperapping does make a difference. My credits per day are going up.

Edit2: Because my hyperapping has raised the cpu temp a bit, I have implemented a program of occasional hyperventilating.

i wonder what the result would be if I hyperapped my "i"?
mmmm.. 16 threads :evil:

 

[Rudy Toody]

Originally posted by: dajeepster

Originally posted by: Rudy Toody
I am hyperapping 12 threads as we speak. Ibercivis can't keep up!

Edit: biodoc--your credits yesterday are the second highest you have had. Maybe hyperapping does make a difference. My credits per day are going up.

Edit2: Because my hyperapping has raised the cpu temp a bit, I have implemented a program of occasional hyperventilating.

i wonder what the result would be if I hyperapped my "i"?
mmmm.. 16 threads :evil:

In cc_config.xml, in the <options> section, <ncpus>16</ncpus> will do it.