S39 Observation Thread

Hermann
Hermann
Joined: 17 Jan 06
Posts: 22
Credit: 220,197
RAC: 0

RE: RE: No, S38 worked

Message 26239 in response to message 26238

Quote:
Quote:
No, S38 worked fine. Since I installed S39 the errors occured. Gone back to S38 and no problems.

On your results for computer I refer to your result id's 22261473,22268751,&22270956.
The 3 (S38) results immediately preceeding the S39 switch.

Yes, you are absolutely right, sorry, my mistake.

TauCeti
TauCeti
Joined: 1 Apr 05
Posts: 16
Credit: 1,336,558
RAC: 0

S39/S38 = 0.69(!) on my

S39/S38 = 0.69(!) on my Northwood P4 2.6 HT with DDR400 (HT enabled, two alberts running)!

Awesome! Thanks a lot akosf!

archae86
archae86
Joined: 6 Dec 05
Posts: 3,158
Credit: 7,244,450,097
RAC: 1,293,922

My diverse Intel fleet has

My diverse Intel fleet has S-39 full results on all four types I own without error. Also each has completed at least one mixed S-38, S-39 result without error.

No validations yet to report.

Speedup initial observations are somewhat variable, with my P4 HT EE (original Northwood-derived type, not Prescott-derived) seeing .70 CPU time on S-39 compare to S-38, my Pentium M (Banias) seeing about .80, but my Pentium III and Pentium II seem somewhat slowed down. As the last two were the most improved by S-38, this means all four now have a more closely matched improvement relative to the project distributed science application.
[pre]
CPU S-39/S-38 S-39/dist
P4 .70 .30
Banias .80 .22
PIII 1.05 .27
PII 1.07 .28
[/pre]

My initial results from my PII and PIII suggest that S-39 actually slowed them down by 5 to 7% compared to S-38. I would be eager to hear from others running these older CPUs, which were so hugely helped by C-37 and S-38. Possibly akosf's S-39 trick is less suitable to those older machines.

Should you check my computers yourself, here are two cautions:
1. I use trux's calibrating client, and it is just kicking in with adjustments, rapidly changing from results to result--making the the summary table CPU times unsuitable for this comparison. One can get the unadjusted CPU time to the nearest second from the individual result report. (very roughly it has taken about 30 results since my original conversion to C-37 from dist for the calibrating client to hit breakeven--the P4 got there a while ago, but the others are just getting there now. With the lag time of the calibration, and further akosf version improvement, none are anywhere stable. I should start another thread to discuss this, but I feel that using the calibrating client will help reduce penalizing non-converted users and the project as a whole for this improvement.)
2. The major datafile in use has switched back and forth--making result sample selection for comparison critical. Ignoring this, or just demarking them as "long" vs. "short" is not adequate for even modestly accurate comparison.

Stef
Stef
Joined: 8 Mar 05
Posts: 206
Credit: 110,568,193
RAC: 0

S39 is 14.6% faster compared

S39 is 14.6% faster compared to S38 on a P3 1GHz..

Superb again, akosf!

Ziran
Ziran
Joined: 26 Nov 04
Posts: 194
Credit: 625,122
RAC: 1,058

First result on my Sempron

First result on my Sempron 3000+ validated fine against 2 windows host running the standard client.
S39/S38=0.93

Then you're really interested in a subject, there is no way to avoid it. You have to read the Manual.

Akos Fekete
Akos Fekete
Joined: 13 Nov 05
Posts: 561
Credit: 4,527,270
RAC: 0

RE: [pre] CPU S-39/S-38

Message 26244 in response to message 26241

Quote:
[pre]
CPU S-39/S-38 S-39/dist
P4 .70 .30
Banias .80 .22
PIII 1.05 .27
PII 1.07 .28
[/pre]My initial results from my PII and PIII suggest that S-39 actually slowed them down by 5 to 7% compared to S-38. I would be eager to hear from others running these older CPUs, which were so hugely helped by C-37 and S-38. Possibly akosf's S-39 trick is less suitable to those older machines.

Yes! I think these values are really correct.
My comment:

Northwood was in trouble at Taylor series (S38), but S39 has a new interpolation method with less multiplications and subtractions. So it is better for its pipelined fpu.

Banias is Banias, it solved the excersise.

PII and PIII come from the same root, so they are to fight same problem. Their L1 data cache size is just 16kB, but the new method requires more (~33kB).

Thanks for the results.

Rojer
Rojer
Joined: 2 Apr 05
Posts: 23
Credit: 49,400,244
RAC: 0

RE: RE: [pre] PII and

Message 26245 in response to message 26244

Quote:
Quote:

[pre]

PII and PIII come from the same root, so they are to fight same problem. Their L1 data cache size is just 16kB, but the new method requires more (~33kB).

Thanks for the results.


I've replaced s38 with s39 in PII,and err rised.That's the reason?

2006-3-22 20:41:06|Einstein@Home|Aborting result r1_1346.0__1716_S4R2a_2: exceeded disk limit: 107477607.000000 > 100000000.000000
2006-3-22 20:41:06|Einstein@Home|Unrecoverable error for result r1_1346.0__1716_S4R2a_2 (Maximum disk usage exceeded)
2006-3-22 20:41:44||Rescheduling CPU: application exited
2006-3-22 20:41:44|Einstein@Home|Computation for result r1_1346.0__1716_S4R2a_2 finished
2006-3-22 20:41:45|Einstein@Home|Starting result r1_1346.0__1698_S4R2a_1 using albert version 437
2006-3-22 21:01:08|Einstein@Home|Aborting result r1_1346.0__1698_S4R2a_1: exceeded disk limit: 104642066.000000 > 100000000.000000
2006-3-22 21:01:08|Einstein@Home|Unrecoverable error for result r1_1346.0__1698_S4R2a_1 (Maximum disk usage exceeded)
2006-3-22 21:01:33||Rescheduling CPU: application exited
2006-3-22 21:01:33|Einstein@Home|Computation for result r1_1346.0__1698_S4R2a_1 finished

Wish you can understand my English:)

B52
B52
Joined: 19 Feb 05
Posts: 45
Credit: 273,899
RAC: 0

RE: My initial results from

Message 26246 in response to message 26241

Quote:
My initial results from my PII and PIII suggest that S-39 actually slowed them down by 5 to 7% compared to S-38. I would be eager to hear from others running these older CPUs, which were so hugely helped by C-37 and S-38. Possibly akosf's S-39 trick is less suitable to those older machines.

I had a VERY VERY small marginal improvement on my PIII 801 mhz, but NOT a slowdown.

My PIII host can be found here

http://einsteinathome.org/host/564412/tasks

The last 2 results in the 3,9xx range are with S39 ONLY-

archae86
archae86
Joined: 6 Dec 05
Posts: 3,158
Credit: 7,244,450,097
RAC: 1,293,922

RE: I've replaced s38 with

Message 26247 in response to message 26245

Quote:


I've replaced s38 with s39 in PII,and err rised.That's the reason?

2006-3-22 20:41:06|Einstein@Home|Aborting result r1_1346.0__1716_S4R2a_2: exceeded disk limit: 107477607.000000 > 100000000.000000

It appears that by bad luck your disk usage exceeded the limit you set in your preferences at this point. You may wish to go to the general preferences link on your account page and adjust the disk limits, assuming your system has room for more. You appear to be allowing 100 Megabytes.

Santas little helper
Santas little helper
Joined: 11 Feb 05
Posts: 36
Credit: 10,001,864
RAC: 25,481

RE: I've replaced s38 with

Message 26248 in response to message 26245

Quote:


I've replaced s38 with s39 in PII,and err rised.That's the reason?

2006-3-22 20:41:06|Einstein@Home|Aborting result r1_1346.0__1716_S4R2a_2: exceeded disk limit: 107477607.000000 > 100000000.000000
2006-3-22 20:41:06|Einstein@Home|Unrecoverable error for result r1_1346.0__1716_S4R2a_2 (Maximum disk usage exceeded)
2006-3-22 20:41:44||Rescheduling CPU: application exited
2006-3-22 20:41:44|Einstein@Home|Computation for result r1_1346.0__1716_S4R2a_2 finished
2006-3-22 20:41:45|Einstein@Home|Starting result r1_1346.0__1698_S4R2a_1 using albert version 437
2006-3-22 21:01:08|Einstein@Home|Aborting result r1_1346.0__1698_S4R2a_1: exceeded disk limit: 104642066.000000 > 100000000.000000
2006-3-22 21:01:08|Einstein@Home|Unrecoverable error for result r1_1346.0__1698_S4R2a_1 (Maximum disk usage exceeded)
2006-3-22 21:01:33||Rescheduling CPU: application exited
2006-3-22 21:01:33|Einstein@Home|Computation for result r1_1346.0__1698_S4R2a_1 finished


edit: oh, archae86 was faster ...

Greetings, Santas little helper

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.