Windows S5R3 SSE power App 4.36 available

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,522
Credit: 739,062,694
RAC: 1,269,208

...and btw, no matter what is

Message 79645 in response to message 79643

...and btw, no matter what is displayed in the messages dialog, BOINC will in fact use the new app, 4.36, that you just installed.

CU
Bikeman

EDIT: Conan was typing faster than me :-)

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2,143
Credit: 2,963,899,025
RAC: 711,776

And you can check that what

And you can check that what Conan and Bikeman both said is true, by using Task Manager to see what's really running on your machine.

Mark Henderson
Mark Henderson
Joined: 19 Feb 05
Posts: 34
Credit: 39,029,369
RAC: 2,909

I just finished more

I just finished more workunits on my 4 different computers with 4.36 vs. 4.32 vs. 4.26

AMD X2 4800+, 1 gb. ram, Win. XP
44411 sec. with 4.26
24551 sec. with 4.32
22000 sec. with 4.36 ;slightly faster

Intel Quad 6600, 4 gb. ram, Win. XP 64 bit.
30502 sec. with 4.26
22749 sec. with 4.32
15000 sec. with 4.36 ;much faster

AMD Athlon 2800, 512 mb. ram, Win. 2000
56253 sec. with ver. 4.26
40352 sec. with ver. 4.32
45000 sec. with ver. 4.36 ;a little slower

Intel P4 Prescott 3.5 with HT on
88000 sec. with ver. 4.26
57000 sec. with ver. 4.32
55000 sec. with ver. 4.36 ;slightly faster

It seems that machines with more L2 cache benefited more from 4.36

Ed1934158
Ed1934158
Joined: 10 Nov 04
Posts: 62
Credit: 14,481,483
RAC: 0

Thank you guys! Unfortunately

Thank you guys!
Unfortunately I can't provide much statistics since just began with beta, but here is comparison from my dual core laptop (GenuineIntel
Intel(R) Core(TM)2 Duo CPU T5450 @ 1.66GHz [x86 Family 6 Model 15 Stepping 13]):
1. normal Einsten application:
44196s
43214s
43687s
49022s
47738s
46170s
45488s
2. version 4.32:
37753s
37028s
38287s
37412s
Now we'll see what 4.36 brings!
Oh yes... Windows Vista

Quote:
It seems that machines with more L2 cache benefited more from 4.36


Or maybe 64bit windows version?

Conan
Conan
Joined: 19 Jun 05
Posts: 172
Credit: 8,380,252
RAC: 9,120

RE: I just finished more

Message 79649 in response to message 79647

Quote:

I just finished more workunits on my 4 different computers with 4.36 vs. 4.32 vs. 4.26

AMD X2 4800+, 1 gb. ram, Win. XP
44411 sec. with 4.26
24551 sec. with 4.32
22000 sec. with 4.36 ;slightly faster

Intel Quad 6600, 4 gb. ram, Win. XP 64 bit.
30502 sec. with 4.26
22749 sec. with 4.32
15000 sec. with 4.36 ;much faster

AMD Athlon 2800, 512 mb. ram, Win. 2000
56253 sec. with ver. 4.26
40352 sec. with ver. 4.32
45000 sec. with ver. 4.36 ;a little slower

Intel P4 Prescott 3.5 with HT on
88000 sec. with ver. 4.26
57000 sec. with ver. 4.32
55000 sec. with ver. 4.36 ;slightly faster

It seems that machines with more L2 cache benefited more from 4.36

The times you are quoting are they for the Peak, the Trough, average between the two or an average over all work units?

You might be quoting a peak time for your AMD 2800 for 4.36 but heading down to a trough time from 4.32.
If your Athlon 2800 supports SSE instructions then it should be faster overall than 4.32.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,872
Credit: 117,918,607,957
RAC: 34,539,335

RE: .... AMD Athlon 2800,

Message 79650 in response to message 79647

Quote:

....
AMD Athlon 2800, 512 mb. ram, Win. 2000
56253 sec. with ver. 4.26
40352 sec. with ver. 4.32
45000 sec. with ver. 4.36 ;a little slower
....

It seems that machines with more L2 cache benefited more from 4.36

You are overlooking the cyclic nature of crunch times. There are several recent high volume threads in this forum that have discussed this in considerable detail. Basically, the crunch time varies considerably with the sequence number of the task and the period of this variation is dependent on the square of the data frequency times a constant.

Your AMD Athlon 2800 is currently crunching data whose skygrid file has a frequency of 840Hz. Using this, the period of the cyclic variation comes out to 144.2 This means there will be peaks in runtimes (slow crunching) for tasks with seq#s around 0, 144, 288, 433, ... and troughs in runtimes (faster crunching) for tasks with seq#s around 72, 216, 361, 505, ....

To see the seq# of a task just look at the name. Here it is highlighted for the last task done with 4.32
h1_0835.55_S5R3__235_S5R3b_0

The seq# of 235 isn't far away from the trough at 216 so that task is expected to crunch reasonably quickly.

Your first 4.36 task is this one
h1_0835.60_S5R3__291_S5R3b_1

The seq# of 291 is very close to the peak at 288 and that is the explanation of why there appears to be a slowdown. You will also have slow crunching for the current task since it has a seq# of 290. The one after that will be faster since it has a seq# of 279 - ie starting to move away from the peak and down towards a trough. As you move further down the seq#s, you will start to see the real performance increase of the latest app.

If you are going to speculate as to why one host has apparently sped up whilst another has apparently slowed down, you might like to read some of the explanations of this behaviour in the recent high volume threads. The thread titles are quite indicative. A little reading will save you from jumping to the wrong conclusions :).

Cheers,
Gary.

rroonnaalldd
rroonnaalldd
Joined: 12 Dec 05
Posts: 116
Credit: 537,221
RAC: 0

Memory for 4.36 changed

Memory for 4.36 changed between 40MB and 45 MB each for ram and virtual.
Starting memory for einstein task i don't know, i was to slowly *rofl*

hotze33
hotze33
Joined: 10 Nov 04
Posts: 100
Credit: 368,387,400
RAC: 0

I have finished some results

I have finished some results on my Q6600 @ 3.28GHz.

4.32
Frequency : 835.4
Period of task cycle = 144.2
Number of points = 9
Minimum runtime in data = 15827.81
Maximum runtime in data = 20745.59
Estimated peak runtime = 20761
Estimated average runtime = 17468
Estimated trough runtime = 15589
Estimated runtime variance = 0.249

4.36
Frequency : 835.4
Period of task cycle = 144.2
Number of points = 8
Minimum runtime in data = 12740.86
Maximum runtime in data = 14269.52
Estimated peak runtime = 15292
Estimated average runtime = 13610
Estimated trough runtime = 12649
Estimated runtime variance = 0.173

Really nice speed up. Also note that the runtime variance is lower.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,872
Credit: 117,918,607,957
RAC: 34,539,335

That's the way to do a

That's the way to do a performance comparison!!

Very useful thanks!

Cheers,
Gary.

Winterknight
Winterknight
Joined: 4 Jun 05
Posts: 1,461
Credit: 377,668,137
RAC: 136,296

Don't have a comparison. New

Don't have a comparison. New computer, here is verbose from three results with computer in steady condition. There are more but I've been tweaking and fiddling with it so times for previous all over the place.

App Version : 4.36
-----------
Frequency : 921.15
Period of task cycle = 174.8
Task sequence number = 385
runtime = 16374
phase = 0.203
principal value = 0.594
Task sequence number = 387
runtime = 15743
phase = 0.214
principal value = 0.623
Task sequence number = 410
runtime = 14347
phase = 0.346
principal value = 0.885
Number of points = 3
Minimum runtime in data = 14347.36
Maximum runtime in data = 16374.41
Estimated peak runtime = 19920
Estimated average runtime = 15893
Estimated trough runtime = 13594
Estimated runtime variance = 0.318

Host is 252515 and before you ask hdd came 'as was' from my old Pent M that released its smoke from psu and mobo. A chicken an egg situation I think, but don't know the answer. Hence old creation date of 4 Jun 2005 19:39:33 UTC for core 2 quad.

Comment viewing options

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