Estimated running time / size given by project vs reality !
log in

Advanced search

Message boards : Number crunching : Estimated running time / size given by project vs reality !

Author Message
Profile marsinph
Send message
Joined: 28 Mar 18
Posts: 17
Credit: 98,039,275
RAC: 0
Message 4936 - Posted: 14 Feb 2019, 16:12:41 UTC

Jello,
A few questions :


What or who) estimated computation size (R351_250_300K_....
Windows gives 80GFLOPS. My host has 3.8GFLOPS

On the post estimated running time 8min-1h20 on avx 3.8Ghz.
I run at 4.2Ghz (3.4Ghz nominal OC to 4.2Ghz). So normùally I would need less than 1h21 !
After 2h20 running, still at 96% in stderr.txt !
It is also not frozen. It increase at each checkpoint.

Then at download estimated running time : 00:16:33 (before start)
I not consider the remaining time given by Windows who is very inaccurate.
Only I check and estimate in stderr.txt.
See WU
http://srbase.my-firewall.org/sr5/workunit.php?wuid=358042596

But also http://srbase.my-firewall.org/sr5/show_host_detail.php?hostid=100501
who takes a little less than 4 (four hours)

What do I wrong ?

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7229
Credit: 42,729,227
RAC: 31
Message 4937 - Posted: 14 Feb 2019, 16:59:44 UTC

It looks like your CPU is overheated and throttled.

(ms / 1000) * iterations total = seconds total

I think my runtime had something with 1,4ms not 4,x

Profile marsinph
Send message
Joined: 28 Mar 18
Posts: 17
Credit: 98,039,275
RAC: 0
Message 4938 - Posted: 14 Feb 2019, 18:38:39 UTC - in response to Message 4937.

It looks like your CPU is overheated and throttled.

(ms / 1000) * iterations total = seconds total

I think my runtime had something with 1,4ms not 4,x



Hello,
It is true my CPU (i7-2600k) is in little overheat 70°C.
But you wrote a few weeks ago no problem to 70°C. At 80°C danger.
On a eight core, I run only four WU at same time.
And no any other project, no any other software. Only Boinc.

I have test with only one WU. I go down to 2.8 ms. Not test a few minutes, one hour !
CPU load : 12% - T° : 35°C


Nothing change.

I will try to capture before finish WU details (R531_250-300K_a_wu_124707)

I also see there is no any checkpoint in BoincReport (well in slot)



Sorry if it is confused, but as blind for 95% I need to use artefact to "see"
I use a other host to write. Not the host who crunh.

Look also R351_250-300k_a_wu_124754_0
and ...........124755_0
I let finish this two WU alone.

Then I will go ahead. I will not monopolyse one host for running one task each two hours. Sorry

Profile marmot
Avatar
Send message
Joined: 17 Nov 16
Posts: 97
Credit: 126,410,450
RAC: 21,911
Message 4940 - Posted: 15 Feb 2019, 10:46:21 UTC

Other possibilities when over clocking, error correction in RAM slows the process down so over clocked system RAM or T3 cache could actually slow work down; step your FSB down 1 or 2 points and see if things speed up.

If running WU on 1 or 2 cores then the CPU can use burst mode to higher frequencies to get faster results. When all cores are in use then burst mode can't be used for all available wattage is in use and thermal limits are reached.

Profile marsinph
Send message
Joined: 28 Mar 18
Posts: 17
Credit: 98,039,275
RAC: 0
Message 4942 - Posted: 15 Feb 2019, 18:39:09 UTC - in response to Message 4940.

Other possibilities when over clocking, error correction in RAM slows the process down so over clocked system RAM or T3 cache could actually slow work down; step your FSB down 1 or 2 points and see if things speed up.

If running WU on 1 or 2 cores then the CPU can use burst mode to higher frequencies to get faster results. When all cores are in use then burst mode can't be used for all available wattage is in use and thermal limits are reached.



Hello Marmot,
I was thinking about something like what you explain.
It is full possible. CPU wait too long results from process.
So L3 is full.


About FSB and RAM speed, all is in adequation.
A stupid comparaison :
RAM at 1600, CPU base clock at 100 and FSB at 1333 !!!
Nice way to slown down a machine !
Or to use only one big RAM instead of 2. So again nice way to slow down because dual channel inactive.

and so one and so one...

But I will check again very accurate.

With two simultanous WU, (and I repeat, nothing else is running. When I say nothing is really nothing exept Win7)
running time per bit go to about 2ms instead 4 )
But global running time is not divided by a ratio of 2, only by about 1.5.

And once again, I repeat i compare the perfect same app, on same host


Post to thread

Message boards : Number crunching : Estimated running time / size given by project vs reality !


Main page · Your account · Message boards


Copyright © 2014-2024 BOINC Confederation / rebirther