Page 1 of 2

Very weird SPH sim times

Posted: Wed Jun 05, 2013 6:56 pm
by Otuama
I've just run a set of simple tests of particles being emitted into an open box object. I'm getting bizarre sim times with SPH.

The scene hasn't changed at all during the process, 100 frames, the particle counts were exactly the same (14k) and I cleaned the data folder before each test too just in case Rf takes a bit longer to overwrite files (not sure if it does or not).

Test 1

With GUI updates on the sim took 4 mins.
With GUI updates off (ALT-D) it took 8 mins.
With command line it took 34 secs.

Test 2

I did another test because maybe I'd gotten the results mixed up.

With GUI updates on the sim took 10:52 mins.
With GUI updates off (ALT-D) it took 14 mins.
With command line it took 42 secs.

Test 3

I restarted Rf:

With GUI updates on the sim took 1:47 mins.
With GUI updates off (ALT-D) it took 5:24 mins.
With command line it took 41 secs.

Test 4

I re-ran the test (without restarting Rf):

With GUI updates on the sim took 8:30 mins.
With GUI updates off (ALT-D) it took 12:16 mins.
With command line it took 40 secs.

In every test having openGL off took longer, which is strange in itself but to get large inconsistent sim times is very strange.

Of course, the CL reigned supreme in each test :D .

Re: Very weird SPH sim times

Posted: Thu Jun 06, 2013 2:02 am
by o.delellis@gmail.com
hi Otuama,

SergeK found Rf slowed down on every re sim, and a re start was needed to match the first sim times...i think the post is/was on the RF2013 beta forum..


but from 4mins to 10 mins!! and display off takes longer...???

have you run the same file with Rf2012 for compaison?

Re: Very weird SPH sim times

Posted: Thu Jun 06, 2013 9:15 am
by Otuama
With 2012. I haven't bothered using the command line for this one.

The scene is set up exactly the same as 2013. I'm getting the same particle counts (14k) too.

Test 1

With GUI updates on the sim took 39 secs.
With GUI updates off (ALT-D) it took 37 secs.

Test 2

With GUI updates on the sim took 44 secs.
With GUI updates off (ALT-D) it took 38 secs.

Test 3

I restarted Rf:

With GUI updates on the sim took 38 secs.
With GUI updates off (ALT-D) it took 40 secs.

Test 4

I re-ran the test (without restarting Rf):

With GUI updates on the sim took 43 secs.
With GUI updates off (ALT-D) it took 45 secs.

Re: Very weird SPH sim times

Posted: Thu Jun 06, 2013 9:48 am
by o.delellis@gmail.com
well that does it!


will keep my install of 2012 for SPH......

Re: Very weird SPH sim times

Posted: Thu Jun 06, 2013 9:54 am
by Otuama
Maybe give it a try for yourself. It may be my Rf install, or the scene itself that's buggy:

https://dl.dropboxusercontent.com/u/7680428/SPH.xml

Re: Very weird SPH sim times

Posted: Thu Jun 06, 2013 10:01 am
by o.delellis@gmail.com
am dealing with enough issues with h2 and retiming here....

will give it a try tomorrow (serenity nooooooooooooooow)

Re: Very weird SPH sim times

Posted: Thu Jun 06, 2013 10:20 am
by Alex
Otuama wrote:Maybe give it a try for yourself. It may be my Rf install, or the scene itself that's buggy:

https://dl.dropboxusercontent.com/u/7680428/SPH.xml
Hi Sean,

should we also get the initial state bin for the emitter? or can we just test it with 0 particles from the beginning?

Additionally, can you please perform the test without exporting any file? It would be interesting to check if the hard disk access is making any difference.

Re: Very weird SPH sim times

Posted: Thu Jun 06, 2013 10:30 am
by Otuama
Hi.

I just ran it from 0.

I'll try the no export thing now.

Re: Very weird SPH sim times

Posted: Thu Jun 06, 2013 11:16 am
by atena
Hello, I did some tests on my side to check that problem although we did it before the launch. I mean a performance test comparing RF2012 versus RF2013. I must say that I got exactly the same times for RF2012 and RF2013 and it is normal as the SPH solver is still the same. I used the following scene that you can load either in RF2012 and RF2013.

https://dl.dropboxusercontent.com/u/183 ... rf2012.flw

I tried with GUI, non-GUI and CMD and got the same times but this is normal as the load of work is significantly small, you will see differences when you crank up the number of particles. I was exporting files, thing that as my colleague Alex mention is not good for testing performance as file caching and other SO tasks related to files can affect the simulation time.

It might be related to the SO? or maybe the hardware? can you give us a full description of your SO and hardware? Here I tested using Windows 7 (Version 6.1 (Build 7600)).

Re: Very weird SPH sim times

Posted: Thu Jun 06, 2013 11:27 am
by Felix
I got similar issues regarding the simulation time with 2013. Switching to a previous graphics driver kind of solved it although I sometimes get random crashes now..