Chaos Corona Forum
Chaos Corona for 3ds Max => [Max] Bug Reporting => [Max] Resolved Bugs => Topic started by: Nekrobul on 2016-05-09, 11:29:41
-
I made some reserach and testing in 1.3 public and 1.4RC2 (I will test RC5 today later).
So it came ou to be that theri is hiding some major bug behignd the skin of the problem.
I used X5550 twin procesor Xeons with 16GB of RAM for the test as a nodes and 3930K with 32 GB of RAM As a lead PC.
And the results are quet dramatic.
The wierdest thing i noticed was the ACC structure building time on the Lead pc 1.3 - 12 SEC 1.4 RC2 - OMG!!!!!!! 5min 32 sec!!!!!!!!!!
UPD - added test for 1.4 RC5
-
I think this must be some system-specific thing. There are some known problems with Xeons, but it's usually with 64+ cores... Maybe 2 processors is enough cause. Are you able to go to task manager and change process priority and affinity of 3dsmax.exe during rendering? I mean, is this possible for you at all, or are you getting some errors like "access denied"?
-
what is the CPU utilization during parsing?
-
what is the CPU utilization during parsing?
In both cases it is slowly growing from almost 0 at the start and to 100%
in 1.3 it takes around 7 mins to grow to the full potencial
in 1.4 i dont exactly know how much it takes to get to full potencial (i was not patient enough)
About ACC structure it is floating from 0 to 60 in 1.4.
I mean, is this possible for you at all, or are you getting some errors like "access denied"?
Have not tryed yet. But it will be inconvinient to change it every time new frame strts as i said before.
---------------------------
Interesting thing it that it only apears when the RAM is beeing overloaded (on nodes) But long ACC building happens on the Master pc wich have enough free memory left (24gigs out of 32 are eaten by the scene)
PS - If it is pausible to somehow Low memory warning to be displayed only once in TOTALY spams out all the log.
-
I mean, is this possible for you at all, or are you getting some errors like "access denied"?
Have not tryed yet. But it will be inconvinient to change it every time new frame strts as i said before.
I am not suggesting this as a solution. Just asking if you are able to do this, and what happens.
Interesting thing it that it only apears when the RAM is beeing overloaded (on nodes) But long ACC building happens on the Master pc wich have enough free memory left (24gigs out of 32 are eaten by the scene)
It is nice of you that you are adding this now. ;)
So what happens if RAM is NOT overloaded on the nodes? Is the ACC building also slow on the master? Some other wrong stuff?
-
I mean, is this possible for you at all, or are you getting some errors like "access denied"?
Have not tryed yet. But it will be inconvinient to change it every time new frame strts as i said before.
I am not suggesting this as a solution. Just asking if you are able to do this, and what happens.
Interesting thing it that it only apears when the RAM is beeing overloaded (on nodes) But long ACC building happens on the Master pc wich have enough free memory left (24gigs out of 32 are eaten by the scene)
It is nice of you that you are adding this now. ;)
So what happens if RAM is NOT overloaded on the nodes? Is the ACC building also slow on the master? Some other wrong stuff?
Acc on master goes long anyway. And the interesting part is that there is 20% difference in time between RC2 and RC5. Wich means that something was done that optimised process.
Btw i can share the scene to play with it is easy to recreate.
I will try to swich the priority as soon as coleges will free the nodes XD
-
ok, I can take a look at the scene, but it is probably something PC-related. It may be some too aggressive energy saving policy
-
ok, I can take a look at the scene, but it is probably something PC-related. It may be some too aggressive energy
May be it could be new embree?
UPD - Uploading scene, filename - FLW2_recoverFinal. Just test out with one condition 32 gb on Main PC and 16 gb on slave. I think it will apear on any processor.
-
oh shit my post got cut off. I hope I will be able to reproduce the issue
-
oh shit my post got cut off. I hope I will be able to reproduce the issue
I will be glad to provide aditional info if neseasry.
UPD - OMG i forgot to make a test with LTP on in 1.3 X_X just to make picture complete.
-
I got very interesting data from this test.
I seems that on master PC Building ACC with ltp switched on takes same time as in 1.4 BUT in 1.3 the render it self with LTP swithed on begins a lot faster (processor starts running ar 100%) it took around 3 min.
PS - Maru i tested out manual priority changing on slave it works just like it should downwards and upwards no problem.
-
moving to resolved as there are many similar threads, but we have this on our internal list of threads related to RAM issues