Long time calculation

Alexis · Post by **Alexis** » Wed Jul 05, 2017 10:18 am

Hello everyone,

I have a problem with my new PC. When I render a rendering, the preparation phase is very long, sometimes longer than the rendering in itself. 3-4 min for preparation GI, less than one minute for rendering and 2-3 min for the "antibanding" phase after the 100%

If I cancel the rendering I sometimes have to wait 8-10 min before I can return to the layout.

It does on all my kray scenes.

You have an idea?

Thank you

Bonjour à tous,

J'ai un problème avec mon nouveau PC. Quand je lance un rendu, la phase de préparation est très longue, parfois plus longue que le rendu en lui même. 3-4 min pour la préparation GI, moins d'une minute pour le rendu et 2-3 min pour la phase "antibanding" après les 100%

Si j'annule le rendu je dois parfois attendre 8-10 min avant de pouvoir retourner sur le layout.

ça le fait sur toutes mes scènes kray.

Vous avez une idée?

Merci

Janusz Biela · Post by **Janusz Biela** » Wed Jul 05, 2017 11:25 am

Probably deep recurse or instances problem.
Send me scene: janusz.biela@gmail.com via wetransfer (confidential of course) I will check.

promity · Post by **promity** » Wed Jul 05, 2017 2:39 pm

Maybe it's related to this or that hardware - CPU, RAM, Motherboard, etc, as it was with me. Symptoms are very similar.
In my case - and I use engineering samples of processors - Gregorz had to add something to the code so that my processors could work in normal mode when rendering through K3.

Alexis · Post by **Alexis** » Wed Jul 05, 2017 5:45 pm

Janusz Biela wrote:Probably deep recurse or instances problem.
Send me scene: janusz.biela@gmail.com via wetransfer (confidential of course) I will check.

You can dowload my 400 Go disk work?

I have this problem with big scene, over 1.5 million poly, many instance veget and objects and i suspect problemb with RAM instructions.

My hardware, bi xeon e5-2696 v3, Ram 4x16gb 2400 cl17 Ecc registered and parity, MB asus Z10PE-D8ws.

I render test frame (960x540 pixel), click on "render frame", 3min for loading and calcul kray, 40s "raytracing calcul" and 5min for return in layout for view render.. For simple texture test is very long.
Even .lws with my i7, I do not wait so long once the calculation is finished and the image go faster in raytracing calculation.

need bios parameter with this RAM ?

Thank you for help.

Alexis · Post by **Alexis** » Wed Jul 05, 2017 5:51 pm

I test three scene with K3_rc1, no change compare K2.62, they all crash lightwave.

Janusz Biela · Post by **Janusz Biela** » Wed Jul 05, 2017 6:15 pm

Alexis wrote:I test three scene with K3_rc1, no change compare K2.62, they all crash lightwave.

Please read Node infrmation what is not supported. Try use Kray3override plugin - disable Nodes there. If scene is running then you use not supported Nodes.

thibault · Post by **thibault** » Wed Jul 05, 2017 6:42 pm

Alexis wrote:I test three scene with K3_rc1, no change compare K2.62, they all crash lightwave.

Ok ! Disable nodes _ sun _ seems light too and pehaps, but i'm not sur, pehaps walls and windows ! After, it's ok !

Janusz Biela · Post by **Janusz Biela** » Wed Jul 05, 2017 7:15 pm

thibault wrote:
Alexis wrote:I test three scene with K3_rc1, no change compare K2.62, they all crash lightwave.
Ok ! Disable nodes _ sun _ seems light too and pehaps, but i'm not sur, pehaps walls and windows ! After, it's ok !

Using K3 for big projects is quite risky yet. It can be some sort of bugs with memory management or strange things...
Please use only in small scenes.

Alexis · Post by **Alexis** » Tue Jul 11, 2017 10:50 am

Hi, I have just re-tested by comparing two PCs.
850pt in cinebench r15 for an i7 (win10 64 family) and 4500pt for the bi-xeon (win10 64 pro).

The same scene without "cache irradiance", 55min on i7 and 10min on the bi-xeon. It's OK

With "cache irradiance" and "shared for all frames" already calculated, 19min on i7 and 10min on the bi-xeon with idle processor !!

I do not understand anything anymore. I do not use plug-in, only classic nodes and classic lightwave surfaces.

I timed when I canceled a rendering in progress. 35s on the i7 and more than 6min on the bi-xeon!

Kray is not compatible with some CPU instruction? Or specification ram (ecc, parity, ...)?

Thank you

Janusz Biela · Post by **Janusz Biela** » Tue Jul 11, 2017 11:23 am

Alexis wrote:Hi, I have just re-tested by comparing two PCs.
850pt in cinebench r15 for an i7 (win10 64 family) and 4500pt for the bi-xeon (win10 64 pro).

The same scene without "cache irradiance", 55min on i7 and 10min on the bi-xeon. It's OK

With "cache irradiance" and "shared for all frames" already calculated, 19min on i7 and 10min on the bi-xeon with idle processor !!

I do not understand anything anymore. I do not use plug-in, only classic nodes and classic lightwave surfaces.

I timed when I canceled a rendering in progress. 35s on the i7 and more than 6min on the bi-xeon!

Kray is not compatible with some CPU instruction? Or specification ram (ecc, parity, ...)?

Thank you

Do not use Final Gathering (cache) The risk of deep recurse is too big.
I forget Final Gathering 1.5 year ago.
Only QMC. With that hardware you do not need FG at all.
Now you will ask me how to share GI........

If you are interested in I will post procedures here.

btw of deep recurse.
Try add this command in Header Cmds:

Code: Select all

lwo2rayslimit 1,0.5;

btw#2
This is reason why I use QMC system: easy, fast, looks almost like Path Tracing.

Alexis · Post by **Alexis** » Tue Jul 11, 2017 12:42 pm

Thanks for return,

It is for an animation, only cam move. With the QMC method (no cache), there are "splotch" in lighting.

I have 2000 frames to render by garagefarm, and their cpu are not as fast as mine.

big money for render...

Yes your procedure interests me

Janusz Biela · Post by **Janusz Biela** » Tue Jul 11, 2017 1:05 pm

Alexis wrote:Thanks for return,

It is for an animation, only cam move. With the QMC method (no cache), there are "splotch" in lighting.

I have 2000 frames to render by garagefarm, and their cpu are not as fast as mine. big money for render...

Yes your procedure interests me

I will try post it today.
General idea is that:
You can not save GI when you use PM+QMC because...doesn`t exists

But you can save Photon Map solution - this is very small file with ONLY information about photon power and positions (a few kilobytes)
This PM solution will be later used like LOAD GI. Because QMC sampling is accurate in his nature behavior , differences between frames are very very little.
The most important is to have Photon Map solution without changing photons cells every frame. This protect you from flickering.
Each frame is render separately like still image loading this PM map created by you.
If I remember I tested this on GarageFarm - so it works. For network system it doesn`t matter what he loads: GI or PM solution.
Noise from render (because of QMC sampling) you can reduce in post process , just catch good balance level: noise/quality.
The quality of render is very close Path Tracing.
Of course whole system works slower but in some situations (complex scenes) you win with this.

artattak tested this in animation in complex scene (I did only synthetic) and even with big noise animation looks good.

Alexis · Post by **Alexis** » Tue Jul 11, 2017 3:45 pm

Janusz Biela wrote: I will try post it today.

I'm looking forward to it!

Janusz Biela · Post by **Janusz Biela** » Wed Jul 12, 2017 10:30 am

Alexis wrote:
Janusz Biela wrote: I will try post it today.
I'm looking forward to it!

I created post here:
http://www.kraytracing.com/phpBB3/viewt ... f=4&t=5333

www.kraytracing.com

Long time calculation

Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Re: Long time calculation

Menu

My account

Kray Tracing Account