Chaos Corona Forum

Chaos Corona for 3ds Max => [Max] I need help! => Topic started by: guest_guest on 2016-04-08, 16:52:54

Title: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-08, 16:52:54
Hi,

here is an screenshot of rendering an huge scene that's not fit to my 16 GB ram ... [in fact need about 80 GB of ram!!]

but ... when used 1.3 in bucket mode vs 1.4 progressive [adaptive + denoise] , the cpu performance always was in 90 - 100 percent .... in 1.4 is in abnormal state !?!

is any solution for this ??
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-08, 19:02:50
maybe embree issue?! i changed the structure to BVH and take better result !!! [my cpu is 4790K + 16GB ram + Win 8.1 x64 + Max 2013 + 1.4 build 2016-04-05]
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-08, 21:21:02
any body there?
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: FrostKiwi on 2016-04-08, 23:23:12
Windows started paging 3dsmax's commits. As you see, Disk 1 shows 100% usage.
That explains low CPU usage, as the occurring paging will be limited to the HDD speed. This also cripples all rendering performance.

7mio instances with 200,000 mio polys is quite a thing. Embree can crunch through any number of geometry, but it does it in a more memory intensive way, I'm not familiar with the details though. Generally the idea of instances is to lower memory usage, but embree is more naive and this is one of the places where intel's render kernel gets it's speed from. What you could try to do is to merge many similar geo's into one object. It improved memory usage in the past for me, although I have not worked with such heavy scenes in corona. I get the feeling though, that god damn 200000mio polys won't fit 16gb anyhow, you are definitely low on memory.

So it kinda goes back to the good old ideas from me: decimate the instanced object's polys, turn off displacements if any and get more Ram. (http://downloadmoreram.com/)
Also there is this very old article for Vray (http://www.vrayworld.com/index.php?section=tutorials&category=useful+tips&tutorial=useful-tips-speed-up-your-vray-rendering) it's not much help but maybe will give you a good idea and is a fun read, while waiting for some official say and tips.
 
any body there?
Generally speaking you are using a daily, so expect to hit some walls here or there, devs are sweating to fix stuff, 24/7 text support has to be sacrificed for that.
Corona does not support out of core rendering, so bucket mode does not lower ram usage and did not when you had 1.3 installed.
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: Ludvik Koutny on 2016-04-09, 00:47:59
Denoising stores a few more internal render passes, which in turn increase memory consumption. If you have 16GB of RAM, then this could push your scene over tipping point, BUT in 720*405 resolution it should make nearly no difference... :|
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-09, 01:23:52
Windows started paging 3dsmax's commits. As you see, Disk 1 shows 100% usage.
That explains low CPU usage, as the occurring paging will be limited to the HDD speed.

Disk 1 is 600 Mbps Write speed SSD ... not f***ing HDD !! :\

But thanks for your reply !!
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-09, 01:38:55
and get more Ram. (http://downloadmoreram.com/)

Nice Joking!! :\
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: FrostKiwi on 2016-04-09, 01:42:15
Disk 1 is 600 Mbps Write speed SSD ... not f***ing HDD !! :\

But thanks for your reply !!
I was about to say, for paged HDD, the performance is still quite good.
Either way, paging to SSD is almost equally devestating too performance. That 600mbs, does only 60mbs  + paging overhead in 4k reads at best, which is still nothing compared to RAM and a huge bottleneck.

Nice Joking!! :\
Ohh, sry I didn't think you would take that seriously :D
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-09, 01:51:44
well ... i hope to see Corona for production purposes!! not only for funny cheap interiors and exteriors!!! Great core and great coding of this engine for funny interiors??? this is a joke!!
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: FrostKiwi on 2016-04-09, 02:58:22
well ... i hope to see Corona for production purposes!! not only for funny cheap interiors and exteriors!!! Great core and great coding of this engine for funny interiors??? this is a joke!!
There are many full animated city scale commercials, with water sims, forests and volumetrics all in ome shot in the gallery. The issue lies within your setup. You are free to use whatever tool you want, but if you believe, that some other renderer can eat through 7mio instances and 200,000 mio Poly's with only 16gb ram, you are gravely wrong.

Better than complaining on the product's forum, put material override to a blank corona material and compress the maxfile in an archive and upload it, if your contract allows it, someone might take a look at it.
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-09, 03:14:11
i was rendered this scene in only 14 min with 1.3 bucket mode at the past ... many of work in gallery was rendered in 1.3 !! not unstable 1.4 ?! i believe that over 75 percent of this works used bucket !!!
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: Christa Noel on 2016-04-09, 03:50:14
hey guest_guest, please don't use too much exclamation mark in forum, it makes you looks like an unstable joke. :)
there is no corona 1.4 yet, if you're using pre 1.4 dailybuild then it means that you're using a WIP features that has a lot of unstable things.
cheers!
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: romullus on 2016-04-09, 09:49:51
and get more Ram. (http://downloadmoreram.com/)

Nice Joking!! :\

Oh boy, i wish my browser would have so little tabs...
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-09, 14:23:44
hey guest_guest, please don't use too much exclamation mark in forum, it makes you looks like an unstable joke. :)

Who are you? are you moderator?? seems that this forum has several moderator that can be block me every time ...
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: FrostKiwi on 2016-04-09, 14:50:18
Who are you? are you moderator?? seems that this forum has several moderator that can blocked me every time ...
I doubt anyone would block you, only time people get banned here is for high level spam. His comment isn't meant in a bad way, every forum user has to keep an acceptable type face to allow for a serious analysis and help, multi bumps and exclamation points will make readers ignore a thread.
I understand your anger with clients breathing down one's neck if a deadline is missed, but just out of respect for the devs, keep it simple and clean and someone will help you.
Also you may get unlucky with timezones and for someone who understands the inner workings well enough to help you, can take some time.

Like previously mentioned try:
Rendering with gray material override,
With big chunks of geometry Molded into one object

if that doesnt speedup things consideribly, you are out of luck and Ram is your limit.
Get more ram, do selective rendering or Roll back to what you know worked for you in the past.
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: Ondra on 2016-04-09, 15:54:50
hey guest_guest, please don't use too much exclamation mark in forum, it makes you looks like an unstable joke. :)

Who are you? are you moderator?? seems that this forum has several moderator that can be block me every time ...
I just did. Please calm down and come back in 2 days.
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: romullus on 2016-04-09, 16:08:05
hey guest_guest, please don't use too much exclamation mark in forum, it makes you looks like an unstable joke. :)

Who are you? are you moderator?? seems that this forum has several moderator that can be block me every time ...

You have two options: A. start acting as grown up responsible adult and receive all benefits this forum can offer B. continue to jerk around and cry every time your new account is banned.
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-11, 23:27:29
You have two options: A. start acting as grown up responsible adult and receive all benefits this forum can offer B. continue to jerk around and cry every time your new account is banned.
hey Moralist !! Psychologist!!
Do you have any options for memory problem?

for example: A.out of core rendering B.out of core rendering

Redshift is GPU based renderer with out of core tech!!

 Corona is proudly cpu base !?! does not have out of core ?!?!?
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: Ondra on 2016-04-11, 23:52:55
Why did you request deletion of your account just to create a new one?
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: FrostKiwi on 2016-04-12, 02:50:18
Corona is proudly cpu base !?! does not have out of core ?!?!?
Although immaturely asked, it's still a valid question worth answering:
CPU and GPU out of core approaches are vastly different.
When VRAM is full, you can offload part of the data set onto RAM and let the GPU work from that set of virtual memory.
Once you go to CPU and you RAM is full, you don't have a RAM 2.0 you can offload to. The only option is slow non volatile memory types, which all require some kind of trickery and hacks not to completely cripple performance. You could make the argument, that Windows Paging is already some kind of out of core rendering, but as you see, offloading everything naively, is awful.
So you need some kind of smart algorithm, maybe camera cone based, or culling based offloading, whatever you can think up, combined with some smart tree cache file you can draw from in a smarter manner than let's randomly throw data onto disk and pray it's the right one.

Anyway you go, you will get a substantial performance hit and thus this is not a priority. Furthermore with CPUs you have the option to upgrade RAM, with GPUs you are stuck. What is vital for GPU renderers to even have the possibility for large scale rendering, is merely a feature of many for CPUs, overshadowed by more important things.
And you are right in some form guest_guest, a large part of the early corona adopter crowd work in ArchViz and their voices for features were heard louder, then more exotic features. The feature request subthread is there for a reason.
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-12, 03:53:03
Although immaturely asked, it's still a valid question worth answering

Hey man,
special thanks for your replies ...

i hope to see valid replies from corona team ... not only ban , block , indifference and ... ;)

and another thing ... i'm not an english or american person ... so my english is not good !! If you have been a misunderstanding is not my fault!!!
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: Ludvik Koutny on 2016-04-12, 09:41:44
so my english is not good !! If you have been a misunderstanding is not my fault!!!

This guy...
(https://media.giphy.com/media/L8XuphFGqlSfe/giphy.gif)
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: guest_guest on 2016-04-12, 18:27:49
This guy...

(http://www.costaricantimes.com/wp-content/uploads/2013/07/the-pink-panther.jpg)
Title: Re: 1.4 performance vs 1.3 bucket performance in full memory situation ???
Post by: Nekrobul on 2016-04-13, 09:16:39
OMG. What the hell is going on here.

Off topic first - We are friendly comunity, we are always glad to help with advice. And we also expect same behaviour from all forum users so please don't act like a jerk.

About the memory consumption problem - There are always ways to siplify scene. Retopology to lowpoly meshes, opacity masks, saving instancing geometry to proxy, lower resolution textures, disabling texture filtering and loooot loooot of other workaround that will help you to fit in to 16 gigs of ram. And there is magical pagefile wich might also help you relocate some ram.