PDA

View Full Version : R10 "Absurd" rendertimes (Local and NET)


tcastudios
02-06-2007, 02:51 PM
Hi.
I need to check this out with you and see if there is something common.
We do work on location making hundreds of effect shorts (loops) used in mediaplayers for live broadcast shows. These are jobs done over a couple of days at the arenas being used.

This last gig did stress me with some "absurd" rendering times.
We used two workstations (Me running Cinema and AE and my companion running AE and god knows what, like Particle Illusion, Combustion etc)
To that we have a small stack of MacQuads as NET Clients, and MacMini as NET Server. My workstation is a MacBookPro to get R10 happening.
And finally some Terabytes of harddiscs, all connected via Gigabit Ethernet/Firewire.
There wasn't any time to really check out what was going on at the gig so I sat down yesterday in my studio for a couple of hours and found some strange things.
To exclude any extra network issues I have only connected "my" stuff and no extra harddiscs.


So, the short version is:
Part 1:
Each time a scene is re-rendered, the rendering time increases. Sometimes to the absurd.
(See render tests below). If I restart Cinema, render times goes back to normal (And then increase again).
Pat 2:
Using NET render with scenes that renders very fast (1-2 seconds per frame) the assembling
takes many times longer than the rendering itself. (See example below).

This did make a Catch22 situation. To get things done I usually just throw the scenes at the Server and then continue with the next scene on my workstation. As that did cause problems
I copied the scenes "never got done" on the Server to Cinema itself, only to find this other increasing rendertime problem.


The longer version, sort of.
Render tests tcastudios.
Checking increasing rendertimes for the same scene.

Part 1: Checking rendering times for CinemaR10 itself.
Part 2: Checking assembling time for R10NET Server.(.b3d to endformat in this case .psd)

For *testing* purposes, the exact same install was done on three different machines(copying the MAXON folder) using Cinema R10.

Part 1:
Testing is done by rendering the scene using CinemaR10, delete the result and emptying the Trash.
Render the scene again, trash the result and so on.
The render settings are set to .psd format using regular alpha.
The test is done using two MacPPC Quads (4.5 resp 2.5 Gb RAM) and a MacProBook(2.16Ghz Intel, 1Gb RAM). All using the same scene running Cinema R10.
All machines updated to latest OS versions(OSX10.4.8), QT etc.
The animation is 800 frames, only have 8 lights using Lenseffect.
No geometry of any kind, so there is only a post effect getting rendered so it normally renders very fast.

What is happening is that the rendering gets longer for each new render.
If Cinema is restarted , the rendering gets shorter again but increase for each new render. Times are in Minutes:Seconds.

Rendertimes PPC Quad1 (4x2.5Ghz , 4.5GbRAM)
1 7.45
2 10.40
3 10.34
Restart of Cinema
4 7.08
5 14.03
6 15.19
7 13.38
8 9.46
9 50.26 (!)
10 38.25 (!)
Restart of Cinema
11 7.42
12 11.36
13 13.19

Rendertimes PPC Quad2 (4x2.5Ghz , 2.5GbRAM)
1 7.28
2 13.04
3 10.18
4 13.31
5 15.49
6 30.44 (!)
Restart of Cinema
7 7.49

Rendertimes MacBookPro Intel 2.16 Dual2 (1Gb RAM)
1 7.51
2 8.19
3 8.39
4 9.18
5 9.25
6 10.10
Restart of Cinema
7 8.01
8 8.26
9 8.54
10 9.19
11 9.09
12 12.15
13 11.07

Part 2:
Using the same scene as above and this time running R10NET.
Server: MacMini 1.83GHz Intel Core Duo, 1GB RAM.
Client 1: Mac PPC Quad.
Client 2: Mac PPC Quad.
Due the current "NET-bug" on OSX, one CPU is turned off on the MacMini
using the CHUD tool.

The rendering of the scene took about 3min for the two Clients.
For the Server to assemble the result (800 frames of .b3d to .psd)
took 48minutes(!).

What is it that makes the assembling take so long? And what can be done do
not have this happening?
If I have a scene that takes about 5-10 seconds per frame,
the assembling seems to "breath" normally and all is done "in time" shortly after
the actual rendering.

The test scene used is attached (R10).

Cheers
Lennart

GruvDOne
02-06-2007, 03:29 PM
Wow Lennart that is f'ed.... I haven't had that specific issue on with my network renders (my set-up is 4 PPC Quads, 3 Core 2 Duo MBPs, with a Dual 800 mHz G4 as the Server). But then, I've only done one big Network render since R10.

The issue I did have though was of about 5% of rendered frames never converted over from .b3d to .tif, and I ultimately had to render them out separately. In this scenario there were no re-renders necessary so the circumstance was a little different.

I hope this is an easily re-producable bug, so that it can be isolated and corrected.

Cheers,
Will

Sneaker
02-06-2007, 03:53 PM
Hi Lennart,

I had a similar problem with R9.6 on a PC with scene motion blurr.
The Problem seems to be that XP/Cinema doesn't clean the memory when file is
written to disk. Therefore the OS needs to page memory to disk, which slows everything down
tremendously.
My idea is, check the memory consumpten when you render / restart the render.
-Michael

smurfted
02-06-2007, 04:14 PM
On my dual intel mac, i don't use the ppc version, i use the mac version as its a lot faster (x2 in places) than the ppc ver.

Are you using any plugins?

tcastudios
02-06-2007, 04:21 PM
Hi.
I'm having a conversion with the support as of now. We're not done yet, but so far
the increasing rendertimes (for the same scene being re-rendered) sounds like a
bug on the Mac side.
There are no plugins involved in the test, and I don't think it is the problem.

And , yup, Intel is the way path from now on.

I'll report as I get more info.

Cheers
Lennart

FredSpeaks
02-06-2007, 04:39 PM
I just ran the file 3 times on a MacPro 2.66

1st 4:45
2nd 7:59
3rd 10:06

There is a definate slowdown. The memory ussage looked pretty steady. There wasn't any more being used each successive time. However the CPU ussage did jump up each successive time. First time through I was only at around 200%. By the third it was up to 300%

Hope this helps

AdamT
02-06-2007, 04:44 PM
It definitely seems to be a Mac problem, and not just in terms of slowdown but in terms of overall speed. I ran the render on my Mac Pro Quad 3Ghz and got the same sort of times and slowdown as reported. Fastest time was 4:15 on the first go and 10:33 on the 5th.

Same hardware but under Win XP Pro (32 bit) I get a consistent render time of 1:44 giver or take a second here or there. So not only is there no slowdown, but it's 3x faster than the fastest OSX render time.

Also noticed that ram use creeps up steadily under OSX, up to 3.65 GB by the end of the render (I have 5 GB ram). Under XP it it doesn't go much above 1.5 GB.

tcastudios
02-06-2007, 05:04 PM
Gosh, that is scary. 3 times faster on the same hardware...
I think support is off till tomorrow so I'll try to get more info then.

I would be awsome if someone running Windows could make a small NET render
to see if that problem also is the same using the same scene.(Assembling takes eons of time)

Cheers and thanks for checking.
Lennart

kromekat
02-06-2007, 05:04 PM
How frustrating! - I am interested to see what comes of this as a current PPC user switching to Intel Mac soon!

I am amazed by your reults Adam - something definitely needs addressing on the Mac side then!

Adam

heathivan
02-06-2007, 05:13 PM
Rendertimes PPC Quad1 (4x2.5Ghz , 4.5GbRAM)
1 7.45
2 10.40
3 10.34
Restart of Cinema
4 7.08
5 14.03
6 15.19
7 13.38
8 9.46
9 50.26 (!)
10 38.25 (!)

Cheers
Lennart

This is xactly what was happening to me in my post GI crashes in v10.
http://forums.cgsociety.org/showthread.php?f=47&t=459579

I noticed every 4th-10th render, the renderlines would stop moving and I would crash out (osx color wheel of death) when trying to close the picture viewer. I knew that the earlier renders were quick so why would the lines stop moving on later renders?

Thanks so much for posting this info as I think it was also my problem with the GI (ahem, sorry for the accusations GI :)

----h

vid2k2
02-06-2007, 05:26 PM
FWIW,

iMac core duo / 2GB ram / v10 demo UB

Time: 5:45 no slowdown
Can't compile images with the demo

acid2002
02-07-2007, 10:29 AM
No problems here G5 1.8
btw
Have not got all the plugins required from your file, but have never had any slow down probs.

Derek

tcastudios
02-07-2007, 10:52 AM
Hi, and thanks for checking.
vid2k2, I -guess- the demo might not have the problem since it doesn't wright anything
to file.

acid2002, is that Mac1.8 a Dual?

Cheers
Lennart

mikeh64
02-07-2007, 01:13 PM
just adding my times...

Quad G5, 2 gigs ram, geforce 6600, C4D R10, OSX 10.4.8

6:50
8:00
13:27
restart C4D
6:51
8:46
13:44

Major Tom to Ground Control - there’s something wrong

acid2002
02-07-2007, 01:42 PM
Hi

No my 1.8 is a single processor net render to 12" 1.5 powerbook.

Derek

Couldn't use the test file as dont have the plugins do you have another file I could test.

thanks.....Derek

Rantin Al
02-07-2007, 01:49 PM
Hi Lennart.

I tried your tests as asked but tests 1>3 were done without a relaunch.
Uptime has been several days at least.
Tests results include 2 relaunches as indicated.

Quad G5 x 2.5, 6GB RAM, OSX 10.4.8, GeForce 6600, QTPro 7.1.3.

Test 1. 14:23
Test 2. 18.32
Test 3. 21:47

Relaunch C4D

Test 4. 06:37
Test 5. 09:11
Test 6. 17:42

Relaunch C4D

Test 7. 6:42

HTH, Alan.

tcastudios
02-07-2007, 01:59 PM
Thanks guys.

acid2002, you can make any scene yourself that is almost " empty".
Like a bunch of lights (not visible) and add lenseffect. Maybe lenseffect is
a part of AdvancedRenderer, I don't know right now.

But I guess only having a simple standard sphere would be fast enough.
Then set to render some 600 to 800 frames.

It would be interesting to see if a singel CPU Mac have this problem or not.

Nothing more from support yet.

Cheers
Lennart

otomo
02-07-2007, 02:15 PM
hmm, I would like to run the testscene on our winXP based renderfarm but its still on 9.6, because we have to finish a pre-10 project.

But what I observe here is very inconsistent render times on our renderfarm. one frame of a scene takes 5 minutes to render and the next frame takes 20 minutes (same machine) without any significant changes in the frame. that is what's driving me nuts at the moment because I dont have an explanation for this (must have to so something with quantum mechanics) and it ruins every clever time-calculation I come up with in order to meet a deadline.

If I find the time I will run the testscene on a r10 winXP renderfarm in the evening.

JoelOtron
02-07-2007, 02:16 PM
1st: 7:38
2nd: 10:54

Dont have time to do more now--but its obvious that the glitch is happenning here too.

Mac OSX 10.48 Quad 2.5 G5, 4gb ram

Erik Heyninck
02-07-2007, 02:17 PM
I have something a bit similar: Cinema slowing down in workflow up to a final freeze. Yet when I reopen the (saved) scene after a freeze, it all goes much better again and at a normal pace. Then again it slows down etc etc.

rel10 on PC, dualcore Pentium, 2GB ram, Nvidia7600 256MB, XPPro

Srek
02-07-2007, 03:20 PM
Hi,
thanks to all for checking, i think we have tracked down the problem. It's only due to the very short render times per frame that it shows this strong, in most other situations the increase in rendertimes is most often not noticable. Nevertheless it needs fixing.

Cheers
Björn

acid2002
02-07-2007, 03:23 PM
Ok

Rendering 0-600 frames (tiffs) over net render G5 1.8 single and G4 1.5 powerbook

getting consistant render times from both and no slow down,

Derek

CGTalk Moderation
02-07-2007, 03:23 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.