View Full Version : Combustion system configurations / format advice.

06 June 2007, 11:57 PM
Hey there,

Aside from a Quadro or FireGL video card are there any other ways to speed up Combustions performance. Also are some formats proven faster to render than others?

I am building a new editing / compositing rig and don't want to short change performance because of my ignorance regarding hardware. A fast processor and 4+ gigs of ram as well as a multi drive raid 0 are my starting point. Any advice would be great!

06 June 2007, 09:27 AM
I've personally had so many problems with hardware OpenGL over the years I've always found myself ending up just using software (CPU + RAM) rendering.

As far as formats:
The less compression the quicker it decodes. But if your RAID can't keep up then some lossless or extremely mild lossy compression might be in order. An extremely broad overview of this will be in an article I wrote for showreel magazine (I think July/August), however, effectively all you need to recognize is that you will hit a bottleneck somewhere in either disk performance or decode speeds, balancing that for maximum return could see some improvements in performance. But overall I've found that the "File In" node of any composite is usually an extremely small % of the overall render time per frame. (Somewhere around 1 second per frame. Obviously unnacceptable for real time, but compositing has never really been real-time.)

Image sequences are significantly faster to decode than movie files is another general good rule.

Be wary of involving Network Attached Storage and not ensuring it;s up to the riggors of high bandwidth random access reads and writes.

To some degree it comes down to a bit of trial and error on what you want to do. If it's going to be a lot of 3D motion graphics with fancy DOF etc... throw more of your money at processing.

If you want dozens of layers of 4k uncompressed DPX files. Make sure you aren't going to choke at the storage end.

P.S. don't forget data recovery into your system cost whether that be redundant systems or an offline backup system.

So pretty much just the obvious stuff, I wouldn't worry a whole lot... unless you're building this yourself in which case: NEVER BUY INEXPENSIVE RAM! NEVER BUY INEXPENSIVE POWERSUPPLIES! Two rules I've established over the years after a half dozen unstable machines. Bad RAM and bad power supplies are the bane of system stability.

06 June 2007, 01:17 AM
Thanks for the tips.

I have started systematically testing different image formats checking for effect on caching frame rate. 24 bit PNG seems to be the winner but that was with DV. I will try HD footage tomorrow and the gaps between formats should widen then. I think I will have to drop .RPF for now though since it seems to cache rather slow in comparison. PNG zdepth passes with G-buffer builder rendered faster than going direct off the .RPF.

I think I will go with a Lacie 2.5 TB 5 disc SATA II array though I was thinking a Raid 0 for speed and your comment on redundency reminded me I will need an equally huge backup. Hmmm.. This is getting pricey.

06 June 2007, 03:58 AM
Be cautious of Combustion, or any application, and codecs. The quicktime PNG codec is 4-5 times slower than other lossless codecs. Basically the more compressed an image/frame is the more time it takes to compress/uncompress it. Also, RP is not a fair codec to compare to as it stores alot more information than standard formats. No other formats, that I know of, will store the camera transform data, though in C4 you can import ASE Camera data, as well as the various extra channels only available to that format.


06 June 2007, 02:29 AM
We ran some quick speed tests today on an 90 frame image sequence @ 720p. We tried out jpeg, tiff (8bit), png (24 and 48 bit) and tga (16, 24, 48). The speed winner was the tiff by a hair. tga 16 was useless due to noise. JPEG came in last. tga and png were a close second to the tiff. We also ran a speed test comparing .rpf with 3D motion blur applied to the zdepth channel against a tiff sequence with a zdepth pass using G-buffer builder. The tiff sequence won here too. I will try adding some more channels/passes and see if the speed gap widens. All these tests were performed with openGL off. Our Quadro 3450 would just display black in some cases at this res with tiff, tga or png. Bummer. Running the comp @ 8 bit so no warnings, just black. Though some particle tests we ran made the Quadro shine.

A though:
perhaps someone should start a thread with a link to sample footage and workspace files for those interested to test out and report their configuration and render times. This was handy when they did this over at Maxwell's forum. There were some interesting results in some cases.

06 June 2007, 05:17 AM
Combustion likes uncompressed file sequences :)

I use mainly DPX, TIF and OpenEXR file sequences

QuickTime files using either Avid Codecs or H.263 when I need to work with something the comes from Avid or Offlines, aside that I avoid using avi or mov files.

Also I have to say that C* uses graphics cards GPU really badly. You can't see the difference between running C* with a Quaddro FX 4500 or a 550.

CGTalk Moderation
06 June 2007, 05:17 AM
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.