4K DI options

[Blair S. Paulsen]

Thanks for the info DANCAD3D. My first response is that I do not expect to add complex filtration and then play back 4K at full frame rate immediately, I would be happy with a single frame for immediate preview and then RAM cache rendering in the background.

In any case, my primary DI objective is to play back 4K material in real time, even if that means that image processing steps like color correction have been rendered out in advance. Ideally some sort of large RAM cache would allow me to make changes per the client request and with a 10 second wait I could play back at least a few seconds from the RAM cache at close to full quality.

My quest is for a beefy desktop solution that gets me 70% of what I can get with "big iron" at a reasonable price point. I am cool with days or even a couple weeks of rendering after sign off for a 90 minute feature.

More important to my pipeline is 4K dailies to check focus, noise levels, how the makeup will look at high resolution, application of a print LUT if film out is in the cards, to check lighting ratios, etc.

[Dan Hudgins]

Thanks for the info DANCAD3D. My first response is that I do not expect to add complex filtration and then play back 4K at full frame rate immediately, I would be happy with a single frame for immediate preview and then RAM cache rendering in the background.

In any case, my primary DI objective is to play back 4K material in real time, even if that means that image processing steps like color correction have been rendered out in advance. Ideally some sort of large RAM cache would allow me to make changes per the client request and with a 10 second wait I could play back at least a few seconds from the RAM cache at close to full quality.

My quest is for a beefy desktop solution that gets me 70% of what I can get with "big iron" at a reasonable price point. I am cool with days or even a couple weeks of rendering after sign off for a 90 minute feature.

More important to my pipeline is 4K dailies to check focus, noise levels, how the makeup will look at high resolution, application of a print LUT if film out is in the cards, to check lighting ratios, etc.

I was thinking about how this could be done, if you have some money it may be possable to make a few mods to what I have done so far and be able to get 4k uncompressed play back by using parallel computers since I have master/slave SMPTE sync anyway, you could use this trick as well, no need to wait until 2011 if you can get the right hardware,

1) Don't try to do it on one computer, figure that you can display 1fps at 2048x1536x32 on each computer. Get a video board maker in China to make video boards that work in banked and linear mode at 2048x1536x32 and have master/slave H&V sync so that all the video from the computers will be in sync on their 15pin output.

2) Make a small change to DANCAD3D.EXE (tm) v3.7N or some other system to display the Modulus of the SMPTE frame number part, so that one computer would hold and load all the SMPTE TC frames that number 0, another computer SMPTE TC frames 1, another SMPTE TC 2, and so on up to SMPTE TC 23. This divides the workload each computer needs to do. 24 computers working in SMPTE LTC slave mode (supported now) could display 1920x1440x 16 or 32 or 2048x1536x16 or 32 at 24fps then, maybe, since each would only realy be going 1fps. By running 96 computers in SMPTE slave mode you could get a four input SVGA monitor of 3840x2880 or 4096x3072 running 24fps uncompresed, so that you would see everything, even things that would not show up in an uncompressed playback system because the compression had removed them.

3) Control the 24 to 96 SMPTE slaved computers from a single SMPTE master computer using proxy frames so that the shot selection and jog controlls would allow all the frames to be selected easer. DANCAD3D.EXE v3.7N (tm) does SMPTE master now, you could also use this idea.

4) Build a multiplexer to switch the video output signals from the 24 to 96 video boards so that the current SMPTE frame computer would be selected. It might be needed to put a 12 frame delay offset (1/2sec) in the sound playback to insure that the frames are loaded, so the multiplexer would be running 12 frames delay also in sync mode and you would enter a 12 frame offset into Magix (tm) for the sound, otherwise the software could have a 12 frame lead on the SMPTE code so that the frame would be loaded when the SMPTE TC for that frame was tripped. In other words, the video board buffers update in a ripple so that even though they are at 1fps, each loads just before the 1/24th of a second that the monitor will see its signal.

If Modulus was also applied to the Insert command, and some master command code was written, then the changes and image processing could be done in parallel on 96 computers to cut the render time to about 1%. That might make something like what you are looking for.

If you keep to one system, I think your render times will be days not seconds, maybe even weeks or months. Since even compressed frames for 88 minutes can take TB of data using a RAM cache could not work since I do not know of any normal computer that has a TB of RAM. If you put the RAM cache outside it would still cost $$$$$$$ to have TB of RAM, and going through one computer would still have problems with bottle necks.

If you use a 3D LUT you can reduce the render times on the updates, but doing a 3D LUT on a 4k image could still take too long for real time changes unless you use a hardware 3D LUT in the video which could be a high cost item at 48bpp. Just updating the 3D LUT at 48bpp would take some time.

So you would need 98 computers, 96 SMPTE slave for quarter frames, one for SMPTE master and jog, one or more for SMPTE slave sound play back. If you use 8 computers for sound playback you can have 16 channel sound playback along with your 4k uncompressed images. So that would be 105 computers total. If each computer costs about $400 * 105 = $42,000 plus a few $1000 more for the wires and multiplexer. The screen and sound system might be $20,000 more? So the total might be less than $75,000?

If you could make a 20 foot wide OLED monitor, you could replace the projector in theaters and have something better than projected film or video, no focus problems or smoke and haze in the optics, no lamp to dim, the future of theater display?

[Thor Wixom]

Exactly. FIRST you want killer RED 4K to current 2K infrastructure.

But for the record ... personal ego and perception run this industry.

Oh yeah ... and stupid committees.

I am selling "F the DCI" T-shirts at NAB.

Where do I send my $20?

-Thor

[mezmo]

Hi Dancad3d,
Make that 105 laptops cause I think Blaire needs this thing
in his DIT/camera van.
Mezmo

[Michael Hastings]

After all Laguun and Luki deserve drink and dinner together in Las Vegas.

Or maybe a metalcage death match in the middle of the RED booth. :clown2: