Any suggestions to improve banding/monitoring? Is firewire truly lossless?

Question

Hello all.
I've finished a fine cut of a feature film shot on DVCPROHD in FCP, and have just sent the project to color.
I'm very impressed with the feature set and useability of colour, but I'm experiencing a couple of serious problems that I hope one of you guys with more experience than I will be able to help me solve.
1) I'll be adding vignettes to darken the edges of many shots in the film. Achieving the vignettes and darkening the outside to achieve the look is very easy, but I'm experiencing terrible problems with banding. The bit depth of the project is set to floating point, which I'd assumed was the best depth, but the only options that are available to me are 8 bit or floating point - the others are greyed out - which makes me wonder if my graphics card is up to the job. My G5 quad has an Nvidia Geforce 6600. Any thoughts?
2) I'm monitoring on an SD Sony 20" broadcast CRT via the SDI output of my Decklink HD extreme card, and the picture quality is not as good as it is when monitoring the output of FCP in the same way. There is a lot of noise in the blacks that just isn't there either on the masters or the FCP output. When I send test shots back to FCP the problem seems to resolve itself. I understand that downconversion is a complex process, but the noise in the image in color does really render the programme borderline useless. Have I maybe missed a setting somewhere? Is the Decklink card just not up to the job? Is it a question of my video card again?
3) Finally, I captured the footage over firewire, and then it occurred to me that I don't know if it is completely lossless. I don't mind investing in a high speed disk array to capture as uncompressed HD via SDI, but only if it's better than capture via firewire. The film will eventually be blown up for cinema projection, so even a tiny loss of quality will be greatly magnified.
I hope I've given you enough information to be able to give me a hand - I'm starting to tear my hair here!
Best.
DC

Answer 1

On point 1.
Yes it is the GPU card. The reason that the output sent back to FCP from COLOR is that COLOR ALWAYS renders in float. But it can only DISPLAY what the graphics card is capable of, and your nVidia is not on the list.
On point 2.
I'm going to postulate that the Decklink output resolution is being bottle-necked by the GPU, which FCP doesn't use, expect for the User Interface, if its got a video card (like Kona, to off-load its higher-bandwidth requirements).
On point 3.
Firewire transfer is not the issue. The codec is. The initial media capture over FireWire will have harvested everything that was on offer from the (tape?). Considering the original 8-bit colour depth, Recompressing to DVCProHD for cinema output is not the path to take. Eventually you will need to produce an RGB 10-bit dpx sequence, although that may vary with your filmout facility requirements. Get on the same page with them ASAP.
You're going to be in for some shocks anyway when you see the show 20 feet wide instead of 20 inches. You have apparently reached picture lock -- have you seen it on a large screen yet? Some scenes may cause you to re-evaluate... There is absolutely no way to synthesize the effect that the change in scale is going to have. Been there, Done it, Have a collection of T-shirts.
jPo