The change from 8 bit to 10 bit doesn't involve a quadrupling of the bitrate, but rather an increase of 25%. You have to code 8 bits anyway, 10 bits gives 2 more, so an increase of 25%. The same principle as when you write 100, you don't have to write 10x as many digits on the page as when you write 10.

My belief is that whereas with coding some improvements (50p v 25p, higher colour space etc) you don't need as much extra bandwidth as may be first thought, with the case of 10 bit you do. Reason is that with the other examples it's possible to take advantage of correlations between the new data and the old, in this case, the new least significant digits are random compared to the 8 bit data.

"One step forward, one step backwards" probably sums it up better. The change to 10 bit (without additional bitrate) will improve things in one way, make them worse in another. "Far inferior image" may be overstating the case.

Vegas now has a 32bit float pipeline. I've only tested this with the 10bit YUV codecs but certainly for them the whole 10bits goes into and out of the pipeline as it should. The downside is not having a 16bit integer mode like Ppro and AE does. All those 32bit float calcs do use up a lot of RAM and CPU.

It's 8 or 10 bits per channel and there are three channels: red, green, and blue. 4x the amount of information per channel means 64x the total amount of information overall. But of course it depends on the codec, etc etc.

The analogy with the digits is a bit misleading because we are talking about binary. Adding two more bits brings the number of colors per channel from 2^8 to 2^10, which is a factor of four. However, if the codec's a YUV where that first channel is for luminance, then maybe only U and V are extended to 10bit, which is only two factors of 4.. Bottom line, it's safe to estimate that jumping from 8bit to 10bit will increase the size of the files by somewhere between 4x-64x. Try it with uncompressed video and you'll see.

Regarding the final delivery, if they accept 10bit then you definitely want to process and deliver in 10bit for the most predictable results. I am operating under the assumption that most delivery formats are 8bit, which may be an incorrect assumption.

Upgrading to 10bit is generally a diminishing return these days, but of course one day in the future it will be completely mainstream.

So if the AJA Ki Pro records 220mbps using 10 bit which 4 times the information is needed how would that compare to the Nanodrive recording at 100mbps in 8 bit? I assume the quality would be like the nanodrive at 50mbps but at ten bit.

Would I be right in thinking that the Ki Pro will be much better in terms of colour correction but the Nanodrive be slightly better at overall defination?

Mark

Uncompressed 8 Bit 1920x1080 4:2:2 video at 25fps has a data rate of 99MB/s and 10 bit is 132MB/s. The extra 2 bits only add 33% more data, not 400%.

Not so, Jad. Alisters quite right with his figures, see Uncompressed video - Wikipedia, the free encyclopedia . A file size increase of about 33% for 10 bit compared to 8 bit, not 4-64x.

Going from 2^8 to 2^10 certainly means 4x as many possible values, but only 2 more bits for every 8 original bits, and (by definition) that's what directly influences the bitrate. Hence 25% more basic bits for a 10 bit system.

The puzzle is why 33%, and not the 25% that the raw bit figures would suggest. I think it may be to do with the fact that Y,U,V (each of 10 bits) are stored as four bytes (32 bits total) for practical reasons, hence "wasting" 2 bits per sample. Hence four bytes per sample for a 10 bit system, three bytes for an 8 bit system - an increase of 33%. (I think there's a bit more to it because of the 4:2:2 sampling, but it's something along those lines.)

Thanks Alister

Think I have enough now to make an informed decision.

All the best

Mark

You're absolutely right, I'm wrong. Sorry to cause confusion! I incorrectly related the number of colors to the number of bytes. You are spot on about the wasted bits causing the 33%.

Mark, accept my apologies for causing you a headache! Listen to the others :)

Thanks Jad

Yes for a while there I was confused!

I appreciate you were trying to help and in the end I got a good understanding that I should remember.

Mark

Attached are a couple of sections of frame grabs from some EX1 and NanoFlash footage that has had a heavy grade applied. The difference is very clear. For a more detailed review of the quality and further frame grabs take a look at my review:
XDCAM-USER.com Convergent Design NanoFlash Review Part 2 – Picture Quality

Before anyone complains that the frame grabs are different sized files I will say that these are good representations of what I saw on my monitor and they were compressed at the same compression ratio. This shows just how much more detail is being retained in the 100Mb/s footage.

Thanks Alistar

This is great! It appears colour correction is improved to a degree that is neglible from 10 bit with the nanoflash.

Mark

Looking at those 2 grabs, I've been wondering: yes, the 35Mbps version has a lot of blockiness to it, but IMHO, the NanoFlash 100 Mbps is not clean either - instead of being blocky, it's very grainy.

Alister, could you comment, please?

Piotr: Do remember that the grabs are from graded footage, one of the results of the heavy grade was to increase the appearance of any noise. The noise is coming from the camera. At 35mb/s the codec is running flat out, as a result the codec tries to save bandwidth by smoothing out the noise which results in the blockyness. At 100Mb/s the codec has enough head room to be able to accurately record the noise so there is no need to do any bandwidth reduction.

When you watch the actual video clips you can see a direct correlation between the noise and the macroblocking. This indicates that the noise is from the camera, not the NanoFlash and it is a very good example of the difference 100Mb/s makes over 35Mb/s.

When discussing the relative merits of 10bit and 8bit you should also consider that 10bit comes in more than just linear. DPX generally uses the Cineon log curve that sets black at 19 and white at 635 leaving usable headroom for superwhites. My understanding is that this is what the Cinegamma curves in the EX cameras are attempting to do however as has been seen with 8bit encoding one can easily run into issues with color banding.
As far as I'm aware the Nanoflash will not record 10bit, for that we'll need the Flash XDR and a lot of CF cards. It would be interesting to see how the EX cameras perform recording uncompressed 10bit with the Flash XDR. My suspicion is the camera lacks the latitude to make the added cost involved worthwhile but I'd love to be proved wrong.

How much latitude do you want? The EX cameras are just about as good as it gets when it comes to latitude. As has already been discussed in this thread most banding issues come from poorly compressed, bit starved images. Certainly 10 bit uncompressed will be better than 8 bit compressed, but it's not as big a difference as you might think.

We could have 10-bit compressed. No reason why we couldn't have JPEG2000 compression in a NanoFlash style device. I believe CD said that they would be working on a 10-bit device a while ago.

I believe the FlashXDR will be getting 10bit uncompressed. The Ki-Pro uses ProRes which is 10 bit but is more compressed than the 100Mb/s of the NanoFlash. A 10 bit codec requires a higher bit rate so storage is going to be more expensive, but then storage is getting cheaper almost daily.

I love my NanoFlash, the files are rock solid and even though they are only 8 bit, banding is not an issue.

Perhaps I should have been more specific and said usable latitude. That refers to how far the recorded image can be pushed or pulled in post. Assuming the camera is capable of delivering then 10bit adds rouhgly 2 stops to 8 bit and that to me is a big difference. You are quite unlikely to see any difference between a 10bit or 8bit recorded image. You might see quite a difference when grading it.
As said before though converting an 8bit image to 10 bit gains you nothing, the damage is done. On the other hand the EXs use 14bit A/D converters and 14 bit processing. The more of that preserved the better when it gets to post IF you intend to do heavy grading.

The banding issues I'm referring to can occur with the cinegammas. They compress 2 stops into the top 10% of the curve. Out of the camera you can get banding as a result, nothing at all to do with the compression.

The number of recording bits makes absolutely no difference to the amount of latitude the camera has, going from 8 bit to 10 bit will not give you 2 more stops of latitude. The latitude of the camera is governed by the sensors and sensor post processing, not the recording system.

While I won't argue that 10 bit with a suitably high bit rate should be better, I don't find banding to be an issue with first generation EX footage. The cinegammas make use of bits 235 to 254 which are not normally used, so while compression is taking place there are more 10% more bits being used. If anything the standard knee will cause more of an issue as this cuts off at 235, so all the knee compression takes place in the existing signal range.

What is clear from my real world tests is that the use of a NanoFlash at 100Mb/s, even though it is only 8 bit dramatically increases what you can do with the footage in post production before blocking and banding become an issue. With the EX and many many cameras it is not the but depth that is the problem but the bit rate.

Indeed the latitude of the camera is governed by the sensors etc, I did not say otherwise.
The dynamic range of the recording system is what I'm talking about. There's no getting away the simple fact of digital signal theory that 10bit has 2 stops more dynamic range than 8bit.

Understanding Dynamic Range in Digital Photography

8bit = 8 stops, 10bit = 10 stops. As said with the standard 2.2 gamma one could in theory fit 18 stops into an 8bit image but the results would be quite horrid. In practice around 11 stops is the limit however using hypergamma more highlights can be fitted into the top 10%. That's fine if they're simply specular highlights, not so good at all if you push sky and clouds into that region. I've certainly managed to 'toast' images with the EX1 by not being careful what I was pushing past the knee.

I don't know if recording 10bit will give us more useable latitude / dynamic range in the images recorded from an EX. Useable latitude is how much the image can be over or under exposed and still recovered. No one would suggest that it's very much at all with 8 bit video, the most optimistic figure I've read anyone suggest for 8 bit video is 2 stops and certainly overexposing at all is generally disasterous.

All of this is theoretical, until someone does meaningful tests I'd certainly not dismiss the idea that recording 10bit from the EX might give us more to work with in post. I think where you're missunderstanding me is you might be thinking I'm saying the image recorded in 10bit will look better than if it's recorded in 8bit. That's not the case at all.

That chart has to do with the sensor A/D bit depth Bob, it's not a measure of 8 vs 10 bit codec latitude. It shows the maximum number of different levels recordable. Alister is correct.

Any codec can cover the full range of any sensor. Black one end, white the other end. That's 100% of any dynamic range. The bit depth, 8, 10, whatever is just a fixed number of discrete luminance levels in between. 8 bit has 256, 10 bit has 1024.

Hi,

This maybe a silly question but..

Love the idea of the Nano Flash but does it have any real benefit if going down to SD? I edit in Final Cut and use Compressor to down convert. Would my SD DVD's look any better for it??

Or is it only worth it for broadcast delivery?

I don't think there is any quality benefit. It just makes the workflow simpler and faster as it removes the need to software down convert.