Some Questions About The Flash XDR & Future Development

[Mark Job]

Hi Mike & Dan:
It was great to learn you have now caught up on the backlog in development you had with the Nano. Along those lines, I have several questions about both existing features and future developments for the Flash XDR.

1. Now that 3:2 pull down removal has been added to the XDR for use with recording Canon's 24 F frame mode, is the simultaneous output which is available for realtime monitoring (HD-SDI out from XDR) also at the adjusted rate of 1080 24p, or is it still 1080i 59.94 Hz like it is out of the HD-SDI camera output ?

2. Are their any secret feature development projects on going for the XDR at this time ?

3. Can you give any time table for the development/enabling of the FireWire 1394a and RS-422 interfaces currently non-functional on our XDR units ?

4. Do you have a time table for the development/release of full uncompressed recording capability on the XDR ?

5. Will it be possible to enable an extra analogue recording setting in the XDR Audio Menu, which would allow recording @ 96Khz 24 bit sampling rate and bit depth ?

(I consider this setting is something which would go nicely with uncompressed video capture, since the application for this mode is cinema 35 mm film out)

6. Now that we have larger capacity - much faster speed CF media available, would it be possible to record Long GOP with much higher data rates (From about 220 to 500 Mb range) ?

* Caveate: NLE Support of course is a question.
** Caveate No. 2: For use in recording high quality dubs from HDCAM or HDCAM SR VTR's in Tv stations.

7. Would it be possible to offer a pure I frame MPEG Time Lapse Mode capture above 220 Mbps ?

[Mike Schell]

Hi Mark-
I can answer some of your questions, but not all. Some of the answers are confidential.

1) If you input 1080i60 and enable 3:2 pulldown, then you get 1080p24 out the HD-SDI.

2) I will leave this one to your imagination!

3) No, not yet. We're up to our eyeballs working on short-term new features. I'll post of list of the features planned for the next release shortly.

4) No, not yet.

5) Yes, this is possible, but not on the development schedule at this time. We can recommend using a seperate sound recorder and synchronizing to the XDR/nano via LTC.

6) Doubtful and really unnecessary. We have done a lot of image overlay tests with uncompressed and 100Mbps Long-GOP. The differences are so small, it's not worth the extra engineering work and storage requirements to bump up the rate. That said, we do offer 140 and 160 Mbps rates for those super complex, super high-motion events.

Note that the 100 Mbps Long-GOP rate is already superior to HDCAM quality.

7) Possibly, but why? Long-GOP is already 2-3X more efficient, so 160 Mbps Long-GOP = 320 Mbps I-Frame in quality.

Best-

[Mark Job]

Hi Mike:
You wrote: "6) Doubtful and really unnecessary. We have done a lot of image overlay tests with uncompressed and 100Mbps Long-GOP. The differences are so small, it's not worth the extra engineering work and storage requirements to bump up the rate. That said, we do offer 140 and 160 Mbps rates for those super complex, super high-motion events.

Note that the 100 Mbps Long-GOP rate is already superior to HDCAM quality.".....Yes. but not HDCAM SR, which is the VTR of choice over at CTV Network in Canada. The high data rate setting could be a Master Dub Only mode application. Think VTR replacement or the XDR as a bridge between VTR/File based post operations.

You Wrote: "7) Possibly, but why? Long-GOP is already 2-3X more efficient, so 160 Mbps Long-GOP = 320 Mbps I-Frame in quality."....Yes. It's more efficient but only so in the recording process, in fact, it's the inverse in post production editing. I-Frame beats Long -GOP MPEG in NLE handling every time. In fact, Long GOP was *never* designed as an editing format (I-Frame was) rather, Long GOP was designed as a delivery format. I-Frame looks just as good, only, you must record at data rates about 2 X that of Long GOP to get equal quality. A good example of this is your excellent 220 Mb I- Frame capture rate in the XDR. This looks pretty much equal to the 100 Mb Long-GOP (just a much higher data rate,thus less recording time per CF card). Now that we have ultra high speed 90 MB per second 64 GB CF Media - who cares ? Crank it up ! :-) :-)

[Alister Chapman]

Interestingly Sony themselves now officially state that XDCAM HD (50Mb/s 4:2:2 long GoP as per 50Mb/s XDR/NanoFlash) to be superior in quality to HDCAM. It's in the latest presentation materials from them. I can't see a lot of point in recording Long GoP at bir rates higher than 160Mb/s as the compressed is already pretty much indistinguishable from uncompressed. At 500Mb/s that's only around 2.5:1 compression and a real waste of a good codec. Also while the cards are getting faster and cheaper at higher bitrates you do still have to think about long term storage and archive plus off load speed will be slower.

100Mb/s does seem to be the sweet spot.

[Mark Job]

Hi Mike & Alister:
Perhaps I haven't been clear about the reasons why or the application for the higher data rate Long GOP settings. (??) Alister, I wasn't aware of any claims by Sony that their XDCAM HD Codec was superior to the quality of their HDCAM *SR* VTR's. You write HDCAM, but this has already been superceeded at the network broadcast level by Sony's new HDCAM SR VTR family. The HDCAM SR VTR records all the way up to 4:4:4 color space HD video. However, the majority of Network HD is still in 4:2:2, so we're good with the XDR.

.....Final Cut Pro 7.0 can now handle data rate video all the way up to 4:4:4 color space and the data rates which go with it (Including HD uncompressed). This is a VERY significant development, which no doubt, will propel FCP forward as a professional editing application.

.....Avid Symphony DS, Nitris DS can also work freely with 2K 4:4:4 color space DI files.

.....The opportunity here for the XDR to act as a bridge between online and clip based editing, and as a VTR replacement, is obvious to me. If I could record in Long GOP (Because, as Mike often points out, it's more efficient than I-Frame encoding), and take dubs from an HDCAM SR VTR onto the XDR @ a Long-GOP based high data rate, then pull the CF card an import that clip into FCP 7.0 or Avid Media Composer, then suddenly I could now re-edit and update a Network program *without the need of a $110,000.00 US Avid Symphony Nitris or DS to do it on without any loss of quality !* Wow ! Would this be sweet ! I then could feed the re-cut back out to the CF card and pop the CF card back into the XDR and play the program back out via HD-SDI into another HDCAM SR deck.

.....The only possible compromises I can see in all of this is twofold.

A) I don't think the 100 Mb Long-GOP setting (As good as it indeed is) is enough for Master quality VTR replacement or dubs from an HDCAM SR VTR. (??) * I need to test the 140 and 160 settings. At least why not give us a 220 Mb Long-GOP setting equal to what CD has already given us for I-Frame MPEG :-)

B) What is the bit level of the XDR ? Is the XDR 8 or 10 bit capable ? I asked this question before, but I don't remember if I got a clear answer or not. (??)

[Dan Keaton]

Dear Mark,

The Flash XDR is 8-bit.

When we add uncompressed, we will support 10-bit.

But, I want to be real clear:

A lot of people want 10-bit since they saw banding in 8-bit.

While this is attributed, by many, to be due to 8-bit versus 10-bit, we feel that it is most like caused by the codec. A bit starved, inefficient codec can produce banding.

No one has ever reported seeing color banding in our Long-GOP 100 Mbps footage.

To the best of our knowledge, no one has ever produced banding in a torture test of our 100 Mbps Long-GOP footage.

Some people have been lead to believe that 8-bit is 256 colors.

It is actually (approximately 256 times 256 times 256, less a few) or around 16.7 million colors.

Planet Earth, with all of its color and glory was broadcast in 8-bit, as is every other television program.

[Dan Keaton]

Dear Mark,

The Flash XDR and nanoFlash at 100 Mbps Long GOP is better than HDCam.

It is not better than HDCam SR.

The Flash XDR and nanoFlash record 4:2:2, not 4:4:4, which the HDCam SR does.

One flavor of HDCam SR is 440 Mbps, and another is 880 Mbps.

Our very best image quality is 160 Mbps Long-GOP (but remember, we recommend 100 Mbps Long-GOP). HDCAM SR is Intraframe not Long-GOP, so we have an advantage in codec efficiency, but they have an advantage in bit-rate.

If recording 4:2:2 to both, we feel that it would be a close race, visually very close, but HDCam SR would win. But a HDCam SR tape deck is not $2,995 either.

Just remember, if no one can tell the difference between live HD-SDI and recorded HD-SDI, as recorded on the nanoFlash at 100 Mbps Long-GOP, isn't that good enough?

[Mark Job]

Hi Dan:
Yes. The 8 bit banding thingy has persisted very heavily in network Tv broadcast circles. In reality, 16.7 million colors is probably way more than enough :-) Hey, if it works for planet earth then who am I to disagree ? ;-)

....Now you wrote that the XDR is 8 bit, but when you go to uncompressed you will "support 10 bit." I want to be crystal clear here, so I must ask what do you mean by "support 10 bit ?" Are you expressing the idea a firmware upgrade will allow the XDR to accept a greater bit depth video signal thereafter, or do you mean something else. Can you elaborate ?

[Mark Job]

Originally Posted by Dan Keaton

Dear Mark,

The Flash XDR and nanoFlash at 100 Mbps Long GOP is better than HDCam.

It is not better than HDCam SR.

The Flash XDR and nanoFlash record 4:2:2, not 4:4:4, which the HDCam SR does.

....No this is incorrect. The Sony HDCAM SR VTRs record *both* 4:2:2 and 4:4:4: Most networks use the HDCAM SR decks @ the 4:2:2 color space setting @ 440 Mbps.

Originally Posted by Dan Keaton

One flavor of HDCam SR is 440 Mbps, and another is 880 Mbps.

....Up the Long-GOP data rate to 440 on the XDR and the XDR is now *Equal* to the HDCAM SR technology @ the 4:2:2 mode 440 Mbps setting.

Originally Posted by Dan Keaton

Our very best image quality is 160 Mbps Long-GOP (but remember, we recommend 100 Mbps Long-GOP). HDCAM SR is Intraframe not Long-GOP, so we have an advantage in codec efficiency, but they have an advantage in bit-rate.

If recording 4:2:2 to both, we feel that it would be a close race, visually very close, but HDCam SR would win. But a HDCam SR tape deck is not $2,995 either.

....No it would be an exact match with uncompressed and 10 bit enablement, plus you are already only a razor's edge away from accomplishing an exact data rate match with your Long - GOP XDCAM HD codec. - Just follow through with a little more data rate and you've done it with the XDR/nano. Turn on that RS-422 interface and Convergent Design is there with every TV Network wanting the XDR as a solid state VTR supplement allowing low cost clip based editing without having to buy the very high end NLE's to do it ! What a great cost cutting tool !

Originally Posted by Dan Keaton

Just remember, if no one can tell the difference between live HD-SDI and recorded HD-SDI, as recorded on the nanoFlash at 100 Mbps Long-GOP, isn't that good enough?

...I can. Network technicians can. Some producers can. I think 100 Mb looks great, it is close, but it's not quite enough. Perhaps the 160 Mbps Long - GOP is, or how about a possible 220 Mbps ? Can you do 440 Mbps on the XDR/nano ?

[Mike Schell]

Originally Posted by Mark Job

Hi Mike:

You Wrote: "7) Possibly, but why? Long-GOP is already 2-3X more efficient, so 160 Mbps Long-GOP = 320 Mbps I-Frame in quality."....Yes. It's more efficient but only so in the recording process, in fact, it's the inverse in post production editing. I-Frame beats Long -GOP MPEG in NLE handling every time. In fact, Long GOP was *never* designed as an editing format (I-Frame was) rather, Long GOP was designed as a delivery format. I-Frame looks just as good, only, you must record at data rates about 2 X that of Long GOP to get equal quality. A good example of this is your excellent 220 Mb I- Frame capture rate in the XDR. This looks pretty much equal to the 100 Mb Long-GOP (just a much higher data rate,thus less recording time per CF card). Now that we have ultra high speed 90 MB per second 64 GB CF Media - who cares ? Crank it up ! :-) :-)

Hi Mark-
I see your point, we'll consider a higher I-Frame rate in the future.

Best-

[Aaron Newsome]

Sorry I wasn't able to join this conversation in progress. There's a few things I disagree with in this thread but I'll play along. 8 bit versus 10 bit is a big deal, a bigger deal than I think this thread will represent.

I don't think anyone has ever thought an 8 bit codec only rendered 256 colors. That's ridiculous.

It really boils down to what are you shooting. The 256 levels per channel is significant, if the majority of your image lies in one of those channels. Like greenscreen work for example. Getting fine edge detail on greenscreen work can be maddening, especially if you don't have full control on the shoot.

Even when your image is evenly spread across all the channels, like shooting smoke on against black for compositing with a luma key. Those extra bits will help get a much better final product.

What I'm most interested in is what Dan said above, that 10 bit is technically possible on this hardware (XDR). Uncompressed of course, since MPEG is an 8bit codec. Is this true? If it's technically possible to do 10bit uncompressed, I would hope this moves to the top of the priority list in development.

My guess is, the RAID'ing of the CF cards would be a requirement for this though since there is no human affordable CF card fast enough to do uncompressed.

[Mark Job]

Hi Aaron & Dan & Mike & Everybody:
Aaron Wrote: "What I'm most interested in is what Dan said above, that 10 bit is technically possible on this hardware (XDR). Uncompressed of course, since MPEG is an 8bit codec. Is this true? If it's technically possible to do 10bit uncompressed, I would hope this moves to the top of the priority list in development."

........Yeah, me too. Specifically, I want Dan to elaborate a little more on what he means by saying CD will support 10 bit in the XDR when they go full uncompressed recording option. What exactly does that mean ? Is it true that the XDR *can* do 10 bit video with a firmware upgrade ? If so, WOOOOOOOOOOOWWWW !!!!!!!

Aaron wrote: "My guess is, the RAID'ing of the CF cards would be a requirement for this though since there is no human affordable CF card fast enough to do uncompressed.
Today 11:00 AM"

........I believe this is incorrect. The 90 MB Sandisk 64 GB cards should be fast enough for a stream of 10 bit uncompressed HD. Striping four x 64 GB 90 MB cards is definitely a really good idea. Striping four 60 MBps 64 GB CF cards is a good idea too !

[Dan Keaton]

Dear Aaron,

Two companies, one frequently, and another one in the past, asked:

What would you want 1024 crayons or 256 crayons? This implies that 8-bit only has 256 colors, unless one clear states 256 levels per color.

Yes, the Flash XDR is designed, hardware wise, to allow for a future full uncompessed, 10-bit option. This applies to the nanoFlash also, now that very fast, large capacity CompactFlash cards are available.

So uncompressed is feasible as a firmware update. Please note, that from day one, we always discussed this as an extra cost feature. We have also quoted $995 (US) for this feature.

Yes, with both the Flash XDR and the nanoFlash, we expect that we will need to stripe the data across at least two cards to achieve the raw bandwidth necessary to do uncompressed.

Aaron, it was nice speaking with you!
(Aaron called just as I was writing this.)

[Aaron Newsome]

Originally Posted by Mark Job

........I believe this is incorrect. The 90 MB Sandisk 64 GB cards should be fast enough for a stream of 10 bit uncompressed HD. Striping four x 64 GB 90 MB cards is definitely a really good idea. Striping four 60 MBps 64 GB CF cards is a good idea too !

i think you skipped right over the "humanly affordable" part. 4x 64GB-90MB/s cards = $$$$

but I guess it's still cheaper than an S.two box :-)

[Alister Chapman]

If I had meant HDCAM SR I would have said HDCAM SR. I was referring to HDCAM which has been the high quality standard for broadcast HD acquisition for several years. However long GoP, while I agree not designed for acquisition (mainly because the processing power need wasn't available at the time) MPEG 2 at 50Mb/s is now regarded by Sony and others as superior to what was previously the de-facto standard for HD shooting.

It is not always true that I frame is easier to decode than Long GoP. At high bitrates to decode I frame only the CPU has to load every frame, requiring large bus bandwidth to decode it. With long GoP the CPU only has to load the I frame and then only needs to use the much smaller B and P frames to decode the image. With fast, modern CPU's in many cases this takes requires a lot less CPU bus bandwidth than having to load every single full frame. With well written software codecs there should not be that big a difference between I frame and long GoP in terms of processing speed. The higher the bitrate the closer the gap between I frame and long GoP gets. With very high bit rates long GoP can become easier and quicker to decode.

Yes, 10 bit is better than 8 bit, IF you give it all the extra bandwidth it needs and you have the hard drives to handle it. The results I am getting from the NanoFlash are nothing short of astounding. I can grade, edit and encode my footage very hard and it holds up solidly and robustly. I can encode my footage to 10 bit ProRes, but I can't see any advantage in my end results. I am convinced that 100Mb/s Mpeg 2 is the sweet spot. Robust enough for heavy post production yet small enough for economic storage and use over a network. Personally I don't want enormous 10 bit uncompressed files that need very fast, very large raid arrays to store and edit. You'll find 2 streams of uncompressed 10 bit HD a lot harder to work with (in computer performance terms) than two streams of compressed long GoP, primarily because of the vast amount of data your dealing with. I've been able to record uncompressed 10 bit via my decklink card for chroma key shoots in my studio for several years, but the advantages over 8 bit are so small that it's never been worth the trouble. However the difference between 35 Mb/s (EX) or 50Mb/s (HD 422) and 100Mb/s NanoFlash is much more significant.

Of course these are just my views and you don't have to agree :)