For most circumstances I agree, but if doing color grading for film out, 10bit can help. But then again I could build a bmd/computer/raid based capture setup for that. Though the Flash XDR recording at 160mb would be preferable. I hope you folks can deliver on that, but I certainly don't consider that a deal killer.
In reality, what will the Convergent Design Nano Flash do for me?
- Thread starter Buck Forester
- Start date
For most circumstances I agree, but if doing color grading for film out, 10bit can help. But then again I could build a bmd/computer/raid based capture setup for that. Though the Flash XDR recording at 160mb would be preferable. I hope you folks can deliver on that, but I certainly don't consider that a deal killer.
I attended a Panasonic presentation a few weeks ago about the new 2700 and 3700 cameras. They put up some graphs that showed that the AVC-intra codec was significantly better than DVCpro HD. The pictures backed this up, too. Beautiful pictures with very smooth gradations. I don't remember seeing any samples of fast pans, though.
Someone needs to compare these recorders head to head when the HPG20 comes out. Maybe split an HD-SDI signal from 1 camera to go to both recorders, or record an uncompressed signal to a computer, then output the same signal to both recorders.
Flash XDR and Nano Flash are made by Convergent Design. You can find more info at their website http://www.convergent-design.com/ They also have discussion forums at other sites. Just follow the links from their page.
Mike Schell, who posted on this thread earlier, works for them, and I'm sure he can help with any questions you may have.
Mike Schell
Active member
Hi Zeke-
Yes, the 160Mbps I-Frame only is definitely on the firmware update roadmap, but probably not until Jan 09. We have MXF and 24p pull-down removal to complete.
I still believe that the 100Mbps Long-GOP will deliver the best video quality, but we'll let you make the final choice, as it's a simple menu selection to choose beteeen 100Mbps Long-GOP or 160Mbps I-Frame only.
Mike Schell
Convergent Design
Mike Schell
Active member
I am anxious to compare our 100Mbps Long-GOP to the AVC-I CODEC. We have images posted on our website that clearly show our 100Mbps outperforming DVCProHD. This is not surprising, since our CODEC is full-raster (1920x1080), while DVCProHD is 1280/1440 x 1080. Also, our CODEC is Long-GOP which is about 2-3X times more efficieint than I-Frame only.
Naturally, this comparison gets more complex when you compare our MPEG2 100Mbps Long-GOP 8-bit CODEC to the AVC-I (MPEG4) 100Mbps I-Frame Only 10-bit CODEC. MPEG4 should be more efficient that MPEG2, but this advantage rolls off with higher bit-rates (it's significant at 10-20Mbps, but much less so at 100). And 10-bit should be better than 8-bit, but this difference is also reduced through any compression.
I suspect the overriding advantage of the Long-GOP recording will make our CODEC very competitive aganist the AVC-I CODEC. But, the only way to answer this question is to run some side by side tests.
Mike Schell
Convergent Design
Stevet
Veteran
I have a feeling both of those codecs will be so good it will be hard to determine the best.
Afterall, we've seen the XDR 100mbps long-GOP compared to uncompressed. I really could not see a difference by eye that is...
I looked at the subtracted frames under no light and did not see really any visual difference. These were the fast motion frames posted.
Times are getting better for all!
Afterall, we've seen the XDR 100mbps long-GOP compared to uncompressed. I really could not see a difference by eye that is...
I looked at the subtracted frames under no light and did not see really any visual difference. These were the fast motion frames posted.
Times are getting better for all!
GuyB
Well-known member
The MPG4 compression would also suck a lot more processing power to encode, and from what I hear it is bit*h to edit and most are having to go back to uncompressed.
I imported a 50Mbps MPG2 MXF file (to simulate the PDW-700) from Clip Browser 2 into Liquid 7.2 and it played it native no problems on my laptop.... I would expect similar results with 100Mbps and Liquid 7.2.
I imported a 50Mbps MPG2 MXF file (to simulate the PDW-700) from Clip Browser 2 into Liquid 7.2 and it played it native no problems on my laptop.... I would expect similar results with 100Mbps and Liquid 7.2.
EIREHotspur
Well-known member
Mike
Your products look very good and the side by side comparison are impressive because thats what counts in the end after all......picture quality.
Can I ask a silly question.
Why can't you shrink the product/codec down to the size of an SDHC card or work with a company who could get it onto such cards?
Having ditched the Sony 60gig HD, which didn't work as was not reliable...I am using SDHC cards now.
It is cumbersome to have to put a HD unit on top of an EX3.
Your products look very good and the side by side comparison are impressive because thats what counts in the end after all......picture quality.
Can I ask a silly question.
Why can't you shrink the product/codec down to the size of an SDHC card or work with a company who could get it onto such cards?
Having ditched the Sony 60gig HD, which didn't work as was not reliable...I am using SDHC cards now.
It is cumbersome to have to put a HD unit on top of an EX3.
Barry_Green
Moderator
I can edit AVC-Intra in realtime on my laptop. It isn't that hard. AVC-HD is significantly more processor-intensive because it's a long-GOP codec, so I can't get AVC-HD in realtime, but AVC-Intra is realtime on EDIUS 4.6.
Jim Arthurs
Well-known member
Hi Steve... I don't currently have access to an XDR, but at the time I had recorded something at a slow pan speed and high detail to both the SxS media and the XDR... here's a link to a layered Photoshop file you can download and play with. The first glance visual differences are subtle, but you will notice a slight softness when you flick between them on the 35Mb/sec image compared to the 100Mb/sec. Of course differencing can be done between them to reveal the whole story...
http://ftp.datausa.com/imageshoppe/outgoing/EX1/SLOW_PAN35MbVRS100MB.zip
This is a very slow pan, coming to a stop 60 frames after this grab...
Regards,
Jim Arthurs
GuyB
Well-known member
Is that editing AVCIntra natively or is Edius transcoding to its own intra codec first?
Jim Arthurs
Well-known member
In case anyone is interested, I did do some tests a while ago comparing uncompressed 8bit vrs uncompressed 10bit coming from the EX1's HD-SDI out while live... I have a thread over on Chris Hurds site, but didn't get any feedback, so I'll recap here.
The bottom line is that it's VERY hard to detect any *worthwhile* difference between them... This could just be an EX specific problem, or maybe a comment on the whole 8 vrs 10bit debate sourcing in general.
What I did was set up a grey card with a subtle gradient of light across it... something like 45 IRE to 55 IRE, with the hope of producing some banding over the image and seeing if the banding was less "steppy" on the 10bit material.
I took the 10bit uncompressed image through a heavy levels adjustment, effectively remapping 45 IRE down to black and the 55 IRE up to white... all keeping the work in Digital Fusion in 16bit float space. I did the same image processing steps on the 8bit uncompressed image, but keeping it in an 8bit space the whole time. Finally, after levels, I dropped the 10bit down to 8bit and split screened it over the 8bit. Here's the results...
http://ftp.datausa.com/imageshoppe/outgoing/EX1/EX1_8bitVRS10bit_Test.png
Here's a zip of the source files, and another set of 8 vrs 10bit images of colored pencils to play with...
http://ftp.datausa.com/imageshoppe/outgoing/EX1/EX1_8bit_vrs_10bit_uncompressed.zip
I don't consider this any sort of final word on the debate, but I've done this a couple of times now and reached the same personal conclusion. I'd love to see someone else take a stab at this sort of test and report back.
What I'm looking for is an example that shows a real-world practical benefit to the 10bit advantage. Please know I'm not arguing the usefulness of a 10bit intermediate file format for color correcting, just the benefit of 10bit original source recording over 8bit in a practical, real world example.
Regards,
The bottom line is that it's VERY hard to detect any *worthwhile* difference between them... This could just be an EX specific problem, or maybe a comment on the whole 8 vrs 10bit debate sourcing in general.
What I did was set up a grey card with a subtle gradient of light across it... something like 45 IRE to 55 IRE, with the hope of producing some banding over the image and seeing if the banding was less "steppy" on the 10bit material.
I took the 10bit uncompressed image through a heavy levels adjustment, effectively remapping 45 IRE down to black and the 55 IRE up to white... all keeping the work in Digital Fusion in 16bit float space. I did the same image processing steps on the 8bit uncompressed image, but keeping it in an 8bit space the whole time. Finally, after levels, I dropped the 10bit down to 8bit and split screened it over the 8bit. Here's the results...
http://ftp.datausa.com/imageshoppe/outgoing/EX1/EX1_8bitVRS10bit_Test.png
Here's a zip of the source files, and another set of 8 vrs 10bit images of colored pencils to play with...
http://ftp.datausa.com/imageshoppe/outgoing/EX1/EX1_8bit_vrs_10bit_uncompressed.zip
I don't consider this any sort of final word on the debate, but I've done this a couple of times now and reached the same personal conclusion. I'd love to see someone else take a stab at this sort of test and report back.
What I'm looking for is an example that shows a real-world practical benefit to the 10bit advantage. Please know I'm not arguing the usefulness of a 10bit intermediate file format for color correcting, just the benefit of 10bit original source recording over 8bit in a practical, real world example.
Regards,
Mike Schell
Active member
The SDHC cards are limited to 32GB, whereas 64 and 100GB CF cards have already been announced by several manufacturers. Also, the CF cards still offer the highest performance (45Mbytes/sec), although SanDisk now has an SDHC card with 30Mbytes/sec R/W.
In all honesty, I doubt we can make the nanoFlash much smaller, even using SDHC cards. There is a certain amount of circuitry outside of the memory cards that will always be required to make the product work.
Mike Schell
Convergent Design
Stevet
Veteran
Jim,
thank you! for this example.
If this is a slow pan (and I'm sure it is), the difference was larger than I expected!
This is exactly what's needed for many to see the benefits if using this codec over the stock HQ 35mb/s codec. The difference is more obvious looking at a 24" 1920x1080 monitor. Like you said, subtracting the layers tell the story. If we're able to see a difference with this slow motion, a faster pan would of yielded an even more obvious difference.
This may be just what Buck is looking for. The difference is subtle. But since the EX has such a high detail image, giving up "some" detail during movement and also adding to that normal motion blur, this may be what Buck needed to hear.
Yes, I remember seeing these 10bit verses 8bit. I agree the benefits from these are not obvious. I do not have much experience with 10bit video. I'd like to see some examples that clearly show a difference. I believe I've seen some grading examples on Cineform's website, but it's been a while.
Thanks again Jim!
Last edited:
Barry_Green
Moderator
Native, playing straight from the card. I hate transcoding. :thumbsup:
Jim Arthurs
Well-known member
Hi Steve... here's the clip I pulled the grab from so you can see the speed of the pan... I think I pulled about 2 seconds before dead stop...
http://ftp.datausa.com/imageshoppe/outgoing/EX1/SLOW_PANexample.mov
In regards to the 8bit vrs 10bit test... I may try it again, but using an IRE ramp range of say, 15 to 25 instead of higher up the food chain (bit-wise) at 45-55. As we get closer to the noise floor the amount of bits per f-stop fall off significantly and any difference should be greater. Yet, the closer you are to the noise floor the more... er... noise you have, masking the banding potential. It's a double-edge sword in testing.
We know for a fact that those extra bits are real coming out of the EX1, not just padding from 8 up to 10. It would be nice if they made a real world difference that we could see and judge.
Regards
EIREHotspur
Well-known member
So the long and the short of these tests are that pans are smoother on the XDR files??