FS7: RAW SDI output recorded in 4KDCI@50p on Shogun Inferno: good news!

moldcad

moldcad

Veteran
OK guys - time for a little summary of my experience so far with the FS7 - Shogun Inferno combo...

As you might remember, I was anxious about the expected quality of the RAW SDI output as many sources have claimed it's not good enough; people kept reporting banding etc. and even suspected the RAW output is not 12 bit (some even said it was less than 10-bit, judging from the color banding and other artifacts, particularly in shadows). Even Dennis (@starcentral) in his PM warned me about it, suggesting my purchase of the Inferno (and the XDCA unit to eable RAW output) might not be the best investment ...

But I'm now after a series of tests and am very glad to report that actually, the Inferno and XDCA - while of course expensive - have been one of the best investments of mine after the purchase of the FS7 last year. I have recorded the same, difficult lighting scenery (a semi-dark room and against quite bright window) in the following settings (all using S-log3 and S-Gamut3.Cine and recording to DNxHR HQX on the Shogun):

1. 2160p25: using HDMI output, which at 25 fps is full 10-bit, 422
2. 2160p50: using HDMI output, which Sony Prime Support confirmed to only be 8-bit, 420 at 50 fps
3. 4KDCIp50: using SDI RAW output

Well - the good news is that comparing 2 and 3, I can clearly see a much better quality of the 10-bit DNxHR clips when feeding my Inferno from RAW SDI than when only using the crippled HDMI output at 50 fps.:

- there are no banding or other artifacts whatsoever with scenario 3
- with scenario 2, I can clearly see banding due to the FS7 output signal being 8bit only.

The banding is very specific in nature; I could describe it as color shades' "peeling" visible on single color objects with varying light intensity falling on them. There is no "continuous" change from one color shade to the other; instead the boarders between the color shades form noise-like, busy pixels... Only the noise isn't the well-known "gain/high ISO noise", as I can see it even on properly lit areas of white! So the only explanation is "color banding noise".

The same areas (as well as even the deepest shadows) are completely noise or banding-free in the RAW SDI output version :)

Piotr
 
With an FS7 and the XDCA I would not care about your option 2) any more. What do you expect from 8bit? Even the internal XAVC I recording will be much better then that.

The scenarios to compare should be
a) record internally with 10bit to XDCA
b) record extrenal with the XDCA to 12 bit linear raw with the Shogun
c) record external with the XDCA to raw and convert it in the Shogun to 10bit ProRes
d) record external with the XDCA to raw and convert it in the Shogun to 10bit DNxHR

For me the comparision between a) and b) but also a) and c) or even c) and d) would make sense.

And some major players like Alister Chapman beliefs today, that for a) and b) the version a) is enough. So I wonder if he is wrong or if there is still an advantage to use the 12bit raw. Maybe for the HDR grading there will be an advantage for b)
 
When it arrives with Shogun fw update, I'll of course test option b) - but just out of curiosity; recording RAW output while converting it to DNxHR HQX is more than enough for me.

The goal of my summary has been to inform those still under the impression the FS7 RAW output is somehow flawed that it's not the case at all. Also, including my option 2) in the comparisons only has been done for the sake of experiment, and it indeed confirms that SDI RAW output actually IS of much greater bit depth than HDMI at framerates >30 fps (some people doubt it, saying that the banding in shadows they see recording RAW to Odyssey would indicate it's not 12, not even 10bit - but more like 8 or even 6bit only). My experiment, comparing this option with HDMI output, clearly shows those reservations were not valid.

Do you understand the choice of my experiment options now, Wolfgang? I hope you do :)

Piotr
 
Oh - I almost forgot to mention that comparing results of scenarios 2 and 3 in my OP also had another important goal:

As I'm still debating over my B-camera choice, trying to decide between GH5 (MFT only, but 10 bit 422 50p) and A7 (Full Frame/APS-C option, but only 8 bit 422 externally) - this experiment proved beyond any doubt 2 important things:

- as I have already pointed out, that my FS7 RAW SDI is going to yield much better results for 50p than I could possibly get using Shogun with only 8bit HDMI output

- also, that even though external recording to the Shogun from an A7 camera would certainly yield better results at 4:2:2 than internal A7 footage which can only be 8bit 4:2:0, it woulld still not be on par with full 10-bit FS7 footage and color-matching them would be difficult and introduce a lot of artifacts (mainly the color banding I described)

So - unless Sony makes it with the new A7 capable of 10bit on time - it's GH5 which is going to be my B-cam, in spite of all the inconvenience of different glass and alike...

Piotr
 
There has been a video recently from Atomos, that the GH5 will be able to deliver UHD 10bit 422 50p by the hdmi interface, and that the Shogun will be able to record that UHD 50p as 10bit 422. So from this side the GH5 is an interesting option, much more interesting then the outdaten 8bit A7s.

There will always some people who still think that the earth is not a globus but a plate. ;) Remember, it was me who shocked some guys in the forum with the news that the FS7 deliveres 8bit 420 by the hdmi interface - for UHD 50p/60p. But nobody thought that this will be true.

The raw capability of the Shogun delivers CDNG, what can be graded great within Resolve. The opinion seems to be that it is somewhere between 11-12 bit, due to some compression losses to be able to deliver it by 3G SDI. So sure, it has some small potential reserves compared to 10bit ProRes. But I think it is enough to be able to record UHD 50p via the raw path/XDCA as 10bit 422. CDNG tends to become huge, since it is not a compressed raw.
 
As one of the people that did the last round of tests showing the FS7's problems with RAW, I would strongly suggest you Wolfgang's advice and do a side-by-side of the 4K 50p RAW out vs the 4K internal XAVC-I.

I haven't heard of any dramatic changes to the RAW since I did my tests, but I'd be thrilled if they did in fact fix the problems. The primary thing that we saw in 4k RAW was a substantial amount of VERY ugly noise in the shadows. I'm attaching a photo of the RAW output (at 4k 24p) with proper exposure that came from my last round of tests. If you expose up by 2 stops, you dramatically reduce the amount of this noise, but you also lose the head room.
RAW Proper Exposure_1.2.1.jpg
 
wschmid said:
The opinion seems to be that it is somewhere between 11-12 bit, due to some compression losses to be able to deliver it by 3G SDI. So sure, it has some small potential reserves compared to 10bit ProRes. But I think it is enough to be able to record UHD 50p via the raw path/XDCA as 10bit 422. CDNG tends to become huge, since it is not a compressed raw.
Click to expand...

Not exactly. The RAW output on the FS7 is 12bit LINEAR. The Prores is a 10-bit LOG.

12-bit linear is mathematically incapable of recording or representing the dynamic range that the Sony FS7 is capable of. I will take 10-bit log over 12-bit linear every day of the week.
 
Hi Christopher, and thanks for chiming in.

I'm not sure whether or not Sony have improved RAW quietly in the meantime, but I can tell you there is no comparison between internal XAVC-I and DNxHR HQX on Inferno fed via RAW SDI from my FS7. The Inferno clips are waaay cleaner - there is no noise whatsoever; the XAVC-I is a nice codec but the implemented version has too high compression ratio, and it shows (the infamous compression artifacts in large areas of single color and low detail, like clear blue sky). I'm not even going to try posting clips or screen grabs to prove my point; it's like night and day.

Cheers

Piotr
 
wschmid said:
And some major players like Alister Chapman beliefs today, that for a) and b) the version a) is enough. So I wonder if he is wrong or if there is still an advantage to use the 12bit raw. Maybe for the HDR grading there will be an advantage for b)
Click to expand...


I know this sounds horrible but I do wish people would stop getting their information from this man as though he is infallible.

He's not a 'major player'. He's a bloke that demos gear for Sony and makes little videos and his opinions carry no more weight than the majority of experienced DVXuser forummers.
 
ChristopherA said:
Not exactly. The RAW output on the FS7 is 12bit LINEAR. The Prores is a 10-bit LOG.

12-bit linear is mathematically incapable of recording or representing the dynamic range that the Sony FS7 is capable of. I will take 10-bit log over 12-bit linear every day of the week.
Click to expand...
Even if the raw is 12bit linear, that is reduced due to some compression losses - that is why I said it is somewhere between 11-12bit only.

And sure there are problems with the DR of 14, to get that in 12bit linear raw. I know that. But I do not see why ProRes/DNxHR should be better, if it is generated from the raw. If there are issues in the raw, then we should find the same issues in the ProRes generated from the linear raw too.

Maybe the ProRes recorded from the HDMI interface is better. BUT this will be 8bit 420 only for UHD 50p/60p. So it can be better only for up to UHD 30p.
 
Ben Scott said:
I know this sounds horrible but I do wish people would stop getting their information from this man as though he is infallible.

He's not a 'major player'. He's a bloke that demos gear for Sony and makes little videos and his opinions carry no more weight than the majority of experienced DVXuser forummers.
Click to expand...

Oh man, I am not the advocate of Mr. Chapman, I never met him. But I do not have the impression that his training material is wrong. So either you are able to explain us why you think so, but please based on facts. Or you better shut up.
 
The countless times he's been wrong should do it. Most recently when he was telling FS5 owners in a big patronising blog post their issue with macro blocking was down to stupid user error and they should get their act together and stop moaning, right up to the point where Sony admitted it was a fault and released a firmware fix. Turns out being paid by Sony to promote their products makes you predisposed to defend their products even when there's an issue with them. He did similar over #12bitlinearraw-gate.

He's not a major player in production in that I (and my colleagues) have never ever heard of him or his company in the UK unless it is in relation to him marketing a Sony product, and it's a relatively small world over here.

I like the implied threat in your last sentence though, although combined with your signature it sends mixed signals :)
 
The complaint about RAW output has been mainly regarding RAW output recorded as RAW. Moldcad's findings are about RAW output but recorded as DNxHR. I'm not sure how those findings are proving that "those reservations were not valid".
 
Indirectly, they do - DNxHR encoding is so clean it's hard to imagine its source (i.e. RAW from FS7) being as bad as has been reported.

Piotr
 
moldcad said:
Indirectly, they do - DNxHR encoding is so clean it's hard to imagine its source (i.e. RAW from FS7) being as bad as has been reported.

Piotr
Click to expand...

When recording RAW to DNxHR in Inferno, doesn't it convert the signal from linear to log before recording?
 
My knowledge is by far not enough to answer this question. Another one which I've been wondering (and would appreciate somebody with enough knowledge to chime in - but with hard fact rather than speculation): why is it still possible to impose a gamma curve on the FS7 RAW footage? I will sound naive, but I thought RAW is linear by definition - and yet I'm happily using my SDI RAW signal in CineEI mode, and S-Log3/S-Gamut3.Cine applied... With this in mind, the footage encoded into DNxHR by Inferno has all the characteristics of S-Slog3 footage! Which is what I needed for those particular experiments of hunting the alleged noise in footage with high dynamic range (shooting into bright window in a darkiroom). Gurus, please explain :)

Piotr
 
Im no expert but: RAW can be linear or log. The FS7 sends out linear 12bit (thats why it sucks even though you claim otherwise) but the recorders in order to interpret this into prores or another non RAW codec has to apply a gamma and gamut. If the RAW is in log that gamma will already be in baked into the RAW in order to get a better distribution of luminance levels in the material. The varicams for example use 10 or 12 bit RAW in log that looks great.
 
No it doesn't. It sends 12-bit linear raw which then has to be interpreted by your recording device which will then apply a gamma curve to it.

Every camera outputting raw needs something to apply a gamma curve to it before it is recorded into a useable format.

Otherwise it is just proprietary data that cannot be read or viewed by anything.
 
If that's the case I must admit the Inferno applies the gamma very well, and encodes DNxHR brilliantly. However, Ben - how does the Shogun "know" which gamma curve to apply? The only way I can imagine is that the RAW stream has a flag, telling Atomos encoder that the user has chosen the S-Log3/S-Gamut3.Cine for his RAW footage...

Piotr
 
You must log in or register to reply here.
Back
Top