Z9 on test - N-RAW no better than H.265?

[Andrew Reid]

On 3/21/2023 at 4:20 PM, Kino said:

All this proves is that NRAW may not be optimized yet in N-LOG gamma on Resolve, as other users have reported much greater DR using different gamma profiles

Sure it's early days, N-RAW might get better software support as we go along... H.265 certainly did.

On 3/21/2023 at 4:20 PM, Kino said:

It is a very new raw format and not user friendly just yet. Your article also fails to take into account the impressive color depth of NEV files, which Resolve lists as 14 bit RAW. Is Nikon secretly giving us 14 bit RAW? All I can say is that the NRAW files display rich color and tonality that I've rarely seen on other mirrorless cams shooting 12 bit RAW

I don't think it does 14bit RAW in video, it'll be 12bit to achieve the higher frame rates especially above 30p.

[Kino]

On 3/22/2023 at 12:31 PM, Fairkid said:

I tried the Z9 on a few jobs but didn't keep it in the end. The colours are beautiful and I loved the ergonomics and battery life but as you mentioned the workflow wasn't ideal. I just couldn't get decent dynamic range out of it using the supplied Nikon conversion LUTs. As it's a bit of a niche camera there wasn't much info out there on how to get the best out of it. 

Nikon really dropped the ball when they released this camera with N-Log, which is not capable of maximizing the sensor's DR. They need to release "N-Log 2" in firmware if they can, since all the non-RAW formats in the camera still suffer from N-Log's limitations. The tests linked above demonstrate the significant difference between N-RAW processed in N-Log gamma and N-RAW when using ARRI Log C in Resolve:

 

[Kino]

5 hours ago, Andrew Reid said:

Sure it's early days, N-RAW might get better software support as we go along... H.265 certainly did.

I don't think it does 14bit RAW in video, it'll be 12bit to achieve the higher frame rates especially above 30p.

Unlike the Canon R5, which switches to a 12 bit readout when engaging the electronic shutter for photo and video, the Z9 electronic shutter is 14 bit, as that is all it has. Therefore, it is possible for the 14 bit readout to carry over to video, which Resolve apparently confirms at 24 fps in this example:

Of course, it could still be a 12 bit source wrapped in a 14 bit file, or there could be an error in how Resolve interprets the files. However, looking at the N-RAW footage, I'm seeing a color depth and tonal range that I don't associate with 12 bit cameras. This is very close to 16 bit R3D files or 14 bit RAW photos:

 

 

 

 

[TomTheDP]

If it is 14 bit it should be identical to 14 bit RAW photos. Does the RAW video and photos have the same dynamic range? 12.5 stops with noise reduction isn't the greatest.

[Kino]

33 minutes ago, TomTheDP said:

If it is 14 bit it should be identical to 14 bit RAW photos. Does the RAW video and photos have the same dynamic range? 12.5 stops with noise reduction isn't the greatest.

The Z9 test shows 12.5 usable from a possible 15.5 overall on an 8K timeline. That is not bad at all. NR can add only about half a stop. Compare that with the R5C in 8K RAW, which shows only 11 usable from a possible 13.4:

If you scale 8K down to 4K, you can improve the Imatest result for both 8K cameras, so you have to keep that in mind when comparing with lower resolutions. For example, the 4K C70 with its DGO sensor attains 13 usable from a total 15.3 with NR applied in-camera:

In terms of DR, Dxomark measured the Z9 RAW photo performance at 14.4 EV at its base 64 ISO and just over 12 EV at ISO 800, which is the native ISO for video:

https://www.dxomark.com/nikon-z9-sensor-test/#:~:text=As for the individual scores,at the same ISO sensitivity.

All that is very consistent with the Imatest DR above and very similar to other 8K cameras at the same ISO. What interests me more is the color depth comparison between photo and video, which is the only way to test whether the color information is retained in the 14 bit NEV files.

[TomTheDP]

3 hours ago, Kino said:

 

This chart mentions NR. Is it just talking about the NR that comes from downsampling 8k to 4k? or is it adding noise reduction in davinci or other program.?

If it is adding NR in post that means the actual dynamic range is about 12 stops. The Panasonic S1 has more without NR. 12 stops still isn't bad.

The 8k Sony venice gets 12.8 stops when downsampling 8k to 4k in post. The RED raptor is getting 12.5 stops. So 12 isn't bad.

The stills you posted do look fantastic though. Would be awesome if it was a 14 bit readout, though it seems like they would advertise that.

 

[Kino]

1 hour ago, TomTheDP said:

This chart mentions NR. Is it just talking about the NR that comes from downsampling 8k to 4k? or is it adding noise reduction in davinci or other program.?

If it is adding NR in post that means the actual dynamic range is about 12 stops. The Panasonic S1 has more without NR. 12 stops still isn't bad.

The 8k Sony venice gets 12.8 stops when downsampling 8k to 4k in post. The RED raptor is getting 12.5 stops. So 12 isn't bad.

The stills you posted do look fantastic though. Would be awesome if it was a 14 bit readout, though it seems like they would advertise that.

 

It seems that he added the NR on the Z9 N-RAW 8K in post using Resolve, but it does not include downsampling. With downsampling to 4K, you should be able to get another .5 stop. So, if you take out the NR but downsample to 4K, you should be back at around 12.5 usable stops.

As for why Nikon did not advertise the 14 bit NEV files, I'm not entirely sure, but I plan to find out . . .

[TomTheDP]

1 hour ago, Kino said:

It seems that he added the NR on the Z9 N-RAW 8K in post using Resolve, but it does not include downsampling. With downsampling to 4K, you should be able to get another .5 stop. So, if you take out the NR but downsample to 4K, you should be back at around 12.5 usable stops.

As for why Nikon did not advertise the 14 bit NEV files, I'm not entirely sure, but I plan to find out . . .

I would definitely be considering it if it is a 14 bit readout. That would be very unique. 12.5 stops with the texture of 8k raw would be very nice.

Still a heck of a lot pricier than my Sigma FP, but for a good reason.

[j_one]

Ever since the D750 I've had hopes that Nikon would pull through with a banger of a release in video specs. I think the Z9 RAW indicates they're going in the right direction, but I just haven't seen enough use or support online to draw a proper conclusion. When NRAW was released I was super hyped; maybe it's just my eyes or bias but I did "perceive" what looked like richer tonality in the few sources of footage online. That kind of color depth from 5D magic lantern, Sigma FP, or Redcode etc. I'd like to find some files of well-lit scenes and push in Resolve.

So I'm looking forward to hearing what both the rumored Z8 and S1Hii releases bring. I'd trade up from my s5ii to either of them for ultimate high-res hybrids with PDAF and color I favor, and just adapt EF glass. I just can't bring myself to settle on the R5C man.

[Kino]

I managed to download a few brief 8K 50fps NRAW NEV samples from another forum and view them in the latest version of Resolve (18.1.4):

https://forum.grassvalley.com/forum/editors/editing-with-edius/572094-n-raw-from-z9-sample-download

They are now listed as 12 bit according to Nikon specs. There must have been a decoding problem that has been solved since Resolve 18.0. Perhaps NEV files are captured in 12 bit log and Resolve previously converted them to 14 bit linear. It is hard to know what the issue was as TicoRAW compression is new to the camera world (hence the RED lawsuit).

The same TicoRAW conversion that the Z9 utilizes when compressing for photos in the "High Efficiency" RAW mode at 14 bit readout is also utilized to create the NRAW files. In photo mode, this is limited to 20fps at full sensor readout (8256 x 5504). Although the video mode uses a crop (8256 x 4644), which produces a 38mp image, it doesn't seem possible that you could get 14 bit RAW video at frame rates higher than 24/25. The Canon R3 could do 14 bit RAW at 30fps with its 6K video, but Canon would never do such a thing! That would obliterate its current cinema line.

In any case, the rich color tonality we were seeing in the Z9 footage is still there and the files edit nicely on a 4K timeline. They are perhaps not as optimized as R3D files, but they should be easier for editing than Canon 8K RAW Light.

[TomTheDP]

25 minutes ago, Kino said:

I managed to download a few brief 8K 50fps NRAW NEV samples from another forum and view them in the latest version of Resolve (18.1.4):

https://forum.grassvalley.com/forum/editors/editing-with-edius/572094-n-raw-from-z9-sample-download

They are now listed as 12 bit according to Nikon specs. There must have been a decoding problem that has been solved since Resolve 18.0. Perhaps NEV files are captured in 12 bit log and Resolve previously converted them to 14 bit linear. It is hard to know what the issue was as TicoRAW compression is new to the camera world (hence the RED lawsuit).

The same TicoRAW conversion that the Z9 utilizes when compressing for photos in the "High Efficiency" RAW mode at 14 bit readout is also utilized to create the NRAW files. In photo mode, this is limited to 20fps at full sensor readout (8256 x 5504). Although the video mode uses a crop (8256 x 4644), which produces a 38mp image, it doesn't seem possible that you could get 14 bit RAW video at frame rates higher than 24/25. The Canon R3 could do 14 bit RAW at 30fps with its 6K video, but Canon would never do such a thing! That would obliterate its current cinema line.

In any case, the rich color tonality we were seeing in the Z9 footage is still there and the files edit nicely on a 4K timeline. They are perhaps not as optimized as R3D files, but they should be easier for editing than Canon 8K RAW Light.

12 bit log has a similar amount of info as 14 bit linear raw. But yeah the image definitely looks very nice regardless.

[Kino]

Nikon should pay this guy! That video will sell cameras.

 

 

[D Verco]

Footage looks clean and with great colours, but not the best example of dynamic range considering the shadows/ blacks. But that could be how it was shot/ graded. 

[MrSMW]

Nice footage and grade, but it’s all sunset and as such, only showcases one very specific use case.

[kye]

21 minutes ago, D Verco said:

Footage looks clean and with great colours, but not the best example of dynamic range considering the shadows/ blacks. But that could be how it was shot/ graded. 

Footage is definitely impressive, but not really a DR test as much as a low-light test.  Most cameras can capture enough DR if you wait long enough after sunset and use a fast enough lens to keep noise low.

Still, it may well sell cameras, that's likely to be true.

Also, you have to be super-careful when it comes to evaluating colour unless the test involves only the manufacturers LUT because the skill of the colourist reigns supreme here.  Most people don't have the faintest idea what is possible in post if you are truly skilled - here's a snippet from a video on the L*a*b colour space (which is completely different to RGB) showing how you can "create" amazing sunset colours almost from nothing:

I hear professional colourists talking about things like this (that are basically mind-bending if you're new to alternative colour spaces) regularly, and these aren't even colour scientists.  If you pay money for professional LUTs or power grades (as opposed to buying the flaming garbage that YT camera bros are selling) then there are all sorts of alternative colour space tricks that get used in these LUTs.  Also, you can see from the above video that it took a single node and single curve to do that, whereas the pros often use dedicated software to build LUTs that go beyond what Resolve can do.  I know colourists that write their own software for building LUTs.

Get the right LUT and the right scene and magic can happen.  I'm just saying, buyer beware unless you know the full pipeline.

[Kino]

Although the grade is always important, NRAW's rich tonality is once again evident in this beach footage, which has been interpreted in linear color space as opposed to N-Log gamma (where DR could be better preserved). Despite the varying workflow, the expansive tonal range is consistent with Nikon's own Venice clip as posted above.

[Emanuel]

I don't understand the complaints, really. Looks like to me mere nerdy geeks talk, that's it, no more no less, unfortunately.

- EAG :- )

[kye]

6 hours ago, Emanuel said:

I don't understand the complaints, really. Looks like to me mere nerdy geeks talk, that's it, no more no less, unfortunately.

- EAG :- )

The images look great.

My point was that you need to be careful when using images like this to decide if you should buy the camera.  There are many cameras out there that can provide a lovely image, and some of those cameras are easy to use and will reliably create lovely images for most users with moderate skill levels, but there are other cameras that require extreme skill or luck to create lovely images and for most users with moderate skill levels they will be frustrating and disappointing.

Those sunset images looked a little too polished and I suspect they're the result of someone that either got lucky with a LUT that happened to do great, or they have a LOT of skill in colour grading.

[BTM_Pix]

5 hours ago, kye said:

There are other cameras that require extreme skill or luck to create lovely images

 

 

[kye]

1 hour ago, BTM_Pix said:

 

 

True, but he used pretty good clubs too.  No "I use the worst clubs I can get because it's all about skill not equipment" going on in the real world....