New for NAB

[Matt Gottshalk]

The AF-100 does NOT use the GH2 chip, and Jan Crittenden has stated that numerous times.

[Les Wilson]

Originally Posted by Matt Gottshalk

The AF-100 does NOT use the GH2 chip, and Jan Crittenden has stated that numerous times.

Originally Posted by Jan Crittenden Livingston

Hi David, Obviously you are not aware of the fact that it is the same silicon in the AF100 as it is in the GH2, but with better DSP to offset the skew and other MOS artifacts.

Best,

jan

Silicon, DSP, Chip .... no wonder people are confused

[David Heath]

Originally Posted by Matt Gottshalk

The AF-100 does NOT use the GH2 chip, and Jan Crittenden has stated that numerous times.

But not on this thread..........! Quite the opposite!

A few posts back she corrects me, and says:

Originally Posted by Jan Crittenden Livingston

Hi David, Obviously you are not aware of the fact that it is the same silicon in the AF100 as it is in the GH2, but with better DSP to offset the skew and other MOS artifacts.

Best,

jan

I now agree with what she says - same physical chip, but think the difference between the AF100 and the GH2 is down to amount of the chip each reads out - the AF100 is windowed or cropped relative to the GH2. As said before: "They do indeed physically share the same sensor, but it seems the AF100 is not making full use of it. The USED area in the GH2 is 17% bigger than in the case of the AF100, 17% more photosites! Who has both cameras and is going to test the angle of view theory?"

There seems to be pretty good evidence from Phillip Blooms Xmas shootout - http://www.dvinfo.net/forum/new-cano...ml#post1706302 . He shoots on the same lens, same view, from same position with both the AF100 (10min 18sec) and the GH2 (13min 40sec) - and it's obvious the GH2 does indeed have a noticeably wider angle of view. (I'm not prepared to draw any other conclusions from that test - the lighting has changed for one thing.)

The explanation provided to me said "It's easy enough to prove all the above if you have a GH2 and AF100. Put then side by side and shoot video with them locked off with the same lens. You will find the angle of view of the GH2 to be a bit wider than that of the AF100, the above theory predicting about 8% wider."

I tried to quantify the difference between the images in Philip Blooms examples and I made it to be approximately 6-7% - a pretty good correlation with the 8% that was predicted, given the margins for error. I'd be interested if anybody is able to redo the comparison under more controlled conditions.

Originally Posted by Sanjin Svajger

David, very interesting. I to didn't know that the AF100 uses the GH2 chip. This makes me wonder what they can do to improve it? A better DSP I guess...? It also makes me think that an upgraded AF100 is not coming.

The obvious thought in my mind is that although they both use the same chip, an upgraded AF100 could still benefit from reading it in the same way as the GH2, read the 4976x2800 window instead of 4608x2592. If the GH2 can do that, why can't the (far more expensive) AF100?

It still won't compete with all the "designed for video" sensors by a long way, but would be an improvement. At least it would mean the quality would be on a par with the GH2.

[Konstantin Kovalev]

Here's an idea I just had: a 4K 2/3" camcorder!

This sounds outrageous at first, but this should be quite doable, considering that with modern advances in sensor design (gapless micro-lenses) it would have about the same pixel pitch as a 1/3" camcorder... when Canon released the the 1D4 a while back, it was claimed as having a higher resolution over it's predecessor (16mp vs. 10mp), but with the exact same pixel size on a same-sized sensor. There aren't many 2/3" lenses that could provide the sharpness needed to make use of 4K, but the extra resolution can allow for over-sampling to 2K/HD until optics catch up.

Edit: while I'm on a roll, would it kill to also ask for some sort of log/linear gamma? I know there's more DR available in these cameras than the software allows for. Low-contrast settings (Cinelike D, low knee, master ped 0 etc.) look like what you'd get from a stills camera with the contrast turned up to 11.

[Sanjin Svajger]

Originally Posted by David Heath

The obvious thought in my mind is that although they both use the same chip, an upgraded AF100 could still benefit from reading it in the same way as the GH2, read the 4976x2800 window instead of 4608x2592. If the GH2 can do that, why can't the (far more expensive) AF100?

It still won't compete with all the "designed for video" sensors by a long way, but would be an improvement. At least it would mean the quality would be on a par with the GH2.

Sure, a little more resolution would not hurt but what would convince me of buying the AF100 is a better DSP and P2 - not more resolution. I don't like the way it handles highlights and I especially don't like the chroma clipping. It looks a loot like my HPX171 in this regard. Same chroma clipping, especially in the red channel.
This two things are a lot more important to me than 50 more lines of resolution. I'm sure that I'm not the only one in this matter.

Anyway, there must be a reason why AF100 uses a lesser amount of the chip area than the GH2...

[Glen Vandermolen]

Originally Posted by Konstantin Kovalev

Here's an idea I just had: a 4K 2/3" camcorder!

This sounds outrageous at first, but this should be quite doable, considering that with modern advances in sensor design (gapless micro-lenses) it would have about the same pixel pitch as a 1/3" camcorder... when Canon released the the 1D4 a while back, it was claimed as having a higher resolution over it's predecessor (16mp vs. 10mp), but with the exact same pixel size on a same-sized sensor. There aren't many 2/3" lenses that could provide the sharpness needed to make use of 4K, but the extra resolution can allow for over-sampling to 2K/HD until optics catch up.

Edit: while I'm on a roll, would it kill to also ask for some sort of log/linear gamma? I know there's more DR available in these cameras than the software allows for. Low-contrast settings (Cinelike D, low knee, master ped 0 etc.) look like what you'd get from a stills camera with the contrast turned up to 11.

If JVC can now squeeze 4K out of a single 1/2" chip on the HMQ10, why not a 2/3" chip? Wasn't this the design of the original Scarlet?

[Sanjin Svajger]

Originally Posted by Glen Vandermolen

If JVC can now squeeze 4K out of a single 1/2" chip on the HMQ10, why not a 2/3" chip? Wasn't this the design of the original Scarlet?

I would rather have a 2K 2/3 chip with a decent DR and good noise performance.

[David Heath]

Originally Posted by Sanjin Svajger

..........what would convince me of buying the AF100 is a better DSP and P2 - not more resolution.

When you say "P2", I suspect what you really mean is "a better codec", isn't it? What is also interesting from Phil Blooms tests with the GH2 is that sensor aside, it's been hacked to record at a higher data rate and his video indicates 170Mbs. (He does say that may not be reliable to his Class 10 card, but seems to indicate the 70Mbs mode is.) So with SDXC, recording 100Mbs AVC-Intra shouldn't be any problem - wouldn't the option of that be a better way forward than P2?

I don't like the way it handles highlights and I especially don't like the chroma clipping.

I've said on many occasions (mainly to do with DSLRs) that there is a far bigger gap than is generally recognised between sensors PRIMARILY designed for stills and those PRIMARILY for video. And highlight handling is one way it shows up. Think about it. There are actually 18 megapixels on the GH2 chip (though the AF100 only windows 12 megapixel), and it's roughly half the size of an s35 sensor. It's an equivalent photosite size to over 30 megapixels on an s35 chip!

Hence each photosite must be only about one tenth the area (very approximately) to that on the F3/FS100 - is it really a surprise that highlight handling is not as good as the "designed for video" sensors?

Anyway, there must be a reason why AF100 uses a lesser amount of the chip area than the GH2...

I've just heard one plausible sounding explanation, that the AF100 started to be designed around a GH1 chip, and the GH2 chip was a late amendment to the spec. I can only assume that to meet release dates etc it wasn't possible to do a total redesign, and some of the original processing couldn't cope with the full 16:9 window of the GH2, so this windowing was the compromise. Doing the same calculations for the GH1 chip suggests that even if the max area was used, the max resolution before aliasing would be only 612 lines, and if it used the windowed area it would only be 562 lines!

So the 648 of the AF100 is still a significant improvement over the 612 it would have been initially.

(And if anybody does still have any doubts, these calculations show that it definately does have the GH2 chip, it wouldn't be able to get to 648 lines with the GH1 sensor, with the processing used.)

[Sanjin Svajger]

Originally Posted by David Heath

When you say "P2", I suspect what you really mean is "a better codec", isn't it?

Yes, a better codec. Bot no, for me personally I did actually meant P2 - as I am already invested in P2 and wouldn't like to sell it.
Off course, ditching P2 and going the CF road or even SDXC as you suggested would be an general improvement. P2 is to expensive to be in this lower end market segment! It's that simple. Remove it. But as we've been through this legacy issues I won't go through them again.

Originally Posted by David Heath

I've said on many occasions (mainly to do with DSLRs) that there is a far bigger gap than is generally recognised between sensors PRIMARILY designed for stills and those PRIMARILY for video. And highlight handling is one way it shows up. Think about it. There are actually 18 megapixels on the GH2 chip (though the AF100 only windows 12 megapixel), and it's roughly half the size of an s35 sensor. It's an equivalent photosite size to over 30 megapixels on an s35 chip!

Hence each photosite must be only about one tenth the area (very approximately) to that on the F3/FS100 - is it really a surprise that highlight handling is not as good as the "designed for video" sensors?

Wow those are actually some quite small photosites! I've never got down to calculating their size on the GH chip. At least Panasonic doesn't delete the ones it doesn't need like Canon does (I'm surprised that they are still doing it...). But still, it isn't all to do with photosite size is it? The DSP is also very important in highlight handling and chroma clipping, yes?

Originally Posted by David Heath

I've just heard one plausible sounding explanation, that the AF100 started to be designed around a GH1 chip, and the GH2 chip was a late amendment to the spec. I can only assume that to meet release dates etc it wasn't possible to do a total redesign, and some of the original processing couldn't cope with the full 16:9 window of the GH2, so this windowing was the compromise. Doing the same calculations for the GH1 chip suggests that even if the max area was used, the max resolution before aliasing would be only 612 lines, and if it used the windowed area it would only be 562 lines!

So the 648 of the AF100 is still a significant improvement over the 612 it would have been initially.

(And if anybody does still have any doubts, these calculations show that it definately does have the GH2 chip, it wouldn't be able to get to 648 lines with the GH1 sensor, with the processing used.)

This does sound like an very plausible explanation yes!

It's funny that all this time we thought that the AF100 uses the GH2 chip!:) There's actually a lot of people walking around wishing that Panasonic would introduce a revised AF100 camera at NAB this year that would feature the GH2 chip!:)

[Konstantin Kovalev]

Originally Posted by Glen Vandermolen

If JVC can now squeeze 4K out of a single 1/2" chip on the HMQ10, why not a 2/3" chip? Wasn't this the design of the original Scarlet?

Oh yeah, the Scarlet used to be 2/3" while in the works, I forgot about that... still sounds like a good idea though, even with the actual Scarlet out now.

Originally Posted by Sanjin Svajger

I would rather have a 2K 2/3 chip with a decent DR and good noise performance.

I doubt noise performance would take a big hit from cramming 4K in, if it's done right, the extra pixels would most likely just quantize the noise floor at worst. 2K w/ bigger pixels would be killer too though.

Originally Posted by Sanjin Svajger

Off course, ditching P2 and going the CF road or even SDXC as you suggested would be an general improvement. P2 is to expensive to be in this lower end market segment! It's that simple. Remove it. But as we've been through this legacy issues I won't go through them again.

There are plenty of CF/SD camcorders in this price range as it is... at least with P2 you're getting a quadruple redundant fail safe system that ensures you never lose more than 4 seconds or so of footage, even if you pull the card out while recording or the camera explodes or something. Also, solid state media is infinitely cheaper than any disk/tape based solution because it can be erased/used practically indefinitely, 95% of everyone can get away with having to own 2~3 cards for the whole life of the device... digital tech has really spoiled this generation.

Wow those are actually some quite small photosites! I've never got down to calculating their size on the GH chip. At least Panasonic doesn't delete the ones it doesn't need like Canon does (I'm surprised that they are still doing it...). But still, it isn't all to do with photosite size is it? The DSP is also very important in highlight handling and chroma clipping, yes?

By popular myth, sensor size seems to have direct correlation to DR, but this isn't always so. The ALEXA and RED push the limits of 14-bit ADCs with 13.5 stops of DR, but sport an S35 sensor that about 1.6x crop equivalent to a full-frame dSLR, conversely, many dSLRs have about 11.5 stops of DR at ISO100, such as the 5D2. Few cameras, like the Nikon D3s, manage to get 13 stops but rarely more so. A typical camcorder has 10 stops of DR, despite having an absolutely tiny sensor in comparison.

[Glen Vandermolen]

Originally Posted by Sanjin Svajger

I would rather have a 2K 2/3 chip with a decent DR and good noise performance.

I sure wouldn't complain about such a camera!
I honestly liked the original design of the Scarlet better than what it is now. Mainly, the $6,000 target price. I'd have been very happy with a 2/3" chip, fixed lens and 4K resolution. I'm waiting for the eventual reviews of the forthcoming HMQ10 to see how close JVC comes to the original Scarlet concept. I work with EX3s all the time, and I don't feel crippled with the 1/2" chip size.
But I'll admit, my FS100 has spoiled me with its Super-35 sensor.

[David Heath]

Originally Posted by Konstantin Kovalev

... at least with P2 you're getting a quadruple redundant fail safe system that ensures you never lose more than 4 seconds or so of footage, even if you pull the card out while recording or the camera explodes or something.

That may be so - but it also applies to other solid state systems. And with the C300 you have the ability to record to dual CF cards at once. The reliability advantages of that (provided the user is sensible enough not to keep them together!) are not just down to card reliabilities, but that even if one got lost/stolen) there's always the other card.

And the economics are such that even two C300 parallel recordings still work out much cheaper/minute than a single one on P2!

Also, solid state media is infinitely cheaper than any disk/tape based solution because it can be erased/used practically indefinitely, 95% of everyone can get away with having to own 2~3 cards for the whole life of the device... digital tech has really spoiled this generation.

Yes - but that equally applies to other solid state solutions. (Though it ignores the cost of the media you download onto.) The real point is that one-off cost adds about $2,000 onto the capital cost of a P2 based system (for approx 3 hours recording) but it's far less with other systems. Hence to be competitive the basic body only price with P2 then has to be about $1,500 less than for other systems - I'd rather put my money into a better camera, not more expensive memory.

By popular myth, sensor size seems to have direct correlation to DR, but this isn't always so.

What I and Sanjin referred to was NOT sensor size, but PHOTOSITE size. It may not be a direct correlation, but generally bigger PHOTOSITES give better dynamic range. And you get bigger photosites by either/or having a bigger sensor and/or FEWER overall sites on the chip.

This is one of the reasons why using a sensor primarily designed for stills is highly unlikely to give as good results as one primarily designed for video. And going for four/thirds (approx half the area of s35) exacerbates the problem.

The ALEXA and RED push the limits of 14-bit ADCs with 13.5 stops of DR, but sport an S35 sensor that about 1.6x crop equivalent to a full-frame dSLR, conversely, many dSLRs have about 11.5 stops of DR at ISO100, such as the 5D2.

Yes - but Alexa has an effective photosite count of 2880x1620 (in 16:9) or about 4.6 megapixel. (Enough to give very good 1080p - but no more.) The Alexa also has about twice the area of the AF100 sensor, and the two factors together mean that EACH photosite should be about (2*18/4.6=7.8) 7-8 times as large on the Alexa as the AF100.

As far as "typical camcorders", then with the EX as an example, it's sensors are each about 1/8 the size of a four/thirds sensor - but it has about 2.2 megapixels each compared to the 18 of the AF100. As 2.2x8=17.6, I'd therefore expect the EX photosites to be (if anything) a bit larger than those of the AF100.

Photosite size may not be the only factor regarding DR - but it's a big one.

[Konstantin Kovalev]

Originally Posted by David Heath

And the economics are such that even two C300 parallel recordings still work out much cheaper/minute than a single one on P2!

Yes - but that equally applies to other solid state solutions. (Though it ignores the cost of the media you download onto.) The real point is that one-off cost adds about $2,000 onto the capital cost of a P2 based system (for approx 3 hours recording) but it's far less with other systems. Hence to be competitive the basic body only price with P2 then has to be about $1,500 less than for other systems - I'd rather put my money into a better camera, not more expensive memory.

There is no such thing as "cheaper/minute" in any digital recording solution unless you're thinking of recording uncompressed 4K or something, under realistic use, storage is cheap enough for it to be a non-issue.
3x P2 costs $1905 at B&H (SxS is a bit more) and 3x CF cards (Sandisk EP 64GB) is $1,110... ok an $800 difference? I know there are cheaper CF cards, but I probably wouldn't want to use those for anything worthwhile.
Considering the useful life of the product, and the likelihood it'll buy itself back, $800 is moot by any standards. Any camera sporting P2 isn't likely to be used by penny-pinching hobbyists.

What I and Sanjin referred to was NOT sensor size, but PHOTOSITE size. It may not be a direct correlation, but generally bigger PHOTOSITES give better dynamic range. And you get bigger photosites by either/or having a bigger sensor and/or FEWER overall sites on the chip.

This is one of the reasons why using a sensor primarily designed for stills is highly unlikely to give as good results as one primarily designed for video. And going for four/thirds (approx half the area of s35) exacerbates the problem.

Yes - but Alexa has an effective photosite count of 2880x1620 (in 16:9) or about 4.6 megapixel. (Enough to give very good 1080p - but no more.) The Alexa also has about twice the area of the AF100 sensor, and the two factors together mean that EACH photosite should be about (2*18/4.6=7.8) 7-8 times as large on the Alexa as the AF100.

As far as "typical camcorders", then with the EX as an example, it's sensors are each about 1/8 the size of a four/thirds sensor - but it has about 2.2 megapixels each compared to the 18 of the AF100. As 2.2x8=17.6, I'd therefore expect the EX photosites to be (if anything) a bit larger than those of the AF100.

Photosite size may not be the only factor regarding DR - but it's a big one.

...Alright, so what about the 14 megapixel RED Epic? It has a small-ish sensor by still-camera standards, but has a higher dynamic range than even medium format digital cameras, which have absolutely massive pixels. To put it in perspective, a medium format camera needs to have 80mp to match the 5D2 in pixel size, which in turn has the same pixel density as an 8mp crop camera; most MFD cams are in the 40~50mp range.

If Red isn't using some secret sauce in their cameras, then I don't know what it is. The RED camera has been tested and proven to have the exact same DR as the Alexa, if they didn't label the step chart I couldn't tell which was which.

[David Heath]

Originally Posted by Konstantin Kovalev

3x P2 costs $1905 at B&H (SxS is a bit more) and 3x CF cards (Sandisk EP 64GB) is $1,110... ok an $800 difference?

But you're forgetting that the codec as used in the C300 puts 2 hours on to a 64GB card, with P2 it's only one hour per 64GB - that's why cost/minute is more significant than cost/GB. Hence the comparison figure really should be more like $550 for CF - so a $1,350 difference - not $800.

Considering the useful life of the product, and the likelihood it'll buy itself back, $800 is moot by any standards. Any camera sporting P2 isn't likely to be used by penny-pinching hobbyists.

You need to look at a total system cost - camera plus memory. Think of the memory as part of the capital cost of the system - I fully agree it's not a consumable cost.

The point I'm making is that I'd rather spend that $1,350 on a better camera - not more expensive memory. If I've got $6,000 to spend, I'd rather it be a $5,500 camera and $500 of memory than a $4,000 camera and $2,000 of memory. This is especially true if you compare something like the HPX250 with the XF305. In many ways comparable, but the 250 is considerably cheaper - until you factor in memory cost when they become similar in price. But the XF305 is better in respects such as the true manual lens (the 250 is servo for iris and focus)

...Alright, so what about the 14 megapixel RED Epic? It has a small-ish sensor by still-camera standards, but has a higher dynamic range than even medium format digital cameras, which have absolutely massive pixels.

I can only repeat what I said before: "Photosite size may not be the only factor regarding DR - but it's a big one."

As far as this discussion goes, then most of the competition to the AF100 are s35 - not four/thirds - and that gives them an advantage to start with. (Twice as big photosites for the same count per chip.) It doesn't surprise me that Panasonic used the GH2 chip in their first camera for reasons of speed and economies of scale, but if they are going to seriously compete from now on, they are going to have to come out with a designed for video chip. And the really big question will be whether it's s35 or four/thirds.

Go for s35 and it's all change for lenses, everything. Stay with four/thirds and it will always struggle against all the s35 rivals in so many ways, always be a runner up. It's a tough decision.

And the news of the FS700 has ratcheted up the stakes considerably.

[Sanjin Svajger]

Originally Posted by David Heath

And the news of the FS700 has ratcheted up the stakes considerably.

Yes, Sony is in a really good place at the moment - hands down. They have all the corners except the sub 5K market where the AF100 and DSLRs come in.
A slightly updated AF100 won't do much now. They must come out with a brand new S35 camcorder!

Have you seen the new Sony PDW 680 ENG type camcorder? What are they doing at Panasonic?