[Rys]

Since its launch last November, Sony's latest gaming platform has given early adopters trouble when attempting to play certain titles in 1080i/p. The console did not automatically upscale its video output to desired resolutions; it was up to either the game software to support these resolutions natively or for users to rely on the internal scalers of their HDTVs. Until now, this forced many people, developers and owners alike, to question the very existence of scaling hardware in the PlayStation 3.

The keywords here would be “until now,” because with the latest PlayStation 3 software development kit (SDK) update, Sony has finally exposed part of the built-in hardware scaler to developers.

Will this mean that most, if not all, future games will support output at 1080p/i resolutions? Moreover --and this is the question that owners of 1080i-only CRT HDTVs crave to see answered-- does this mean that current PS3 games may eventually support the native HD resolution of their televisions? Well, the answer requires some good old fashioned explanation, so let’s start already!

[DemoCoder]

Is it really hardware scaling, or are they simply playing tricks with the TMDS transmitter, RAMDAC, or maybe some I/O chip? E.g. mess with some clock timings so that each "pixel" is held for twice as long. I still doubt the "hardware scaler".

The resolution choice is curious, since it seems to suggest they're just going to double each pixel on output. While this will give you 1920 pixels on a scan line, it's more like a point-sampled magnification.

[Zaphod]

Do (m)any PS3 games use 2D art or overlays for menus, HUDs or whatever (do they support different aspect ratios at all for that matter)? If so, it is not simply a matter of rendering to a 960x1080 display buffer as the non-linear stretch will make such elements look atrotcious without remaking the art to suit the anamorphic upscaling. Thay'll basically have to redo the game to suit the faux aspect ratio.

[Arun]

Originally Posted by Zaphod

Do (m)any PS3 games use 2D art or overlays for menus, HUDs or whatever (do they support different aspect ratios at all for that matter)? If so, it is not simply a matter of rendering to a 960x1080 display buffer as the non-linear stretch will make such elements look atrotcious without remaking the art to suit the anamorphic upscaling. Thay'll basically have to redo the game to suit the faux aspect ratio.

The same is true for 720p->480p, however, so if the game can handle that, I'm sure it can handle this too.
DemoCoder: I wouldn't be surprised if it wasn't really a scaler per-se, although I *think* that's how it is being described to developers. We ran the story with what we knew - not with what we didn't. So hopefully we'll get to know more about this in the (near) future...


Uttar

[Zaphod]

Originally Posted by Uttar

The same is true for 720p->480p, however, so if the game can handle that, I'm sure it can handle this too.

Why? Having designed a game for both 4:3 and 16:9 doesn't necessarily imply that you can render to any arbitrary aspect ratio and still have it look all right after stretching in one direction. Doesn't games most games that display on 4:3 SD sets use either a narrower field of view or letterboxing anyway? Sounds to me like this will require (not insignificant) effort from the developers.

[Titanio]

Originally Posted by DemoCoder

Is it really hardware scaling, or are they simply playing tricks with the TMDS transmitter, RAMDAC, or maybe some I/O chip? E.g. mess with some clock timings so that each "pixel" is held for twice as long. I still doubt the "hardware scaler".

The resolution choice is curious, since it seems to suggest they're just going to double each pixel on output.

This doesn't seem to work so well at the other new resolution modes, though. They could (I guess?) replicate every second and every third pixel in those modes, but it would not be an even 'stretching' across the whole image as in the 960x1080 case. Maybe they don't care, but..

[[maven]]

For the memory calculations for the front buffer: Why use 32bit if 24bit (or even less, like the Gamecube demonstrates with its 4:2:2 (IIRC) YUV front-buffer in main-memory) is sufficient?

[mczak]

Originally Posted by [maven]

For the memory calculations for the front buffer: Why use 32bit if 24bit (or even less, like the Gamecube demonstrates with its 4:2:2 (IIRC) YUV front-buffer in main-memory) is sufficient?

Can RSX do scanout from yuv or 24bit buffers? Old pc gpus sure couldn't (though I don't know about newer ones). I also think it's a bit strange there is no "real" hw scaler - geforce gpus sure do? Even a good old radeon 7000 had a hw scaler (aptly named RMX).

[Jawed]

Originally Posted by [maven]

For the memory calculations for the front buffer: Why use 32bit if 24bit (or even less, like the Gamecube demonstrates with its 4:2:2 (IIRC) YUV front-buffer in main-memory) is sufficient?

I asked a similar question on the other thread, but Vysez seems to be dodging the question.

Jawed

[pux]

It seems PS3 has hw scaler in RSX, but it can't down-scale well.
Note that 720p to 1080i is down-scaling vertically.

[Farid]

Originally Posted by Jawed

I asked a similar question on the other thread, but Vysez seems to be dodging the question.

Jawed

Heh heh, I'm not dodging anything, your question was interesting, so I ran some more fact checks around. Just in case.
We chose to go with 32bits frontbuffers because it's the common depth used by most developers we asked about their frontbuffers. It also appear that there might be some aligment reasons explaining that choice of theirs.

[Arun]

Originally Posted by Jawed

I asked a similar question on the other thread, but Vysez seems to be dodging the question.

I could be mistaken, but it seems to me that if the hardware has no knowledge of the programming doing software scaling, the backbuffer and frontbuffer must surely be at the same resolution and depth. Furthermore, 24 isn't a power of two, and at least on PC APIs, true non-pow-of-2 rendertargets don't tend to exist... No idea about consoles, but I'd assume the same.


Uttar

[Carl B]

Originally Posted by pux

It seems PS3 has hw scaler in RSX, but it can't down-scale well.
Note that 720p to 1080i is down-scaling vertically.

The scaling is horizontal-only at this time. I think what you're being thrown off by is the interlaced resolution being 'half' of what the framebuffer itself actually contains; as far as memory costs go, consider 1080i to be the same as 1080p. Field rendering would give you that situation I believe you're refering to, but PS3 can't go that route.

[pux]

Originally Posted by xbd

The scaling is horizontal-only at this time. I think what you're being thrown off by is the interlaced resolution being 'half' of what the framebuffer itself actually contains; as far as memory costs go, consider 1080i to be the same as 1080p. Field rendering would give you that situation I believe you're refering to, but PS3 can't go that route.

You are right.
The scaling is horizontal-only at this time, because the down-scale is necessary for scaling the vertical direction (for 1080i TV), and PS3 seems have a problem there.
And, the hw-scaler is sure to work when scanned out from the frame buffer.

[nelg]

Please excuse if this a stupid question........ does this have anything to do with HDCP, and perhaps a way to circumvent any licensing restrictions?

[Jawed]

Well, I admit I really can't tell which items of hardware are involved in the chain from backbuffer to display:

http://www.beyond3d.com/forum/showpo...&postcount=225

http://www.beyond3d.com/forum/showpo...&postcount=255

as it seems that PS3 is not using standard NVidia G7x hardware.

Jawed

[NeoTechni]

Could they not make one buffer for the 1080p image, leave the 720p image in the buffer, but so it's right against the end of it. And start scaling, by the time the output buffer starts overlapping the 720p image it would already be done with that area. It a little at the end gets overlapped before it's used they could extend the 1080p buffer a bit more.

Thus using one 1080p (or slightly larger buffer) for both images

[Pete]

Originally Posted by nelg

Please excuse if this a stupid question........ does this have anything to do with HDCP, and perhaps a way to circumvent any licensing restrictions?

Interesting, but don't HD-DVD players allow for variable output resolutions via HDMI-HDCP?

BTW, Vy kept me wondering right up to the last half-sentence when he'd add his trademark ingredient.

[DemoCoder]

Originally Posted by Titanio

This doesn't seem to work so well at the other new resolution modes, though. They could (I guess?) replicate every second and every third pixel in those modes, but it would not be an even 'stretching' across the whole image as in the 960x1080 case. Maybe they don't care, but..

I don't think they're claiming it can deal with the other resolution modes? This seems to be a 960x1080 -> 1080i/p transformation. I bet the SDK will actually use 1280x720p if the OS is configured for 720p output or the developer is asked to first detect the output and choose either 720p or 960x1080. Since real 720p is less of a workload than 960x1080, it's likely that the transformation will not produce any performance hiccups. For 480p, you can either render at that res, or simply downsample the backbuffer, since the memory requirements to hold a 480p buffer are not significant.

I think if Sony really had a true hardware scaler, they would have enabled it at the OS level and it would not need SDK support, nor would it be only a "half scaler"

My guess is, they came up with a trick for "scan doubling" the horizontal that would not require too much extra performance or developer headache (e.g. 960x1080 res) using some aspects of the RAMDAC or TMDS system. I'll admit, I'm not really sure how they're doing it, but I remember plenty of tricks you could do by playing with VGA clocks, so I would not be surprised if they are using some flexibility of the scan-out devices to implement this.

The reason this smells like a hack and not a real hardware scaler is:

1) it requires explicit SDK support rather than being something you can just toggle on and off (a HW scaler would sit between the framebuffer and HDMI out and would not need explicit SDK control)

2) the idea of a HW scaler that can only deal with one dimensional axis, and requires a nice, even power of 2 resolution to scale is very strange

3) no one who has done teardowns of the PS3 has been able to identify any such HW

4) if Sony really had a HW scaler, they would be releasing PR press releases, it would be in the marketing materials/spec sheets, and they would not allow MS to get away with all of the attacks on PS3 resolution issues

5) It's really stupid to launch a console worldwide in a heated race with MS, already 1 year behind, and not fully reveal all of your HW capabilities to buyers. If you were already 1 year ahead, you might choose to keep secrets since you're in the lead and don't need them. But if you're playing catch up?

Like the hopes and dreams of people who thought there would be an HDMI cable for the Xbox360, since there might be "secret" HDMI support or A/V pins on the 360 (again, why not launch with an HDMI cable accessory then if you went through the trouble of including TMDS hardware that is costing you on every box shipped!!) This "PS3 has a hardware scaler" issue will turn out to be a big myth IMHO, unless you want to play semantic games with the definition of "hw scaler"

[Carl B]

Originally Posted by DemoCoder

3) no one who has done teardowns of the PS3 has been able to identify any such HW

Demo I agree that whatever it is, it is not taking the normal form of what one would expect from a scaler, but that said at the same time I don't think teardown analysis is the thing that's going to prove anything one way or the other. Half the chips in there, the teardown firms have no idea what they are, and even now a southbridge chip that may or may not be Toshiba's Super Companion Chip is an utter mystery in terms of both its functionality and its size.

[Geo]

The whole thing does have a sort of Hardy Boys-ish "The Case of the Curious Hardware Scaler" feel to it, doesn't it? I suppose this puts Vysez in the Joe Hardy/Shaun Cassidy role?

The next time we're wrong won't be the first time, but we're feeling pretty good there's a real hw scaler in there. I suspect at some point we'll be able to revisit this one for a post-mortem, and I'm sure we all look forward to that.

[nAo]

Originally Posted by [maven]

For the memory calculations for the front buffer: Why use 32bit if 24bit (or even less, like the Gamecube demonstrates with its 4:2:2 (IIRC) YUV front-buffer in main-memory) is sufficient?

The GC demonstrates quite awful banding, all the time... but that's cause even the back buffer can be 24 bits..
And also how many modern GPUs that render to a 24 bit render target do you know?

[nelg]

Originally Posted by Pete

Interesting, but don't HD-DVD players allow for variable output resolutions via HDMI-HDCP?

Seeing that the HDCP 1.3 revision was finalized on Dec. 21 2006 and the timing of this update someone might want to look here to see if they can find a possible answer.

[Titanio]

Originally Posted by DemoCoder

I don't think they're claiming it can deal with the other resolution modes?

The article says:

Additional 1080 rendering modes also supported now include 1280x1080, 1440x1080 and 1600x1080. These modes, similar to 960x1080, are capable of being upscaled by the hardware into 1080p/i.

As for the rest, there's certainly a lot of mystery about it all, but re. Sony PR or whatever..I don't think there's any point second-guessing their thoughts on this, since their approach seems to be to acknowledge no issues and just quietly fix them. Unless you're Watch Impress and have access to Ken Kutaragi..he seems more than happy to discuss anything and everything with them in finer technical detail, but that asides..

My own theory, or one of them, is that Sony is cautious about what it enables or how it enables it because they have an eye on cost reduction. There may be hardware in there that allows for more, but they may have an eye on a point in time where it won't be there, or can be substituted for something cheaper (and perhaps less capable). So they're being careful about what they commit too. The other commonly presented explanation is software issues. I haven't a clue, it is very mysterious.

[[maven]]

Originally Posted by nAo

The GC demonstrates quite awful banding, all the time... but that's cause even the back buffer can be 24 bits..
And also how many modern GPUs that render to a 24 bit render target do you know?

As you know, the banding is due to the 6:6:6:6 mode, not the YUV conversion. And if you look closely, I was talking of the the front-buffer (in main-memory), not the back-buffer or any render targets in the embedded RAM...