[B3D Article] "Ripping off the veil: The mysterious PS3 hardware scaler exposed"

[Panajev2001a]

No, scaling to an entire buffer is wasteful and unnecessary.
You can do horizontal scaling entirely as an inline process. You just interpolate between the pixels as they come in and insert new ones into the output stream. For a (reasonably good quality) cubic filter you need just four pixels as inputs. Chump change.
Vertical inline scaling would need much more storage, because it requires you to keep multiple lines of pixels available as inputs.

Now that's still something I'd most definitely call a hardware scaler. It's not just an accidental capability of a bog-standard RAMDAC, you need to include those transistors into the design somewhere. NVIDIA's PC chips could do it for a while though.

I think you hit the nail on the head, it sounds like a perfectly reasonable explanation as to why they enabled horizontal scaling only (4 pixel inputs for a reasonably nice cubic filter does not seem too bad in terms of storage needed). The question is why it took them this much time to feel safe about developers using this functionality in games... was the driver that allowed access to the HW scaler slow or were there some software related bugs ?

 

[jetrii]

To quote Precis from AVS:

"My guess is that the intended implementation was in violation of the HDCP 1.3 revision spec (published Dec. 21 06). The current 'work around' is a means to circumvent HDCP 1.3 non compliance."

Hmmm, that seems plausible.

 

[Carl B]

Do you think Sony will respect the difference between Breaking NDA and Seriously Breaking NDA?

Well, in truth he's not breaking the NDA at all by confirming a hardware scaler, because as of this article, it's now public knowledge. The only doubt really rests with those that wonder whether it's 'actually' a hardware scaler; but the article was explicit in this regard.

 

[Shifty Geezer]

Well, in truth he's not breaking the NDA at all by confirming a hardware scaler, because as of this article, it's now public knowledge. The only doubt really rests with those that wonder whether it's 'actually' a hardware scaler; but the article was explicit in this regard.

Would that count though, as the article was describing something of a 'best guess'? Putting it another way, consider the clocke speed of RSX which is all top secret... Before a dev can say outright it's 650 MHz (honest guv!) that knowledge needs to be public, right? So if a dozen blog sites and article say RAS is clocked at 400 MHz, 450 , 500, 550, 600, 650, 700, 750...can the dev say 'ah, such and such an article alredy made it public knowledge that RSX was 650 MHz so I can talk about that all I like'? I don't think Sony would be too happy with that! Unless the info comes from an official release, it might still be under NDA. Or if it's been officially confirmed after a leak, perhaps, but I imagine legal departments would be pretty unforgiving.

In this example, I'd say Darknight was in danger of a breach of NDA by both confirming the hardware scaler, and saying that devs knew it was there from Day 1 but weren't allowed to use it. Certainly that latter information has repercussions for Sony. Was it broken? Are they holding back? Is their software incomplete? Whatever the reason, it looks bad on Sony. Whereas if the presence of a hardware scaler were not confirmed, and certainly that it wasn't pointed out to devs in the beginning, Sony wouldn't be the subject of these negative opinions.

 

[Carl B]

Would that count though, as the article was describing something of a 'best guess'? Putting it another way, consider the clocke speed of RSX which is all top secret... Before a dev can say outright it's 650 MHz (honest guv!) that knowledge needs to be public, right? So if a dozen blog sites and article say RAS is clocked at 400 MHz, 450 , 500, 550, 600, 650, 700, 750...can the dev say 'ah, such and such an article alredy made it public knowledge that RSX was 650 MHz so I can talk about that all I like'? I don't think Sony would be too happy with that! Unless the info comes from an official release, it might still be under NDA. Or if it's been officially confirmed after a leak, perhaps, but I imagine legal departments would be pretty unforgiving.

Well, but this scaler information was given to B3D with the express purpose of making it public, something I don't think people totally realize. I think that's a big difference between the scaler and the RSX scenario.

And the RSX is clocked at 500MHz, just to repeat that message!

In this example, I'd say Darknight was in danger of a breach of NDA by both confirming the hardware scaler, and saying that devs knew it was there from Day 1 but weren't allowed to use it. Certainly that latter information has repercussions for Sony. Was it broken? Are they holding back? Is their software incomplete? Whatever the reason, it looks bad on Sony.

Well, I guess it comes down to this - is the newly revealed scaler good news for Sony or bad news? And if it's good, then how can Sony possibly look worse off for it, secrecy or no?

Anyway I think Darknight is 100% in the clear, but the truth will of course be born out by anything happening (or not happening) to him in that regard.

 

[Shifty Geezer]

Anyway I think Darknight is 100% in the clear, but the truth will of course be born out by anything happening (or not happening) to him in that regard.

As long as identities are kept secret, talk of NDA limits are all hypothetical anyways

 

[nelg]

To quote Precis from AVS:

"My guess is that the intended implementation was in violation of the HDCP 1.3 revision spec (published Dec. 21 06). The current 'work around' is a means to circumvent HDCP 1.3 non compliance."

Hmmm, that seems plausible.

That fuc*ing plagiarizing SOB.
He is copying exactly what I said here.

That is OK though because Precis@AVS = nelg@beyond3d

 

[archie4oz]

Or it could be a historical precedent set by their last console

Actually it wouldn't be a historical precedent with the PS2... Console history (and computer history for that matter) is littered similar cases of functionality being progressively exposed, or a platform vendor being anal about exposing features (or demanding that certain previously exposed features not be used). Hell Commodore ended up having to re-implement a bunch of C64 design bugs in the C128 (even though during the design of the C128 the bugs were fixed) to maintain backwards compatibility simply because devs ended up writing software to work around or take advantage of those bugs...

 

[Fafalada]

Console history (and computer history for that matter) is littered similar cases of functionality being progressively exposed, or a platform vendor being anal about exposing features

Oh I'm not arguing with that - but that makes it even more of a historical precedent doesn't it?
Mainly I was talking about dodgy attitude about exact same features that were set as "standard" by their direct competition a year or so in advance though...

was the driver that allowed access to the HW scaler slow or were there some software related bugs

My guess would be the unit probably only scales purple pixels vertically or something to that effect.

 

[archie4oz]

Oh I'm not arguing with that - but that makes it even more of a historical precedent doesn't it?
Mainly I was talking about dodgy attitude about exact same features that were set as "standard" by their direct competition a year or so in advance though...

So you're saying the HDD should be made optional now?

My guess would be the unit probably only scales purple pixels vertically or something to that effect.

Funny that you mention purple and scaling in the same sentence...

 

[Rolf N]

Purple is everyone's favourite key color, isn't it? I admit to having used it too

 

[inefficient]

Purple is everyone's favourite key color, isn't it? I admit to having used it too

Do you mean Magenta? FF00FF

 

[Crossbar]

My guess would be the unit probably only scales purple pixels vertically or something to that effect.

So you are assuming the vertical scaler is broken somehow?

From a layman perspective it sounds like a pretty fundamental and simple function that should be fairly easy to test during development. Don't you think it would be strange if it was so hampered with bugs that it was rendered non-usable?

 

[Panajev2001a]

Oh I'm not arguing with that - but that makes it even more of a historical precedent doesn't it?
Mainly I was talking about dodgy attitude about exact same features that were set as "standard" by their direct competition a year or so in advance though...



My guess would be the unit probably only scales purple pixels vertically or something to that effect.

All right, I was thinking about doing a game based on Prince, whoohoooooo free scaling for me!

 

[nelg]

All right, I was thinking about doing a game based on Prince, whoohoooooo free scaling for me!

Thankfully the "Purple One" is in need of vertical scaling.

 

[bobthebub]

I think you hit the nail on the head, it sounds like a perfectly reasonable explanation as to why they enabled horizontal scaling only (4 pixel inputs for a reasonably nice cubic filter does not seem too bad in terms of storage needed). The question is why it took them this much time to feel safe about developers using this functionality in games... was the driver that allowed access to the HW scaler slow or were there some software related bugs ?

Wild speculation time: As you and others have mentioned horizontal scaling is relatively easy, how about Sony have a scaler in there (part of the southbridge) but they don't want Dev's to use it because they are planning to remove the component in a future cost-reduced version of the PS3.

If that were to be the case hopefully they plan to replace it with something that has enhanced functionality but may not be directly compatible with the existing SDK functions, they may however be confident that the horizontal scaling functionality could be easily emulated on the new component.

That would potentially leave existing PS3 owners feeling shortchanged if they are missing some features but if it means Sony could sell a cheaper box with greater functionality it could explain the current situation.

 

[Rangers]

http://ozymandias.com/archive/2007/01/29/Bruce_2700_s-Thoughts-on-PS3-Horizontal-Scaling-in-SDK.aspx

 

[inefficient]

http://ozymandias.com/archive/2007/01/29/Bruce_2700_s-Thoughts-on-PS3-Horizontal-Scaling-in-SDK.aspx

He is a Microsoft empolyee. Of course his pov is going to be negative for the PS3.

 

[phat]

He is a Microsoft empolyee. Of course his pov is going to be negative for the PS3.

His "horizontal scaling is cheap and easy and says nothing about a general HW scaler" assertion is perfectly reasonable.

Phat

 

[Crossbar]

http://ozymandias.com/archive/2007/01/29/Bruce_2700_s-Thoughts-on-PS3-Horizontal-Scaling-in-SDK.aspx

Except, it will be pretty weak 1080 support with an odd and substandard result. The horizontal resolution will be worse than with 720p (960 across instead of 1280), and the images will be twice as blurry horizontally as vertically. 960x1080 is going to look worse than 1280x720

Is he saying that a 960x1080 scaled to 1080p will look worse than a 720p scaled to 1080p, or what is he trying to say?
How is it possible that by adding 12.5% more pixel information it will make the picture look worse (edit: assuming you have good scaling filters) ?
Am I missing something here?

I found this text in ozymandias about page:

With full credit to Major Nelson, I’m going to crib a bit of his About text as it will apply to what I say and do as well.
...

I don´t want to sound prejudiced, but it does not really add to his credibility in my opinion, but maybe they know their stuff?

When I first got to hear about this added scaling function I thought, that maybe that was the reason why Motorstorm got the extra delay, and maybe Peter Dille was not totally lieing at CES when he was talking about Motorstorm as a 1080p title. Will be interesting to see.