So what do you think about S3's DeltaChrome DX9 chip

[DemoCoder]

Of course, given that 8 concurrent pixels are being evaluated, and that the order of evaluation is not guaranteed during rasterization, it's probably of no use anyway since if you try to smple read from a framebuffer pixel, it could have either not been evaluated yet, is currently being evaluated by a neighboring pipeline, has been evaluated but not written over by an occluder (yet), or has been evaluated but written over by an occluder, etc.

If the pipeline doesn't have logic in to sort out overlapping read/writes, then it is faulty. Essentially, each pixel has to be 'checked out' and 'checked in' - it can't be read if it's checked out. Therefore, overlapping rendering generates pipeline bubbles, not incorrect rendering.

It's not the issue of overlapping writes, the the issue of not being guaranteed an evaluation order. If I write a pixel shader and I try to utilize the value of the pixel that is to the left of the current pixel as a piece of dependent data, I can't be guaranteed the state of that pixel.

Imagine trying to use the frame buffer to compute a fibonacci sequence, but reading the two frame buffer pixels before you, and using that to output the current pixel.

So what's the use trying to read from the current framebuffer you are rendering to? It breaks the stateless model of no-interdependencies between the pipelines.

 

[Inspirer]

please challenge my theory at:
this thread

 

[Dave Baumann]

Nappe1, WRT Creative/3Dlabs consumer parts, errr cough

 

[Ailuros]

:drumroll: I am just trying to be realistic on my comments this time. S3 is the only one (from SiS, PowerVR, S3, 3DLabs.) that has introduced / released something.

Introduced as in a paperspec announcement. If it makes it on shelves within a couple of months then fine. In any other case it'll resemble more to the XP/Trident "flop" which was "introduced" when and will be available when again?

I prefer to hear announcements from any IHV or the marketing hype machine to start only when any of them is 100% sure that the product is ready for mass production. I am willing to believe that it's the case with S3 here.

As far as the rest of the IHV's listed goes, we'll see when they're ready and what they've been working on for the past years. As much as there is a chance that the specs could be lower than Delta Chrome, there's also the chance that they're significantly higher too.

In terms of reality I'll take that "something" for granted when it's sitting on shelves.

 

[Basic]

DemoCoder:
It can probably read the color-/z-value of the old pixel at the same pixel position as current. That should be a lot easier. Only question is how it should behave with MSAA. Use average of all subpixels, or all subpixels that the fragment eventually will be written to, or maybe something else? (See _centroid modifyer in PS 3.0.)

 

[Dio]

What he just said. Basically I don't forsee it having offset reads - just 'same pixel' reads.

To do offset reads I imagine you will still need to render to a different target from the source, for exactly the reasons you state.

As regards what's the use - that's a philosophical question, not a technical one

 

[Nappe1]

:drumroll: I am just trying to be realistic on my comments this time. S3 is the only one (from SiS, PowerVR, S3, 3DLabs.) that has introduced / released something.

Introduced as in a paperspec announcement. If it makes it on shelves within a couple of months then fine. In any other case it'll resemble more to the XP/Trident "flop" which was "introduced" when and will be available when again?

argh...
yes, but YESTERDAY (because situation has changed again.) the others on my list included, didn't have even those papers out! S3 was the only one having even paper stuff out with some kind of shipping and availability schedules.

I am pretty much full of all this. First , I am not liked because I trust companies that HAVE NOTHING. And now when I based my opinions raw facts and that only company that had SOMETHING, (instead 3 others that did have yesterday nothing.) I am again an asshole. Plus, my excelent superior unbelievable luck, brings 3DLabs consumer launch to the next day after I have been saying that I don't see 3DLabs coming gamers market and puts me on laughing stock completely. Maybe I should say "I do exist" and next day I would be vanished like fart to the Sahara desert.

I am finished for this week. see you on Monday.

 

[Tahir2]

:drumroll: I am just trying to be realistic on my comments this time. S3 is the only one (from SiS, PowerVR, S3, 3DLabs.) that has introduced / released something.

Introduced as in a paperspec announcement. If it makes it on shelves within a couple of months then fine. In any other case it'll resemble more to the XP/Trident "flop" which was "introduced" when and will be available when again?

argh...
yes, but YESTERDAY (because situation has changed again.) the others on my list included, didn't have even those papers out! S3 was the only one having even paper stuff out with some kind of shipping and availability schedules.

I am pretty much full of all this. First , I am not liked because I trust companies that HAVE NOTHING. And now when I based my opinions raw facts and that only company that had SOMETHING, (instead 3 others that did have yesterday nothing.) I am again an asshole. Plus, my excelent superior unbelievable luck, brings 3DLabs consumer launch to the next day after I have been saying that I don't see 3DLabs coming gamers market and puts me on laughing stock completely. Maybe I should say "I do exist" and next day I would be vanished like fart to the Sahara desert.

I am finished for this week. see you on Monday.

Does this mean that you have cursed the S3 DeltaChrome now as well?

 

[BoddoZerg]

Anyone think this is a bit imbalanced?

Initially, both the core and the 128-bit DDR-I memory interface will run at 300-MHz, generating a 2.4 gigapixel fill rate and memory bandwidth of 11 Mbytes per second.

I hope for their sake that their deferred rendering is really damn good...

Then again, if they plan on releasing the $20 version with 16 MB of RAM and clocked at 120 MHz, it might not matter.

 

[Thowllly]

Anyone think this is a bit imbalanced?

Initially, both the core and the 128-bit DDR-I memory interface will run at 300-MHz, generating a 2.4 gigapixel fill rate and memory bandwidth of 11 Mbytes per second.

I hope for their sake that their deferred rendering is really damn good...

Then again, if they plan on releasing the $20 version with 16 MB of RAM and clocked at 120 MHz, it might not matter.

That's no more imbalanced than the GforceFX or the Radeon 9500PRO

(the memory bandwidth would of course be 300*2*16=9.6GBps, not 11MBps)

 

[Randell]

Well people are accusing the Fx of being unbalanced and it has a similar fillrate:bandwidth ratio. Of course nVidia has its occlusion detection, memory controllers etc to get more effective bandwidth - so yes in comparison to the 9700Pro it does look bandwidth starved. Its deferred rendering needs to be efficient to compete.

 

[Ailuros]

I am pretty much full of all this. First , I am not liked because I trust companies that HAVE NOTHING. And now when I based my opinions raw facts and that only company that had SOMETHING, (instead 3 others that did have yesterday nothing.) I am again an asshole. Plus, my excelent superior unbelievable luck, brings 3DLabs consumer launch to the next day after I have been saying that I don't see 3DLabs coming gamers market and puts me on laughing stock completely. Maybe I should say "I do exist" and next day I would be vanished like fart to the Sahara desert.

I haven't the slighest idea where all the aggrevation (or should I say temper) comes from.

To put this into perspective and before it causes any more misunderstandings: I want to see products on shelves and that affects pretty much ANY vendor out there. On paper they all look fine, once they get tested by intependant parties and sufficiently analyzed 9 out of 10 they're usually less than what I could call even exiting.

I'm tired of powerpoint presentations and paper launches and wether you like it or not S3 is amongst those companies that will have to put up some extra effort to regain some confidence. Last time I spend my hard earned money on one of their products, the best way I could describe it was a hardware virus. (edit: for those with a weak memory the ViperII's were officially replaced back then).

Before I call a S3 product even "something" it has to be on time and very well deserve it's price; any other case it's just another SiS trash effort that leaves me fairly in the cold.

 

[Randell]

Ail - all this rooting for someone to break the nVidia/ATI duopoly has left Nappe1 highly strung this past year or so - Bitboys/Matrox/PVR/Trident have all left their mark

Dont worry Nappe1 I dont see anybody hating you or calling you names - unless you want to add drama queen to the list

 

[OVERLORD]

I am pretty much full of all this. First , I am not liked because I trust companies that HAVE NOTHING. And now when I based my opinions raw facts and that only company that had SOMETHING, (instead 3 others that did have yesterday nothing.) I am again an asshole. Plus, my excelent superior unbelievable luck, brings 3DLabs consumer launch to the next day after I have been saying that I don't see 3DLabs coming gamers market and puts me on laughing stock completely. Maybe I should say "I do exist" and next day I would be vanished like fart to the Sahara desert.

I am finished for this week. see you on Monday.

Chill out dude. Everyones opinions still appreciated even yours. I just hope they're not the GPU equivalent to CYRIX!

 

[McElvis]

More info over at EETimes -

The first incarnation of the core will be a mobile graphics processor that the company plans to announce on January 16.

Still, the company said it had to cut some fat to stay within the power guidelines of notebook computers, currently S3's biggest market. Displacement mapping, which the company said is rarely used, was left out. So was 128-bit rendering capability, which S3 considers overkill.

S3 will offer three versions of DeltaChrome graphics processors for mobile systems, two of which will be delivered in a multichip package that includes frame buffer memory.

http://www.eetimes.com/printableArticle?doc_id=OEG20030108S0042

 

[OpenGL guy]

I just hope they're not the GPU equivalent to CYRIX!

Funny you should mention Cyrix... Via owns them too

 

[Humus]

What he just said. Basically I don't forsee it having offset reads - just 'same pixel' reads.

To do offset reads I imagine you will still need to render to a different target from the source, for exactly the reasons you state.

As regards what's the use - that's a philosophical question, not a technical one

I think having the stored depth/color values accesible in the fragment shader would be a good idea. It solves the problem of blending on floating point buffers (we could implement any blending of our choise) and adds a lot of flexibility.

 

[psurge]

What about multi-sampling? Wouldn't the blend have to be performed multiple times for a given shader result?

 

[Ailuros]

Ail - all this rooting for someone to break the nVidia/ATI duopoly has left Nappe1 highly strung this past year or so - Bitboys/Matrox/PVR/Trident have all left their mark

Dont worry Nappe1 I dont see anybody hating you or calling you names - unless you want to add drama queen to the list

Randell,

I realize that, but it'll take one hell of resources/time/efforts for any of the other IHV's to come even close. With the market being as critical as it is, it won't be an easy task for any of them; of course will some of them attempt, but if success is going to ring in then not with the first attempt even if it truly can be considered as a far better option/alternative.

It'll take years to gain marketshare and complete consistancy in releases. A product - how good it might be - once in a blue moon won't cut the cake; if it turns out underwhelming on top of that then the efforts is more likely doomed.

Consumers want to feel safe with their purchases and you don't gain their confidence from one day to the other.

 

[JD]

I tend to agree with trident when they've said that programable hw is easier to design and write drivers for and this makes it possible for other smaller companies to compete. It's also nice that MS is doing the hlsl that ihv can plug into. Maybe this is why s3 went with 2+ shaders over 2.