Cure For The Common Graphics Processing

[nAo]

It's more probable Jenna Jameson is a virgin than G70 is TBDR, imho

 

[j^aws]

It's more probable Jenna Jameson is a virgin than G70 is TBDR, imho

Maybe so...but if true, you'll luv-it. LUV-IT, I tell ya!

 

[Farid]

First things first, great article lazy8s, that was a good read.

Nevertheless, I regret that your article didn't relate the fact that, at the exception of one of the Intel portable solution, none of the "big" IHVs choosed the Deferred path for their graphical solutions, and why is that.

As is, the article sounds just like a description of the theorical advantages of the TBDRs.

You sould try to expose the reasons of why the IMRs are still being majoritarily used todays. By talking about the IMRs own methods to overcomes the bandwidth problems, such as Z-Buffer compression, tile rendering , etc...

And then ultimately, do a comparison of the both architectures and technological choices. Explaining why you think TBDRs are the way to go for the future graphical needs.

That would have create an interresting discussion, too.

But as I said, it was a good reading, lazy8s. As far as I'm concerned, don't hesitate to post any of your future articles.
That'd change from the fawful threads started by trolls pretending being Bill Gates or whatever.


Now about the PS3GPU, potentially being a TBDR, I would say the chance are, really, extremely low. Seeing that the part is going to share the same architetcture as its PC counterpart.

But that would be killer! IF true...

 

[Inane_Dork]

I'm guessing that deferred rendering has not been chosen by the big dogs because it has to be implemented transparently to the app, and that raises lots of... issues. I'm just going by failed attempts, here. I have no knowledge of what the issues actually are.

If you set out to make a dedicated platform and decide it will use a deferred renderer, I'm sure it would be completely feasible. Worth it? Well, I dunno. The so many games anymore are using depth-only first passes seems to indicate that deferred rendering is fundamentally the better algorithm. It probably depends on implementation.

 

[FiggyG]

That was a great read. I really wonder why TBDR didn't really make it on the desktop scene?

 

[MechanizedDeath]

Argh! How much does NEC pay your viral marketers to continue pushing their propoganda on the public? Deferred rendering (which this is) has been repeatedly ignored by NVidia and ATI for a good reason. What that reason is may remain unknown for many years. But the bottom line (dollars and cents) obviously won't be helped. Let it go. It won't happen, probably ever. Let it go. Besides, PVR tech has never been featured in a world-class card. Is this a coincidence? Are they sandbagging intentionally, or is it simply not adequate for the task at hand? It's always used in low-end products. Probably b/c it's a low-end solution. Sorry if it sounds harsh, but it needs to be said. This is becoming a damn pipedream. PEACE.

 

[Iron Tiger]

Argh! How much does NEC pay your viral marketers to continue pushing their propoganda on the public? Deferred rendering (which this is) has been repeatedly ignored by NVidia and ATI for a good reason. What that reason is may remain unknown for many years. But the bottom line (dollars and cents) obviously won't be helped. Let it go. It won't happen, probably ever. Let it go. Besides, PVR tech has never been featured in a world-class card. Is this a coincidence? Are they sandbagging intentionally, or is it simply not adequate for the task at hand? It's always used in low-end products. Probably b/c it's a low-end solution. Sorry if it sounds harsh, but it needs to be said. This is becoming a damn pipedream. PEACE.

PowerVR's console deal threw off their PC timetables. Were it not for Dreamcast related delays in their development cycle, the KYRO could have beaten the GeForce 256 to the market and become a mainstream hit. Yeah, it's all conjecture, but so is saying that a TBDR isn't a mainstream hit because ATi and NV aren't using it.

 

[Lazy8s]

Every timely PowerVR release has led all comparable chips decisively. Dreamcast wasn't matched when Voodoo3 and TNT2 were released months later. No cards in Kyro II's class scored anywhere near it. MBX is still the most advanced handheld graphics technology.

Technology is rated by how much capability is delivered for a given set of costs like price (chip size), power consumption, and heat dissipation. Saying that the only good graphics chip is a high-end graphics chip is like saying that the only impressive computer is a supercomputer. Cost is inextricably involved, and an argument not taking that into account ends up making no sense.

NEC's partnership with Imagination Technologies and their PowerVR graphics division to bring the technology to all applicable markets ended a long time ago. NEC is just a customer now, but they'll probably be expanding their involvement from the legacy products they currently do to an MBX license soon and possibly even to new arcade systems like Lindbergh/NAOMI3.

 

[Neeyik]

Neeyik:

it's usually polite to quote sources for such articles.

I wrote it trying to provide a very general overview of tile-based display list rendering since the topic is only first being introduced to many people.

I wasn't inferring that you didn't write it all - I was trying to point out that if you are going to pinch images from a site, you should at the very least offer links to the sources; that way you're not just leeching bandwidth for your sake. A nice bit work by the way, but rather wasted in a forum!

edit: typo corrected

 

[Lazy8s]

You're right; I hadn't much considered the implications of linking borrowed pictures before. Thanks for the tip. I'll edit the sources in.

 

[Squeak]

AFAIK. the two major problems with TBDR is unequal distribution if polys in the different tiles, and how too deal with sorting and storing an increasingly large number of polys.

The solution to the first problem is of course to subdivide polygon heavy tiles into two or more.
The solution to the second problem is to use simple geometry when sorting, and then in the rendering process, with ondie TnL, adding the extra polygons through either HOS or displacement mapping (losing a little of the non overdraw advantage, but not critically much).
Neither of these techniques are trivial to implement, and the space all that extra logic takes up is perhaps what is holding nVidia and ATI back, TBDR simply being deemed "not worth the trouble".

 

[PC-Engine]

AFAIK. the two major problems with TBDR is unequal distribution if polys in the different tiles, and how too deal with sorting and storing an increasingly large number of polys.

The solution to the first problem is of course to subdivide polygon heavy tiles into two or more.
The solution to the second problem is to use simple geometry when sorting, and then in the rendering process, with ondie TnL, adding the extra polygons through either HOS or displacement mapping (losing a little of the non overdraw advantage, but not critically much).
Neither of these techniques are trivial to implement, and the space all that extra logic takes up is perhaps what is holding nVidia and ATI back, TBDR simply being deemed "not worth the trouble".

I believe Imgtech already have solutions to the issues you mentioned above, however, I don't know how effective it is or how expensive it is in terms of logic. IIRC they also patented the solutions.