PS3 - who's really benefitting?

[ninelven]

18+ months? or was that a rhetorical question :? ?

 

[jvd]

18+ months? or was that a rhetorical question :? ?

and that is from the first parts that were capble of it came out right

18 months after 8500 , 18 months after r300 ?

so i guess ati has 18 months to be as cutting edge as nvidia

 

[Eronarn]

so i guess ati has 18 months to be as cutting edge as nvidia

Gasp, how ever will they do it?

 

[DemoCoder]

The shame of it is, one of the big reasons to want to use VS3.0 would be for displacement mapping. However NV40 doesn't have any higher order surface support, so it greatly limits the potential. Even Npatches would have been nice, even with as many faults as they have. And of course ATI doesn't even support VS3.0 which is just disappointing.

You could get away with using a pre-tesselated mesh, especially for terrain displacement maps, you could combine a pre-tesselated mesh with geometry instancing. Of course, you won't get LOD dependant tesselation, but these cards are fast enough to deal with it.

Mint,
I think you'll find the're something strange going on with the RTHDRIBL demo that will be rectified in the future.

 

[WaltC]

ATI has higher image quality and speed, and while we don't have PS3 I am confident that we'll look back on this a year from now with the R420 as the winner, even if only by a slight margin. Sure, in the long run, ATI may suffer a bit for not having it... but I doubt any games will go PS3-only and not have an alternative in PS2.

What I understand in reviewing the specifications for R4x0, is that R4x0 does indeed include hardware feature support for several ps3.0 specifications--just not all of the ps3.0 specifications--generally declining to support those that ATi apparently felt weren't worth their weight in transistors so as to justify their inclusion in R4x0.

What I get from reading various comments attributed to ATi personnel is that certain ps3.0 functionality wasn't supported for the sake of performance because ATi believed that implementing that support would negatively impact performance should developers opt to exploit that support, which meant they had to ask themselves this question at some point: "Do we want to add transistors that support features we don't think developers will actually choose to exploit since that support will adversely impact the performance of their software while providing no other benefits, and further, do we want to risk yields by adding circuitry we don't believe will actually be utilized by developers working with our hardware?"

I'm guessing that ATi opted for "yields/processing efficiency" over a marketing bullet here, while at the same time it appears to me that they did support in R4x0 the ps3.0 functionality they deemed worthwhile. It doesn't appear to me that R4x0 categorically doesn't support ps3.0 functionalty, just that it doesn't support all of it. If ATi's approach is correct, and since I've been assured by numerous parties that ps2.x and ps3.0 can be compiled for simultaneously with a minimum of bother, then how could ATi lose?...

What I'm looking forward to is David Kirk's upcoming talk at the UN, where he'll be delivering a speech entitled: "Why We at nVidia Today Believe that ps1.x is Not the Future of 3d-Gaming, After All, Even Though We Stated That Often Several Times Last Year." That's one I don't want to miss. What I hear is that he'll be following that up with an impromptu Q&A session at the White House, after delivering a brief speech to the White House press corps which is advanced as: "Why nVidia Never Actually Quit the FutureMark Program Last Year, Even Though You Might Have Thought We Did: A Case of the Tail Wagging the Dog". Now that's interesting stuff...

 

[ninelven]

and that is from the first parts that were capble of it came out right

18 months after 8500 , 18 months after r300 ?

so i guess ati has 18 months to be as cutting edge as nvidia

Well with regard to the 8500 and ps1.4 it never really came out to be a significant advantage. On the other hand, NV certainly suffered for its lack of PS2.0 performance over the past 18months. The question is how big of a performance advantage will SM3.0 be able to provide (comparitivly speaking... not absolutely). It could turn out to be something akin to ps1.4 or something much more significant like SM2.0 vs SM1.x today. I doubt it will be either extreme, probably somewhere in the middle.

 

[jvd]

and that is from the first parts that were capble of it came out right

18 months after 8500 , 18 months after r300 ?

so i guess ati has 18 months to be as cutting edge as nvidia

Well with regard to the 8500 and ps1.4 it never really came out to be a significant advantage. On the other hand, NV certainly suffered for its lack of PS2.0 performance over the past 18months. The question is how big of a performance advantage will SM3.0 be able to provide (comparitivly speaking... not absolutely). It could turn out to be something akin to ps1.4 or something much more significant like SM2.0 vs SM1.x today. I doubt it will be either extreme, probably somewhere in the middle.

well everyone agrees that branching , one of the big things on p.s 3 is going to cause a major hit .

But regardless . For two generations ati pushed features ahead of nvidia and now they take a break for a generation and everyone is upset .

Imagine how it would be if the geforce 4 line had ps 1.4 and if the geforce fx line could do 2.0 at the same speed in full persicion like ati with out all those headaches

 

[Mintmaster]

Mint,
I think you'll find the're something strange going on with the RTHDRIBL demo that will be rectified in the future.

Yeah, probably. I think there may be something strange with R420's multitexturing too, and hopefully that'll get fixed. Never expected those results.

I'll definately admit that on a per clock basis, the NV40 pixel shader architecture is awesome. We still have to see what compiler improvements will do, though.

Time will tell...

 

[ninelven]

well everyone agrees that branching , one of the big things on p.s 3 is going to cause a major hit

I'm not sure who this "everyone" is, but the obvious answer is always there: only use branching if it is a performance win. From my understanding, there are indeed useful situations where this occurs.

But regardless . For two generations ati pushed features ahead of nvidia and now they take a break for a generation and everyone is upset .

That's probably just because most people want to have their cake and eat it too. It will always be performance versus features. IMHO, as a gamer, the current situation is ideal because the two major IHV have chosen quite different paths (and both are being very successful with them). This provides more options for me such that I can make my purchasing decisions based upon my own preferences/needs.

 

[Ostsol]

The shame of it is, one of the big reasons to want to use VS3.0 would be for displacement mapping. However NV40 doesn't have any higher order surface support, so it greatly limits the potential. Even Npatches would have been nice, even with as many faults as they have. And of course ATI doesn't even support VS3.0 which is just disappointing.

You could get away with using a pre-tesselated mesh, especially for terrain displacement maps, you could combine a pre-tesselated mesh with geometry instancing. Of course, you won't get LOD dependant tesselation, but these cards are fast enough to deal with it.

Or there's OpenGL's Uber-buffers and render-to-vertex-buffer -- if the extension spec ever gets finished and only for OpenGL, of course. . . :?

 

[christoph]

dave orton just confirmed in daves interview that r400 was removed from the pc roadmap because of execution problems. so r420 basically is a gap filler, but a real good one at that.