Nvidia's next graphics chip: GPU or VPU

[alexsok]

I expect NV30 to be more flexible than P10.

Will it support branches, loops and subroutines in the pixel shader?

No, but it supports Instruction predicates, Unlimited dependent textures, Swizzling, advanced instructions, conditional write masks among other features.

Allow me to remind u that P10 doesn't even support Pixel Shaders 2.0 and NV30 goes beyond them.

Besides, vertex processing doesn't need many instructions, just loops & branches, etc, while in fragment processing, u don't need all of that, u only need many instructions support and NV30 offers just that.

 

[Dave Baumann]

No, but it supports Instruction predicates, Unlimited dependent textures, Swizzling, advanced instructions, conditional write masks among other features.

P10 supports many / most of these AFAIK, and R300 also supports some as well. Instruction predicates are more difficult for the developer than just supporting conditional logic.

Allow me to remind u that P10 doesn't even support Pixel Shaders 2.0 and NV30 goes beyond them.

That’s more of a technicality due to the lack of the floating point pipeline (although bit stacking can be used to reach formats beyond 32bit int). AFAIK P10 supports or goes beyond the programmable requirements of PS2.0.

 

[Hyp-X]

Edit: cut most of the post since Dave beat me to state most of the things

Just one example where branches/loops comes handy:
http://www.wildfred.dsl.pipex.com/raycast/index.html

 

[DemoCoder]

No, but it supports Instruction predicates, Unlimited dependent textures, Swizzling, advanced instructions, conditional write masks among other features.

P10 supports many / most of these AFAIK, and R300 also supports some as well. Instruction predicates are more difficult for the developer than just supporting conditional logic.

This is only important if you're assembly language programmer. Smart HLSL compilers will use predicates where appropriate. For both the NV30 and R300, multipass will be used to handle the conditionals that can't be done simply by using predicates.

Allow me to remind u that P10 doesn't even support Pixel Shaders 2.0 and NV30 goes beyond them.

That’s more of a technicality due to the lack of the floating point pipeline (although bit stacking can be used to reach formats beyond 32bit int). AFAIK P10 supports or goes beyond the programmable requirements of PS2.0.

The P10 is more flexible in the way that a CPU is more flexible, but the question is, what's its performance like when executing these shaders? For simpler shaders, I would guess the performance will trail behind the R300/NV30. For more complex shaders, it's a fight between what's gonna be slower: multipassed conditionals on R300/NV30, or faked 128-bit FP using integer math on the P10.

 

[alexsok]

Edit: cut most of the post since Dave beat me to state most of the things

Just one example where branches/loops comes handy:
http://www.wildfred.dsl.pipex.com/raycast/index.html

Alright, I edited my previous post, it wasn't written very well I agree

Anyway, I never argued loops & branches are not important, they are, but for vertex processing, not fragment processing.

A well written review of Wildcat VP870 based on P10:
http://www.digit-life.com/articles2/p10/index.html

 

[alexsok]

BTW Dave, CHECK YOUR PM!!

 

[TheMightyPuck]

What exactly does GPU or VPU tell us about the chip(s). Not much. But it does suggest that the chip be treated on a level with the CPU. Nvidia or ATI would love it if people shopping for computers cared as much about the G or V PU as the CPU. Instead of Intel inside you could have Nvidia Nside. The fact is that for a heck of a lot of people the video card is more important than the CPU and they should pay more attention. As for naming conventions, I don't really know. ATI has used a word with an order of magnitude that captures pretty much what the chip does. Visuals. If you told someone a movie had great visuals you could be talkng about anything whereas if you said it had great graphics you'd probably be talking about something to do with special effects or titles, yada yada. I suspect if you asked a thousand people what graphics were and another thousand people what visuals were you'd see a better understanding of the latter term. Also, I think graphics is a bit anachronistic in describing what computers do visually. In the end arguing for one TLA over another is a bit silly. Like CPU, once the marketing term takes root, the actual words underpinning the acronym quickly become meaningless. When is the last time you heard someone say central processing unit instead of cpu? Of course when is the last time you heard someone say "hey, nice graphics, what kind of GPU do you have?"

 

[Dave Baumann]

The P10 is more flexible in the way that a CPU is more flexible, but the question is, what's its performance like when executing these shaders? For simpler shaders, I would guess the performance will trail behind the R300/NV30. For more complex shaders, it's a fight between what's gonna be slower: multipassed conditionals on R300/NV30, or faked 128-bit FP using integer math on the P10.

<Shrug> I was just pointing out the flexability in relation to the context of the thread.

BTW Dave, CHECK YOUR PM!!

I do, regularly - I don't think you've sent it to the right Dave.

 

[alexsok]

I do, regularly - I don't think you've sent it to the right Dave.

I doubt that, but I just sent it again, check now

 

[jb]

Embargo.

Hmmmm with the 9700 due to be at QuakeCON which is this week and the fact that Mr. B over at rage3d said that some reviewers have them. I whould expect that reviews are really close. Can you confirm this or is it safe to say reviews of the 9700 will be out soon?

Ty

 

[Joe DeFuria]

Well, since the 9700 is officially shipping on the 19th, I would have to believe that's the latest possible date to lift the embargo. I'm hoping it's lifeted before the weekend though...

 

[PC-Engine]

I've always liked the PowerVR acronymn of ISP and TSP. Maybe something like VSP?

 

[Sabastian]

This must be the epitome of what many here are talking about with regards to "noize".. Yet more banner waving by the Nvidia throng with regards to merchandise virtually a half a year away? I find it unappealing really and not note worthy. Simply put … who cares what Nvidia will endeavor to use as a catch phrase for their next product offering? Is this an exercise in mental masturbation? I believe so.

 

[Tahir2]

First time I heard the term GPU it was for the PSX.

 

[Colourless]

Since when has the Beyond3D forums cared for Marketing? Unless i'm thinking of another Beyond3D forum, things have been generally firmly anti-marketing.

This topic serves no purpose.

 

[Fuz]

Well, I'm sorry I bored you. Maybe I should have started yet another NV30 thread instead?

Thats ok, don't let it happen again! Let that be the first and final warning young man!

Nagorak, I did not mean to sound unfriendly or the like, even though it may seem that way.

 

[Kristof]

I've always liked the PowerVR acronymn of ISP and TSP. Maybe something like VSP?

Hmmm.. how about VGP :

Optional Vertex Geometry Processor (VGP) - Hardware transform and lighting (TnL) engine

http://www.arm.com/armtech.nsf/html/MBX3D?OpenDocument

K-

 

[OpenGL guy]

I prefer VPU: Vaporware Processing Unit.

P.S. Go ahead and flame if you have no sense of humor.

 

[LittlePenny]

Since when has the Beyond3D forums cared for Marketing? Unless i'm thinking of another Beyond3D forum, things have been generally firmly anti-marketing.

This topic serves no purpose.

I agree.

 

[Nagorak]

This must be the epitome of what many here are talking about with regards to "noize".. Yet more banner waving by the Nvidia throng with regards to merchandise virtually a half a year away? I find it unappealing really and not note worthy. Simply put … who cares what Nvidia will endeavor to use as a catch phrase for their next product offering? Is this an exercise in mental masturbation? I believe so.

LOL, if you're implying I'm an Nvidiot, you are totally off base.

As to this being just marketing, I think some of you are taking this topic too seriously. It was just a curiosity I had, nothing really firm. I don't really "care" what they call their GPU/VPU, I just thought it might be interesting to discuss a little. Consider it a "fluff" topic.