A thought on X-Box 2 and Creative Labs.

Brimstone

B3D Shockwave Rider
Veteran
This is just speculation on my part, but isn't Creative Labs in a position to get in on the X-Box 2 design proposals now? They certainly have a deep patent portfolio on the audio side of things, and 3D Labs is certainly in no slouch when it comes to graphic processor design.

If Microsoft is taking proposals on a X-Box 2 design, it would seem to me Creative Labs could give ATI and Nvidia a run for the money. Am I wrong?
 
Who's going to produce the chipset (north and south bridge), motherboard design and memory architecture?
 
From what I understand the X-Box isn't the total result of Nvidia labor. The motherboard design was by Intel wasn't it? I doubt Creative Labs would be the complete solution, but they can certainly throw their hat into the ring for some critical components couldn't they?
 
Who's going to produce the chipset (north and south bridge), motherboard design and memory architecture?

Its been said before that some of the things in the North / South bridge are fairly off the self designs anyway.
 
you know what would be really cool? Intel gets to supply the CPU, but must also fab the gfx chip. They are making CPUs in .18u with over 200 million transistors (McKinley).

Can you imagine an XBox2 chip on an Intel .09u copper process?

hmmm. How fast would a 400 million transitor gfx chip running at 1GHz
be? (no power constraints since it's not being run over AGP).

Ah well. seriously though, shouldn't Intel be able to come up with much much better than i845G?
 
psurge,

The fact that Intel can pack a significant amount of trans. on a single chip isn't necessarily the make/break deal...It's how they can balance this with cost...

There are a lot of outfits, nVidia/ATI included, that could produce some Mega huge chip...but at a cost that would make it prohibitive...and I think if you take a look at what Intel is charging for those processors, it pretty much says it all.

The trick is in trying to balance complexity, cost, yields, execution (hehe...let's not even discuss how ridiculously long/delayed that part was).
 
M$ would be mad to not opt for PowerVR in the X-box 2, especially as they are moaning to Nvidia about the cost of the GFX core.

The X-box uses unified ram, where powervr shines. Via are also rumoured (lol) to be buying the STM graphics division which comes with PowerVR licences. With Via having the tech licenced they can go with via for the whole X-box design (cept the audio) and it wouldn't require an enourmous amount of RnD because a lot of what would be the design has already been designed for want of a better expression.

This adds up to cheap hardware and they need their hardware cheap, especially if they are putting Intel CPU's in the damn thing;)

Dave
 
Dave B(TotalVR) said:
M$ would be mad to not opt for PowerVR in the X-box 2, especially as they are moaning to Nvidia about the cost of the GFX core.

Actually, they'd be smart to not use PowerVR. Connecting the cost [silicon]of hardware to the price nVidia is charging is pretty crazy. MS should have outlined a multi-year plan that included price drops that will accompany a manditory decrease in die size - using the president that as costs are lowered and the 'box matures slaes will increase dramtically.

nVidia is just raping MS, plain and simple... Sony has already started production of their GS [eDRAM] using a 0.13um process, and has been highly agressive in moving to smaller processes and will soon consolidate the EE & GS into a single 0.1um chip... While nVidia has done nothing. Thats the problem.


Besides, the NG console will need to be enormously flexible while maintaing a high sustained preformance - I don't see PowerVR doing this. Sony's CELL looks to compleatly remove any fixed functionality from the entire 3D pipeline.

PowerVR's region based, deffered rendering scheme wouldn't be my choice for CELL competition. nVidia's by far the company with the more agressive and forward-looking vision; It's me belief that the combined nVidia-3dfx team will soon prove to be the preemptive and unrivaled PC graphics company, beginning with the release of their first new architecture since the RIVA TNT.

Besides, if your covering the entire screen with polygons that are a pixel in size, what good is a tiler? Hell, with a sustained TFlop/sec of power, you could design your own tiler...
 
Actually, they'd be smart to not use PowerVR. Connecting the cost [silicon]of hardware to the price nVidia is charging is pretty crazy."

Well, regardless the silicon will be cheaper for the same performance so they can still charge less and that is a very important factor in consoles.

MS should have outlined a multi-year plan that included price drops that will accompany a manditory decrease in die size - using the president that as costs are lowered and the 'box matures slaes will increase dramtically.

Well, this is besides the point, but yeah.

nVidia is just raping MS, plain and simple... Sony has already started production of their GS [eDRAM] using a 0.13um process, and has been highly agressive in moving to smaller processes and will soon consolidate the EE & GS into a single 0.1um chip... While nVidia has done nothing. Thats the problem.

The whole console is too expensive, the GPU is only part of that. Using PowerVR tech can reduce the overall cost because memory speed (and hence price) need not be as high and also certain graphics processes can be done in hardware on tilers but not on traditionals, such as translucency sorting. This can mean getting away with a slower main processer, or just living with that extra performance.

Besides, the NG console will need to be enormously flexible while maintaing a high sustained preformance - I don't see PowerVR doing this.

Why on earth not? It is not effected by opaque overdraw which goes up and down in-game, it has no trouble with the memory inefficiency associated with small polygons and the fundamental thing to remember is that while current PowerVR incarnations do not offer flexible graphics fucntions such as pixel shaders and the like there is no reason why these cannot be implimented with the technology. Indeed there is no doubt prodcut designs with these technologies included in the RnD pipeline. The design of PowerVR series 4 is complete, even if it doesn't have the features you can bet series 5 will.

Sony's CELL looks to compleatly remove any fixed functionality from the entire 3D pipeline.

Again there is no reason why such a concept could not be applied to PowerVR technology, indeed the fact that the display list is stored could allow for additional effects. A more programmable version of the old generalised modifier volumes (tiler only again) found on the Neon250 and DC could offer inviting oppertunities.


PowerVR's region based, deffered rendering scheme wouldn't be my choice for CELL competition.

That's like saying ATi's heirachal z-buffer technology would not compete well against Nvidias FSAA techniques.


nVidia's by far the company with the more agressive and forward-looking vision;

In the case on Nvidia you are correct, but you overlook that the 'aggressiveness' of the application of PowerVR tech is down to the licencee, not PowerVR technologies or IMGTEC. I think it would be safe to say that Via are reasonably aggressive in the PC space.

It's me belief that the combined nVidia-3dfx team will soon prove to be the preemptive and unrivaled PC graphics company, beginning with the release of their first new architecture since the RIVA TNT.

I dont really know the specifics, but didn't Nvidia just buy all the assets and such, not the employees? I suppose it would have been silly to not at least try and pick up a few boffins on the way.

Besides, if your covering the entire screen with polygons that are a pixel in size, what good is a tiler? Hell, with a sustained TFlop/sec of power, you could design your own tiler...

A tiler would be fine in that situation, in the arcades PowerVR has already been combined with very powerful TnL units.

Inside NAOMI2 NAOMI2 is a highly optimized parallel rendering architecture featuring twin PowerVR rendering engines, using adaptive load balancing across the dual engine and a full hardware Geometry Transformation and Lighting processor. With no-penalty for up to six fully featured lights NAOMI2 can provide 10M triangles/sec sustained throughput in real gaming applications.
 
nVidia is just raping MS, plain and simple... Sony has already started production of their GS [eDRAM] using a 0.13um process, and has been highly agressive in moving to smaller processes and will soon consolidate the EE & GS into a single 0.1um chip... While nVidia has done nothing. Thats the problem.

I find it amazing that Nvidia isn't alarmed by lower than forecasted sales in Europe. They are in such a strong postion in so many markets, it comes across as short sited not to do whatever it takes to make the X-Box triumph. Of course they are out to make money, but Microsofts gripes with the pricing of Nvidias silicon doesn't bode well. The X-Box is about high volume sales not large profit margins. Microsoft has the financial stamina to take a loss on each console sold, but they aren't going to be stupid about it. If at the end of the day this dispute changes the warm friendly relationship between M$ and Nvidia to something frigid, it would be the first real blunder Nvidia has made in a long time.


The whole console is too expensive, the GPU is only part of that. Using PowerVR tech can reduce the overall cost because memory speed (and hence price) need not be as high and also certain graphics processes can be done in hardware on tilers but not on traditionals, such as translucency sorting. This can mean getting away with a slower main processer, or just living with that extra performance.

Microsoft got a very nice console for the money. The hype of having Nvidia in the console helped make the X-Box seem like a legitimate attemp at the home console market. If the first generation games only have scratched the surface of the consoles power, I'm sure some incredible games are on the way.
 
PowerVR's region based, deffered rendering scheme wouldn't be my choice for CELL competition.


You responded: That's like saying ATi's heirachal z-buffer technology would not compete well against Nvidias FSAA techniques


Not at all - I think your not grasping what the NG console will look like - Well what i believe it will be, and I'm pretty good at putting it together.

What I talk of programmability, I'm talking general computing on the level of a CPU, not these shaders that share some CPU qualities. Basically, I see a TFLOP/sec of sustained preformance comming from a single chip which has a massivly parrallel array supported by eDDRAM and wide internal busses. There would be NO fixed functionality at all in this processor. Fragment shaders would be absent, instead it will probobly run a high-level vertex shading language that achieves near sub-pixel accuracy due to the high polygon rate. The developer than can opt to design a fully software tailer made pipeline that suites his/her game or scene - perhaps this will even be a form of 'chunking' or region based system.

Faf from the Console Forum is really good at this kind of conversation, he seems to be excited by the posibilities that the NG Consoles will provide for programmers.

is that while current PowerVR incarnations do not offer flexible graphics fucntions such as pixel shaders and the like there is no reason why these cannot be implimented with the technology. Indeed there is no doubt prodcut designs with these technologies included in the RnD pipeline. The design of PowerVR series 4 is complete, even if it doesn't have the features you can bet series 5 will

I sort of adressed this above... But, if anything, a hardwired pixel pipeline like that used in Tiling/Chunking would be the exact opposite of what I'm talking about. Your basically locked into the specific rasterization method with tiling... not so with other ideas that are being proposed.

nVidia's long term goals that I saw (4-5 years, as of last year) show them moving to a high-programmable state; beyond the limited 'shaders' of today. It's only a matter of time untill 'features' disappear completely and the developer has nothing but massive amounts of sustained preformance to play with... Pretty much the form of dynamic media processing envisioned by Diefendorff. I don't see PowerVR is that kind of marketplace, but I do an nVidia.

I dont really know the specifics, but didn't Nvidia just buy all the assets and such, not the employees? I suppose it would have been silly to not at least try and pick up a few boffins on the way.

AFAIK, over a 150 people ended up there.
 
At the time of RnD, I think Sega would have had something to say about that;P

:) yeah you might be right, was the contract with Sega that restricting? PowerVR worked really well with Dreamcast and Naomi, I think it has proven it self to be a viable thing.
 
Well vince, for either of us to really know we would have to have details of this architecture you speak of. Also, whilst the scene is split into tiles and HSR done, the actual unit of the chip that rasterises, shades and such is the same in design as any IMR, except the 'resolution' is always 32x16 and most of the memory access is done on chip istead of going ot external RAM.

Just because you clip all the polygons intot separate 32x16 tiles why cant you have a fully programmable pipeline, with no fixed functions except that of the clipping (hell, can you say dynamic tile sizes:)?
 
Vince,

Have you checked Img Tec's "Meta" ( www.metagence.com ) technology? There's not only a nice audio processor for a next gen console, there's ( somewhat more theoretically ) also multi-purpose programmability that might reveal something about Series 4 or 5, possibly maybe...

Dave B,

I expected you to bring Meta in to this discussion ;)
 
A snip from the Metagence homepage for any lazy people:

"Using true hardware multi-threading, the META family of processors deliver, in a unified architecture, general purpose processing, complex DSP and multimedia capabilities, real-time operation and low power consumption. These advanced capabilities make the META processor cores ideal for the next generation of entertainment and communications products."

(Bold mine. New post because for some reason Edit didn't seem to work.)
 
Dave B(TotalVR) said:
Oh christ yes, I had completely forgotten about metagence!

Ok, it's a SoC device that seems to use some a form of fine-grained multithreading. It's reminescent of CELL, but on a smaller scale and I question their ability to upscale to 750GFlop/sec and above. Realise that it's just SMT, and sharing the same resources, and that it's never going to be as fast as a massive CMP array with the sustained bandwith to back it up. hell, IBM was even looking at multiple thread execution per cell, of which each cell is but a single part of a larger CMP array.

I'd still say nVidia, who will be churning out 250+Million transistor beasts by then will be the better partner... they have the infastructure, are proven and have the technology to back it.

Also, what does this have to do with PowerVR's Tiling that you're pimping for the 'box2 before? This Metagence seems to be more consumer orientated - perhaps is PowerVR executed brillinatly they could, but their track-record preceeds them.
 
Also, what does this have to do with PowerVR's Tiling that you're pimping for the 'box2 before? This Metagence seems to be more consumer orientated - perhaps is PowerVR executed brillinatly they could, but their track-record preceeds them.

Well, metagence could easily be used in the graphics core. It would allow for massive parallelism for a start IIRC. Also, load balancing with the CPU. Im no expert on metagence though, you'd have to ask some of the more informed people here about that;P
 
Back
Top