So is PS3 GPU one generation beyond Xenon graphics chip?

[Shinjisan]

Since you're so well informed here I have a question:

http://www.digitimes.com/news/a20041220A7039.html

"Nvidia has not given up on the NV48 and is expected to roll out the GPU in the second quarter of next year, despite market rumors saying the company has cancelled the chip, according to sources at Taiwan graphics card makers.

However, Nvidia plans to have Taiwan Semiconductor Manufacturing Company (TSMC) produce the NV48, instead of IBM, as originally planned, the sources noted. The NV48 will be built using a 0.11-micron process, the sources added.

Improved yields of the GeForce 6600 and GeForce 6200 manufactured at TSMC would probably be the reason for Nvidia to shift the production of the NV48 back to TSMC, the sources said.

The NV48 will take on ATI’s next-generation high-end graphics chip, the R520, which is also expected to be launched in the second quarter of 2005. The R520, however, will be the industry’s first GPU produced using a 90-nm process at TSMC."


Xenon GPU is code named R500,if ATI R520 is out in Q2 2005 competing with NV48 and the PS3 GPU will be based on NV5X (or equivalent technology),does this put the PS3 GPU potentially one generation beyond the Xenon graphics processor?
Of course we talk about a PC generation so a 6 months difference but if NVidia and Sony decide to use 65nm for PS3 GPU the gap could be really consistent.

 

[Qroach]

In the end does more powerful mean better? Will anyone see the difference is the real question.

 

[Shinjisan]

More powerful doesn't mean better games but better graphics for sure.

 

[arhra]

The R520 is completely unrelated to the xenon gpu.

 

[Brimstone]

My guess is both ATI and Nvidia GPU's destined for the Xenon and PS3 are designed for .90 nm. The amount of eDRAM will depend on if it is fabbed on .90 nm or .65 nm. The target for Sony would seem to be .65 nm, while .90 nm is more likely for MS. Supposedly TSMC will have .65 nm available to customers at the end of 2005, so the Xe R500 GPU has the possibility of being a .65nm product.


Xenon .90 nm CPU and GPU with 512mb of GDDR-4 or 1024mb of DDR-II with QBM-2 integrated into the SIS chipset to double the bandwidth of the standard DDR-II DRAM.


SIS jons the QBM alliance

 

[aaronspink]

My guess is both ATI and Nvidia GPU's destined for the Xenon and PS3 are designed for .90 nm. The amount of eDRAM will depend on if it is fabbed on .90 nm or .65 nm. The target for Sony would seem to be .65 nm, while .90 nm is more likely for MS. Supposedly TSMC will have .65 nm available to customers at the end of 2005, so the Xe R500 GPU has the possibility of being a .65nm product.

Probably a better way to think of it is that they are both targetted for 65 nM but that the initial implementation will be done in 90 nM for timing reasons. There is a tradition of console companies launching the console with die sizes that are expensive to seed the market while working on shrinking the design to the next process.

Xenon .90 nm CPU and GPU with 512mb of GDDR-4 or 1024mb of DDR-II with QBM-2 integrated into the SIS chipset to double the bandwidth of the standard DDR-II DRAM.

AFAIK, QBM never existed and QBM-2 certainly doesn't exist. Xenon is most likely using 4 32b wide 256Mb or 512Mb GDDR3/4 chips @ 1600 Mtps.

Aaron Spink
speaking for myself inc.

 

[pc999]

QBM-2 certainly doesn't exist

Why do you say that, IMO would be very good for Xe 2X the bandwidth at almost the same price, they win over the PCs in cost!

 

[Acert93]

Maybe some of the HW gurus can help me with this.

If R500 is at 90nm, the PS3 GPU would need to be at ~ 45nm to have twice as many transistors on the same die size, right?

And at 65nm it would be what? 30-50% more transistors?

Or am I way off...

Anyhow, I know that clock speed, die size, and the architecture make a big difference. If R500 taped out recently, and the PS3 will tape out at the end of 2005, Nvidia has about a 12 month area to do some work. Not sure if 12 months is enough to double the power... but all is speculation since we know nothing really about both chips. Who knows, maybe R500 will be some killer chip. I wonder what features, other than VS/PS, are being planned for these chips. I think some great new features will be more important than just raw speed (although that is important too), but that is just an opinion. Exciting time

 

[Brimstone]

My guess is both ATI and Nvidia GPU's destined for the Xenon and PS3 are designed for .90 nm. The amount of eDRAM will depend on if it is fabbed on .90 nm or .65 nm. The target for Sony would seem to be .65 nm, while .90 nm is more likely for MS. Supposedly TSMC will have .65 nm available to customers at the end of 2005, so the Xe R500 GPU has the possibility of being a .65nm product.

Probably a better way to think of it is that they are both targetted for 65 nM but that the initial implementation will be done in 90 nM for timing reasons. There is a tradition of console companies launching the console with die sizes that are expensive to seed the market while working on shrinking the design to the next process.

Xenon .90 nm CPU and GPU with 512mb of GDDR-4 or 1024mb of DDR-II with QBM-2 integrated into the SIS chipset to double the bandwidth of the standard DDR-II DRAM.

AFAIK, QBM never existed and QBM-2 certainly doesn't exist. Xenon is most likely using 4 32b wide 256Mb or 512Mb GDDR3/4 chips @ 1600 Mtps.

Aaron Spink
speaking for myself inc.

The QBM concept is real. You wouldn't have all those big name companies apart of the alliance if it didn't work.

I don't know if there's any technical reason stopping GDDR-3/4 from getting the QBM treatment.

 

[j^aws]

Maybe some of the HW gurus can help me with this.

If R500 is at 90nm, the PS3 GPU would need to be at ~ 45nm to have twice as many transistors on the same die size, right?

And at 65nm it would be what? 30-50% more transistors?
....
)

(90/65)^2 ~ 2x

The area available to transistors would be a square of the process drop...

 

[V3]

There is a tradition of console companies launching the console with die sizes that are expensive to seed the market while working on shrinking the design to the next process.

Where did you get that idea from ? It isn't a tradition, that's just what Sony did with PS2.

Looking at the shortage in the early stages, it isn't a good idea IMO. They're lucky that Sega managed to screw up Dreamcast marketing. This time around good supply is more essential than technology.

 

[Acert93]

Maybe some of the HW gurus can help me with this.

If R500 is at 90nm, the PS3 GPU would need to be at ~ 45nm to have twice as many transistors on the same die size, right?

And at 65nm it would be what? 30-50% more transistors?
....
)

(90/65)^2 ~ 2x

The area available to transistors would be a square of the process drop...

Thanks Jaws

 

[aaronspink]

QBM-2 certainly doesn't exist

Why do you say that, IMO would be very good for Xe 2X the bandwidth at almost the same price, they win over the PCs in cost!

QBM doesn't provide any benefit in an application that isn't DIMM based. For an application with directly connected memory, you will also be targetting to run at a much higher data rate than the QBM asics can handle.

Also QBM hasn't to my knowledge EVER been shipped.


Aaron Spink
speaking for myself inc.

 

[aaronspink]

The QBM concept is real. You wouldn't have all those big name companies apart of the alliance if it didn't work.

I don't know if there's any technical reason stopping GDDR-3/4 from getting the QBM treatment.

What big names? Is AMD shipping support for it? Is Intel shipping support for it? Is Nvidia shipping support for it? Is ATI shipping support for it?

looking at the QBM web page, I don't see any big names.

As far as technical reasons... Signalling speeds.

Aaron Spink
speaking for myself inc.

 

[London Geezer]

Maybe some of the HW gurus can help me with this.

If R500 is at 90nm, the PS3 GPU would need to be at ~ 45nm to have twice as many transistors on the same die size, right?

And at 65nm it would be what? 30-50% more transistors?

Or am I way off...

Hey this is elementary geometry we're talking about!!

Edit: Oh, Jaws answered...

 

[Guden Oden]

And at 65nm it would be what? 30-50% more transistors?

The actual question has already been answered, but then there's the issue of actual die size.

If we assume PS3's GPU uses a deeper process than nextbox (quite an assumption already), and we assume die size is exactly the same, then PS3 GPU will have more trannies. But we could just as well assume Sony wants a similar-class GPU as what ATi is designing for MS, so they'll end up having roughly same amount of trannies. PS3 die size will therefore be smaller.

Then again, looking at PS2, we know Sony isn't afraid of large chip dies, so it could be a monster chip too, who knows. We're assuming left and right with no solid foundation to base any of this on.

 

[London Geezer]

In the end it all comes down to cost, like i've said so many times before.
Sony will draw a line somewhere, when they think they have enough performance for the money they'll be spending, and anything more would be a luxury they might not be able to afford.

I don't think Sony should worry too much about being a lot more powerful than MS, they'll end up spending a lot of money for something that really doesn't affect the return of investment to them as much as GTA or FF do.

It will be valuable to them to have ze bigz numberz, but look at what happened with PS2, many people still think PS2 is more powerful than the competition.

Sony and everyone else better spend their money somewhere they know they'll make a profit out of. GTA. Pokemon. Halo. Those are the things that make the most money for them, people buy consoles based on what big game they can play on them, not because one GPU runs 500MHz faster than the other one.

 

[Brimstone]

The QBM concept is real. You wouldn't have all those big name companies apart of the alliance if it didn't work.

I don't know if there's any technical reason stopping GDDR-3/4 from getting the QBM treatment.

What big names? Is AMD shipping support for it? Is Intel shipping support for it? Is Nvidia shipping support for it? Is ATI shipping support for it?

looking at the QBM web page, I don't see any big names.

As far as technical reasons... Signalling speeds.

Aaron Spink
speaking for myself inc.

VIA and SIS are a among those listed as supporters and they're certainly chipset middleweights.

Kentron isn't seeking royalties on the patents they have, so I assume companies can customize the technologys as they please. However, assuming what a company would allow is dangerous though.

 

[hovz]

in the end it doesnt matter cuz sony is going to dominate next gen.

 

[aaronspink]

VIA and SIS are a among those listed as supporters and they're certainly chipset middleweights.

Kentron isn't seeking royalties on the patents they have, so I assume companies can customize the technologys as they please. However, assuming what a company would allow is dangerous though.

Neither VIA nor SIS have support for QBM in their chipsets. This is a technology that has so far been all hype with no products on the market.

But this is all irrelevent for the area of consoles and graphics chips since they use directly connected on board memory and would not benefit from any intermediate chip. If they wanted faster memory, they would order faster memory.

Aaron Spink
speaking for myself inc.