Why NV30 beats 9700 Pro?

[legion88]

Originally posted by Jandar


reread that link...

8 Pixel Pipelines
1 Texture UNIT per Pipe
8 Textures PER Texture unit.

lot more than just 8 textures in a pass....

:shrugs:

What are you trying to say? The 9700 can do 64 textures?

The "8 textures per texture unit" (I saw that somewhere besides your post but don't remember where) looks to be an error.

It probably means "8 samples per texture unit" (a buzz phrase meaning it can do trilinear). That is an actual improvement in design since most graphics card requires two units to do trilinear. NVIDIA had this capability with the GeForce256 but went backwards with the GF2 when they went back to the two texture units per pipeline setup.

[Bigus Dickus]

Originally posted by legion88


What are you trying to say? The 9700 can do 64 textures?

The "8 textures per texture unit" (I saw that somewhere besides your post but don't remember where) looks to be an error.

It probably means "8 samples per texture unit" (a buzz phrase meaning it can do trilinear). That is an actual improvement in design since most graphics card requires two units to do trilinear. NVIDIA had this capability with the GeForce256 but went backwards with the GF2 when they went back to the two texture units per pipeline setup.

Or perhaps he was talking about loopbacks instead of multipass rendering? ;)

[Uttar]

Originally posted by PreservedSwine


Ever hear of backwards engineering? Not to mention TSMC will have that much more exprierience on the .13 fab process...

1. Backward engineering is illegal and nVidia most certainly got A LOT more lawyers than ATI. ATI would suffer so much from nVidia legal power they'd never get back on their feet again. That trick won't work here.

2. Remember nVidia is TSMC biggest fabless customer. If nVidia wants TSMC highest-yielding 0.13 factories, they'll have them directly. ATI might have problems if nVidia uses all TSMC 0.13 fabs ( which is unlikely, but who knows... )


Uttar

[ZenOps]

Whoa, hey there. Its not like ATi does not have a relationship with TSMC. From intial design to rollout it takes about 2-3 years.

ATi is independantly developing a .13 chip with TSMC. TSMC bows to noone, ATi, Nvidia, Lucent or Cisco, and many others, all of which use their fabrication plant.

The .13 micron fab is a process more than a patent. Chip design is a 100 percent patent.

The .13 process is not ready, its taking longer to figure out than expected.

Nvidia risked that the turkey would be cooked by now so to speak, but its not... And just like an oven, you can't just turn up the heat to see if you can speed up the process, it doesn't work that way.


Side note: Even though ATi supposedly made *a lot* of .15 micron chips in its first production run, they are already internally sold out of cards only after a few days.

http://mirror.ati.com/shopati/promotions/atigamer/index.html

At 1 percent production and low yields, Nvidia is probably cursing the tech who suggested they try moving to .13 micron early.

[Uttar]

Originally posted by ZenOps
Whoa, hey there. Its not like ATi does not have a relationship with TSMC.

I'm not saying that. I'm just saying that more important customers are more important. Or you want to say that's incorrect, maybe? :P


Uttar

[Bigus Dickus]

Originally posted by Uttar


1. Backward engineering is illegal... No, not really. It's perfectly legal to decompile code, take apart mechanical things, or look at a chip under a microscope (for whatever that's worth) and see how someone else made something. It's also perfectly fine to learn during the process.

What is illegal is copying or nearly copying particulars of the design that are protected under IP laws (patents etc.). Looking at an NV30 under a microscope and saying "hey, they used wider interconnects here than I would have thought" is perfectly fine, and you can use that knowledge towards your own chip design if you think it's useful, so long as NVIDIA hasn't obtained a patent for "increased interconnect design and application in .13u ASIC fabrication" that specifically covers their usage as a unique case. Whether they can patent such "process tweaks" is highly suspect. Sure, they can patent the actual logic, but that hardly defines the specifics of the fab process.

In short, if you are Ford and you notice that Chevrolet has started using aluminum 6061 for chassis (you haven't made it that far in engineering, and are still using steel), it's perfectly legal to examine their chassis to see how they handled welds, stampings, coatings, etc. to get a handle on the problems and benefits of using aluminum vs. steel.