Why is nvidia backtracking in die size?

[Slaimus]

The original GF3 was .13u, but the new GF3Ti's and the NV17M are both .15u
I did not heard of bad enough yield problems to cause them to actually backtrack. Is TSMC having yield problems now or something?

 

[Rand]

Their not backtracking, the original GF3 was made on the same .15u (150nm) process that the Titanium series is currently being fabbed on. TSMC was in no way capable of providing sufficient volumes of 0.13u chips upon the initial launch of the GF3, even were they able to Transmeta would have had first dibs on that for their TM5500 and TM5800.

TSMC has been having issues with their 130nm process technology, many manufacturers among which nVidia, Broadcom, VIA, and Transmeta have been very upset with the problems TSMC is having implementing its CVD-based system for their 130nm process.
nVidia and TSMC havent been getting along to well of late after nVidia's decision to go with UMC as an alternative supplier for their nForce and XBox related chip development.

 

[Finality]

What makes you think the original GeForce3 was 0.13 micron? It was more like 0.15 micron.

FYI 8 months ago no one had a decent 0.13 micron process. Even today TSMC has major yield problems with 0.13 micron. Intel themselves are just moving to 0.13.

UMC got the new contract for 0.13 micron parts (for the time being at least) but seeing as how expensive it is starting a foundry its probably at least a mid-term contract.

Edited its actually 0.15 micron not 0.18.

 

[Rand]

<< What makes you think the original GeForce3 was 0.13 micron? It was more like 0.18 micron. >>



It was made on the same .15u process the current Titanium chips are fabbing on, though undoubtedly the process had been tweaked somewhat since then, likely correlating with stepping changes the GF3 has undergone.



<< FYI 8 months ago no one had a decent 0.13 micron process. Even today TSMC has major yield problems with 0.13 micron. Intel themselves are just moving to 0.13.
>>



Intel has been shipping .13u processors in the mobile arena and their desktop Tualatin for 4-5 months now, I'd hardly call that "just moving to .13u"
The mainstream desktop Pentium 4 line is their last product to transfer to said process, and even it has been fabbing in .13u for awhile now, and is approximately a month from shipping in the Northwood core.

That said, your most definitely right in that no one could provide fab capacity and adequate yields on a .13u to sip the GF3 chips in the volume required upon it's initial launch.

 

[Finality]

<< It was made on the same .15u process the current Titanium chips are fabbing on, though undoubtedly the process had been tweaked somewhat since then, likely correlating with stepping changes the GF3 has undergone. >>

Your right I edited that 5 minutes before you posted



<< Intel has been shipping .13u processors in the mobile arena and their desktop Tualatin for 4-5 months now, I'd hardly call that "just moving to .13u" >>

Doh of course they where Still doesn't make it 8 months though.

However looking back if they shipped the GF3 around April that means the chips would have to be in production (full swing mind you) around February or so. With sampling going on in January? So its more like 10-12 months.



<< The mainstream desktop Pentium 4 line is their last product to transfer to said process, and even it has been fabbing in .13u for awhile now, and is approximately a month from shipping in the Northwood core. >>

Is the complete shift to Northwood or are they planning on leaving some 0.18 micron stragglers behind?

I'm guessing they still have legacy plants though for items like there networking products etc? Or is it more economical to move everything to 0.13 mircon?

 

[FishTankX]

I think the older processes do better with higher voltage (PCI anyone?) And the money they would spend on moving to .13U would have to be *less* than they're eventual savings (more dies on the same wafer) and aren't networking chips dirt cheap anyways? Wouldn't make *that* much difference if you were taking the board into account. I'm sure that once .18 matures nuff they'll move they're networking over to .18 but here's a question for ya. If process was so important why wouldn't they have moved to .18 with the 850 chipset? That's still .25. I'm sure what intel does is just switch to a new process when it's convenient. But with networking and low profit chips the benefit might be too little to justify the cost of the upgrade. I betchaya it would take 3 years to make back the money they would spend on migrating to a new process in they're networking process then have all these headaches with yields and redesigning all this crap. I think the most critical market for die shirnks is CPU's and Flash, really... Flash is AMD's domain though. Flash and transistors. AMD rules that arena.