EETimes reports on reasons behind high initial 360 failure rate

[AzBat]

I put "recall" in quotes because technically it's not a recall. Yes, it's as expensive as a recall, but no specific issue was addressed and they didn't require every unit to be sent in to be replaced. Anyway, EETimes are reporting that while Bryan Lewis, of Gartner Research, was at the Design Automation Conference this week, he reasoned that the problems with the 360 were due to Microsoft wanting to "avoid an ASIC vendor" by designing the graphics chip on their own and sending it directly to TSMC. Lewis then said that Microsoft went back to a unnamed ASIC vendor based in the US and redesigned the chip. EETimes believes that company is ATI, based on a report last year they made about TSMC fabbing their own 90nm version of the eDRAM chip.

Lewis further added: "Had Microsoft left the graphics processor design to an ASIC vendor in the first place, would they have been able to avoid this problem? Probably. The ASIC vendor could have been able to design a graphics processor that dissipates much less power."

Not alot of new info, but it basically confirms some suspicions about the GPU being the issue. I just wonder if Microsoft really had more of role in the design process than we thought? Chalk it up to a $1billion lessoned learned?

Source: EETimes

Tommy McClain

 

[Mintmaster]

That info seems a bit odd. ATI always was heavily involved with the GPU design. Moreover, TSMC only acquired the ability to do eDRAM recently, so changing fabs isn't really evidence of what they're claiming it to be.

 

[Mize]

What "recall"? RROD fix once busted?

 

[AzBat]

"Recall" is per the EETimes article and yes they're referencing the $1billion RROD warranty extension. As for the claim, I believe the change in fabs is EETimes proof that ATI is the unnamed ASIC vendor hired to fix the issue. I don't get the connection either.

Tommy McClain

 

[Zaphod]

Seems to me like the article is laying out this timeline:

1. MS: Hey, ATI! Could you sell us some IP for our next Xbox? We're going to lay out the chip ourselves 'cause we're the roxxorzz!
2. ATI: Sure.
3. MS: Psst, AMD! Remember that IP you sold us? It turns out we're not quite as roxxorzz as we thought we were... Could you have a look at our design and help us fix this thing for a respin?

Doesn't make much sense, though.

 

[thambos]

With Falcon besides the CPU shrinking to 65nm, I remember the GPU being slighty smaller. People speculated it might have moved to 80nm.

Is it possible the GPU was just redesigned to be more efficient? Or is the 80nm shrinkage more realistic?

source:
http://www.anandtech.com/showdoc.aspx?i=3152&p=5

Microsoft never released die sizes for the Xbox's chips so we had to break out the trusty ruler and measure both the old and new chips, to get an idea of how things have changed. The table below shows our findings:

 Chip 	Old Die Size 	New Die Size 	% of Old Die
CPU 	176 mm^2 	133 mm ^2 	75.5%
GPU 	182 mm^2 	156 mm^2 	85.7%
eDRAM 	80 mm^2 	68 mm^2 	85%

....It's possible that all three chips are now 65nm, or a mixture of 65nm and 80nm (TSMC's 80nm half-node process was used in ATI's R600 GPU). Needless to say, the chips are all smaller, which should yield some nice power savings.

 

[3dcgi]

Is it possible the GPU was just redesigned to be more efficient?

Yes.

Or is the 80nm shrinkage more realistic?

No.

 

[thambos]

Yes.


No.

So then we should expect around a 118 mm^2 GPU for the Jasper model which is supposed to arrive the first week of Sept.

Hope Anandtech does another disassembly/analysis so we can have a definitive answer to the EETimes article.