Another interesting bullet point is how they say TSMC has, once again, solved their 40nm manufacturing problems and production has ramped up. They also say that TSMC has new machines since September and that production is now starting to come online. The new machines are probably a major reason for the production of 5800's to go from a few thousand a week to tens of thousands per week.
I'm reading the HardOCP article but I am not seeing anything stating yields have improved. What am I missing?
Only thing I've read about TSMC's 40nm yields this week are these:
Quote:
TSMC 40nm yields are at 50 percent
Industry sources have confirmed Fudzilla that TSMC’s 40nm process is definitely far from the numbers they should be hitting. It’s been more than a year, probably even year and a half since TSMC started 40nm test production, and even today their yields are extremely poor.
Silicon foundry giant Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) claims that its 40-nm yields will improve by early 2010.
In response to the reports, TSMC officials said: ''The 40-nm yield didn't drop as reported. As a matter of fact, yield on the 40-nm process remained flat. TSMC is confident that the 40-nm yield will improve at the beginning of next year.''
Quote:
Originally Posted by BTRY B 529th FA BN View Post
bout 10yrs from now, we'll be reading how nvidia settles out of court to AMD/ATI for paying off TSMC to hinder 58xx sales
guy1: per...
guy2: ice...
guy3: less!
edit: sigh, i can't quote worth dookie with this vb thing yet
__________________
Rig1: Q9650 3Ghz, 4g, Asus P5Q Prem, Revo 950w, 1900XTX Vraptor 300g email
Rig2: i7 920 D0, ASRock Extreme, Mushkin 1333 1.5-1.6v @ CL6, Revo 950w, 4870, 300g 15k SCSI Gaming
Rig3: i7 920 D0, EVGA 760 Classified, 6g DDR3 2000 CL8 Soon to be gaming but currently still building
Retired: Q9550, 4g, UD3R F9 - possible hackintosh Heat 15,0,0
Last edited by BTRY B 529th FA BN; 11-13-2009 at 07:49 PM.
These two articles suggest the yields are fixed and it is another issue affecting production.
Sometimes I get my info mixed up or it might be missing cause I might read about something and not post it a few days later. Either way, the 40nm issues are affecting both nVidia and ATI.
These two articles suggest the yields are fixed and it is another issue affecting production.
Sometimes I get my info mixed up or it might be missing cause I might read about something and not post it a few days later. Either way, the 40nm issues are affecting both nVidia and ATI.
The xbitlabs link is a rehash of the EETimes article I linked above, and your EETimes Asia article requires login to view for some reason so I'm guessing its a bad link but probably says the same thing as my EETimes links.
That said, I believe you are getting the semantics mixed up here. TSMC is saying yield never dropped, which isn't the same thing as saying the currently dismal low yield numbers have been resolved and now yields are above the 80% entitlement levels.
Everything I have read so far speaks to low 40nm yields, no one is saying they have deteriorated per say, but everyone is saying they aren't where they were promised to be by TSMC going back months ago.
What perked my ears up when I read your post was that you said HardOCP is claiming the yields problems are solved which is a word I can't find anybody using outside your post, not in your HardOCP link nor any of the EETimes links. I'm not trying to come down on you about this, I was really hoping good news was out there and I had missed it.
Suffice to say I'm not seeing any good news here. Am I wrong?
I think we'll be seeing a light at the end of the tunnel soon according to this HardOCP article. At least as far as 5800 series card availability.
Looking at the bullet points, it's surprising how little cards have shipped. The good thing is TSMC seems to have gotten their crap together and ATI expects "tens of thousands" of cards to ship per week in late November. I think we'll see most of the pent up demand alleviated in early December. But for now it seems you guys looking for cards will still have to do some scrounging.
Another interesting bullet point is how they say TSMC has, once again, solved their 40nm manufacturing problems and production has ramped up. They also say that TSMC has new machines since September and that production is now starting to come online. The new machines are probably a major reason for the production of 5800's to go from a few thousand a week to tens of thousands per week.
So this bodes well for anyone looking for 5800 cards next month but you still seem to be SOL if you're looking for one in the next 2-3 weeks.
For ATI, this is not the best situation to be in, being supply constrained, when you have your competitor in a "Johnny come lately" status but at least they will still have some time before nVidia comes out with Fermi.
tsmc's yield problems should also end up delaying fermi. nvidia can't trouble shoot if tsmc can't reliably build their chips.
Not to derail the thread about TSMC... but the Intel - AMD settlement will likely open up the field for nVidia to use Global Foundries in the future. AMD can now go completely fabless and does not have to own a majority stake in Global Foundries to produce x86 chips there.
Time will tell if AMD holds on to their interest in Global Foundries or if they dump the stock.
I don't see how yield problems would affect debugging. It's a huge problem for satiating a hungry market, but when your "market" is your tech department, I don't see it.
If someone here knows better, now's the best time to chime in.
Yields aren't a big deal for design verification/validation provided obviously that the yield isn't zero and the design guys can get back enough samples to populate their testers and get things going.
That said there really is a crazy amount of things that need to be tested and verified with new designs, not just the question of does the math work properly (does 1+1=2) but also the lifetime stability testing as well.
So yes in a sample shortage situation after tapeout if yields are low then the scale and scope of the validation work is pruned back per priorities but that comes with the penalty of lengthening the verification timeline as they have fewer samples to work with while waiting to get more samples back from the fab.
In a well run fab this is transparent to the customer (be it external like NV/TSMC or internal like Intel) as the fab's resource planning and wafer start team factors in all the needs of the customer as well as risks that could jeopardize the fab's ability to meet those needs prior to determining how many wafers are to be started and allocated to the particular device in question.
What perked my ears up when I read your post was that you said HardOCP is claiming the yields problems are solved which is a word I can't find anybody using outside your post, not in your HardOCP link nor any of the EETimes links. I'm not trying to come down on you about this, I was really hoping good news was out there and I had missed it.
I wrote the HardOCP article and I don't remember reporting any yield issues being solved.
