SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Technology Stocks : AMD, ARMH, INTC, NVDA -- Ignore unavailable to you. Want to Upgrade?

To: Vattila who wrote (16110)6/2/2016 12:44:04 PM
From: rzborusaRead Replies (1) | Respond to of 72371
 
According to the info I can find with a quick search, the GP104 die in GeForce GX 1080 (and presumably 1070) is estimated (by videocardz.com) to be ~333 mm², while the Polaris 10 die in Radeon RX 480 is estimated to be ~234 mm². Can anyone speak to the yield and cost advantage of Polaris, assuming equivalent process?
My simple layman understanding of this: There will be a number of defects per wafer. These will be distributed over the number of die. To compare effects for different die sizes go directly to the die count per wafer. The one benefit of larger die is one die is more likely to absorb multiple defects. But, it is all about averages. HTH Oh, there is 333/234=1.42. And it probably would be even higher than 1.42 given the edge loss, round wafer and rectangular die. So my wag would be almost 50% more candidate parts. I welcome correction.



To: Vattila who wrote (16110)6/2/2016 1:59:46 PM
From: neolibRespond to of 72371
 
I'm guessing 300mm2 at TSMC yield better than 200mm2 at GF.



To: Vattila who wrote (16110)6/2/2016 5:34:20 PM
From: THE WATSONYOUTHRespond to of 72371
 
According to the info I can find with a quick search, the GP104 die in GeForce GX 1080 (and presumably 1070) is estimated (by videocardz.com) to be ~333 mm², while the Polaris 10 die in Radeon RX 480 is estimated to be ~234 mm². Can anyone speak to the yield and cost advantage of Polaris, assuming equivalent process?

.....I'll take a crude stab at it........assuming equal defect density, (around .2D/cm^2), I'd guess maybe a 10%-15% difference in yield (say 67% vs 55%) for the two die sizes.........the bigger cost effect is the number of potential die per wafer. I'd guess the larger die (with a bit more edge loss) at perhaps 190 die/wafer and the smaller die at perhaps 270 die/wafer. .....so, together with the yield delta, the smaller die probably costs about 58% of what the larger die costs. So for a cost per wafer of $5K..........the smaller die would cost about $28 and maybe $48 for the large die... What does NVIDIA charge for their board
(GeForce GX 1080) with the 333mm^2 die? Of course the defect densities in the two fabs could be somewhat different and I'd expect GF to charge less than TSMC. But certainly, the smaller die is a BIG advantage to AMD.

.....there's a yield vs (defect density x Area) for various models here:

smithsonianchips.si.edu

.....note a 100mm^2 die would yield above 85%


HomeMailHotSubjectMarksPeopleMarksKeepersSettings
Terms Of UseContact UsCopyright/IP PolicyPrivacy PolicyAbout UsFAQAdvertise on SI
© 2025 Knight Sac Media.  Data provided by Twelve Data, Alpha Vantage, and CityFALCON News