Fermi yields were bad but they had a pricing agreement with TSMC, they only paid for functional chips, and not per wafer, which only hurts TSMC's bottom line and not NVs. This is the critical difference now.
Typically a 40nm wafer costs ~$5,000, probably much more on 28nm. Actually, not probably but definitely given the news surrounding TSMC's 28nm production.
I guess they can squeeze ~160 gk104 onto a wafer given its die size estimates. Raw die cost are only a fraction of the final product costs though.
Edit: whatever the case, i'm delaying gpu upgrades until i see gk104 benchmarks. With something concrete to work with, one could speculate on the big kepler's potential performance as well.
Charlie at SA stated they paid by the wafer cost in his infamous article :TSMC sells chips in two ways: you can purchase the whole wafer, in which the cost per die can get as low as possible - but you're responsible for the yields in question. Second mode is purchasing dies alone, i.e. you pay much more per single die but ultimately - you don't care what the yield is. Most of TSMC customers chose either one or another but with clients as big as AMD and nVidia are - those pricing models change on product range. However, in terms of die size and overall number wafers, nobody comes close to these two. Our sources claim that in case of GPU dies such as Cypress and GF100, both AMD and nVidia are ordering per wafer and they're paying around $5000 per single wafer manufactured on 40nm Performance node.
Even if Nvidia beats the initial production targets by ten times, its yields are still in the single digit range. At $5,000 per wafer, 10 good dies per wafer, with good being a very relative term, that puts cost at around $500 per chip, over ten times ATI’s cost. The BoM cost for a GTX480 is more than the retail price of an ATI HD5970, a card that will slap it silly in the benchmarks. At these prices, even the workstation and compute cards start to have their margins squeezed.
Excellent post IDC, I learned a bunch. However, I'd like to point out that it may in fact be a combined issue (both functional and parametric) as numerous TSMC customers are having difficulties, not just NVIDIA: http://www.eetimes.com/electronics-news/4234961/TSMC-manufacturing-process-in-trouble
Obviously, AMD was able to get 79xx and 77xx parts out the door, so they're suffering less if at all, but unless they're that 1 of 7, I think both companies are getting hit hard and the consumer is going to pay for it.
??????Wow spin city is working hard tonight.
Fermi yields were bad but they had a pricing agreement with TSMC, they only paid for functional chips, and not per wafer, which only hurts TSMC's bottom line and not NVs. This is the critical difference now.
Do you have a citation for the wafer purchase agreement.
Both. Simplified, AMD is going to price the 7970 as high as they can to reduce the demand to meet their limited supply. I've said for months now, until TSMC gets its ducks in order, there will be no price war unless either company is willing to take a big hit on the bottom line.So, this begs to ask - are the current pricing issues on the HD 7K series really a money grab by AMD or a production issue causing their own prices to increase?
So much for a price war I'd still hope to see GTX 580 like performance in the $350 price range. Hopefully.
Remember how well AMD navigated the design-for-manufacturing aspects of 40nm (double vias) in comparison to Nvidia ("we need zero via defects!").
Yeah, reading between the lines of JHH's comments it sure sounds like NV is still having problems with their implementations. From what I've read Tegra3 is bloated for an SOC. NV needs to get there head around this problem, because it is only going to become more difficult on smaller nodes.
Speaking of : Along with Ice Cream Sandwich and a 4.7-inch high-definition screen, LG's high-end flagship phone will also feature a Tegra 3 chipTegra 3 is a quadcore ARM CPU with a die size of 80mm squared. For comparison purposes, the ipad 2's cpu, A5, is a dual core CPU with a die size of 120mm squared. Nvidia manufactured Tegra3 on 40nm instead of 28 because 1) it was a known quantity and 2) it allowed them to be the first company in the world with a quad core ARM cpu.
....The GPU in the iPad 2 owns the Tegra 3....Tegra 3 is a quadcore ARM CPU with a die size of 80mm squared. For comparison purposes, the ipad 2's cpu, A5, is a dual core CPU with a die size of 120mm squared. Nvidia manufactured Tegra3 on 40nm instead of 28 because 1) it was a known quantity and 2) it allowed them to be the first company in the world with a quad core ARM cpu.
<speculation> With NV's operational expenses increasing beyond expectations, I'm inclined to believe that there will be a Kepler refresh b/4 Maxwell and that would square up with Maxwell's debut being push out as well. </speculation>
....The GPU in the iPad 2 owns the Tegra 3....
Also despite being a quadcore, the CPU performance between the A5 and Tegra 3 is really close too.
Well the truth is of course that it is always a combined issue.
Within the same market space though, AMD has not cited 28nm yields as a concern. That would suggest whatever issues NV is having they are due to the aspects of the chips which are not shared (die-size and chip design/layout).
Remember how well AMD navigated the design-for-manufacturing aspects of 40nm (double vias) in comparison to Nvidia ("we need zero via defects!").
If NV has a parametric issue then I would expect it to be one of power-usage versus clockspeed. Same as was the case for Fermi. Functional yields aside (they harvested their 480's), just getting them to clock high enough without having TDP issues was a big problem. That's all parametric yield.
Hopefully Kepler is not having those kinds of issues, but yeah we'll never know for sure. Just guessing from outside.
The A5 also uses a third party GPU core from Lucid and is manufactured on Samsung's fabs. So it's not a direct apples to apples comparison with the Tegra which is fully designed in house and uses TSMC I belive.
AMD might not be bragging that the yields are poor but the price on the 7900 series might be a direct consequence.
However, there are additional hurdles Nvidia has to take. Huang noted that the hard disk drive supply issues remain and the shortage of 28-nanometer wafers as well as lower than expected production yields continue, which negatively impacts the company's core GPU business. It does not seem to be an issue that can be resolved anytime soon as Huang explained that "the amount of 28-nanometer capacity in the world is not enough." He expects 28 nm shortages to persist throughout this year.
so according to tomshardware it was true. and i think it will be more than 6 months before we can expect it to launch
at least AMD doing right by not pricing their flagship GPU to low.
Again, just excellent info; this is very insightful. It would seem to me then that NVIDIA will have difficulties in both yield arenas considering their approach. If functional yields are low for everyone because TSMC's 28nm process isn't doing so well, that's bad. Couple that simply with a physically bigger chip, I imagine yield plummets. If they are facing a parametric issue like they did with Fermi, then good luck seeing anything until there's been several respins.Well the truth is of course that it is always a combined issue.
Within the same market space though, AMD has not cited 28nm yields as a concern. That would suggest whatever issues NV is having they are due to the aspects of the chips which are not shared (die-size and chip design/layout).
Remember how well AMD navigated the design-for-manufacturing aspects of 40nm (double vias) in comparison to Nvidia ("we need zero via defects!").
If NV has a parametric issue then I would expect it to be one of power-usage versus clockspeed. Same as was the case for Fermi. Functional yields aside (they harvested their 480's), just getting them to clock high enough without having TDP issues was a big problem. That's all parametric yield.
Hopefully Kepler is not having those kinds of issues, but yeah we'll never know for sure. Just guessing from outside.