News - Apple looks set to choose TSMC over Samsung for A7 chips

[3dcandy]

I really doubt Samsung would be too bothered by Apple turning to TSMC, historically they've been a tough company to work with long term as they shift their allegiances far too easily. This will impact AMD more than anyone else I feel, they will need to go to GloFlo for more products because effectively they will be 3rd in line at TSMC

[keithwalton]

Didn't they try to move production with the A6 to TSMC's 28nm process and TSMC basically told them where to stick there capacity demands as basically TSMC already had more orders than production capacity and apple wanted sole use of over half the plant.
If they do move over to TSMC expect the A7 to be late, run hot and consume too much power and have extremely limited avaliability. AKA the TSMC way

[watercooled]

TSMC refused to dedicate capacity to Apple as it would risk damaging relationships with other companies. IIRC it was Qualcomm who asked for a similar deal and were also rebuffed.

It's unfair to blame those problems on TSMC - they have plenty of huge customers, and the likes of Xilinx and Altera who make some HUGE dies (FPGA logic isn't directly comparable to GPU but still). AMD also make GPUs and don't seem to have some of the problems Nvidia blame solely on TSMC (presumably that's what you're referring to).

[Biscuit]

TSMC refused to dedicate capacity to Apple as it would risk damaging relationships with other companies. IIRC it was Qualcomm who asked for a similar deal and were also rebuffed.

It's unfair to blame those problems on TSMC - they have plenty of huge customers, and the likes of Xilinx and Altera who make some HUGE dies (FPGA logic isn't directly comparable to GPU but still). AMD also make GPUs and don't seem to have some of the problems Nvidia blame solely on TSMC (presumably that's what you're referring to).

Well either that or they dont make as much fuss about it

[watercooled]

Developing new process nodes isn't easy to do, so it's reasonable to expect some teething problems and a production ramp i.e. they can't just switch on 28nm and immediately have full capacity with perfect yields. Other companies had huge dies out at 28nm before Nvidia with reasonable capacity, Nvidia kept their 28nm GPU dies relatively small initially but still complained about TSMC, as usual - of course it's never Nvidia's fault.

There's more to fabbing than producing some logic and pressing print at the foundry; yield can also be heavily impacted by the logic design. Essentially what I'm saying is, Nvidia combined with TSMC seem to have problems, problems that other big customers clearly don't have to the same scale - it's not just about how much they complain about it, Nvidia's problems have been evident anyway in the form of low availability, late arrival, poor efficiency etc. Although the problems at 28nm seem less severe than some previous nodes. Again, smaller initial dies probably helped somewhat.

[kalniel]

Well AMD also suffered from availability issues with the last bunch of GPU chips, but were rather more taciturn in their press releases.

[3dcandy]

Other reports are quoted as saying that Samsung are privately quite pleased with this because they can put more capacity aside for their Exynos chips which they have been forced to not use in a few products because of the vast numbers of A5 and A6 chips they make. Seems the S4 has to use Snapdragons in some variants for example whilst some early benchies say the Exynos is faster in almost every way.
They did increase prices by just over 20% recently and I'm guessing it was a calculated move to distance them from Apple.

[watercooled]

Well AMD also suffered from availability issues with the last bunch of GPU chips, but were rather more taciturn in their press releases.

Oh I don't dispute that, or the fact that initial availability can be, and often is, below what customers would like. I'm just saying I don't think it's fair to jump on the Nvidia-led 'blame TSMC for everything' bandwagon.

[3dcandy]

Oh I don't dispute that, or the fact that initial availability can be, and often is, below what customers would like. I'm just saying I don't think it's fair to jump on the Nvidia-led 'blame TSMC for everything' bandwagon.

I just feel both of you are right, but nVidia are always more vocal in blaming others

[mercyground]

Serves Apple right. Also anyone who DIDNT see Samsung going to put the boot in has rose tinted glasses on.

[mercyground]

Nv's ****ty yields are down to their inability to follow process technology. NV are the ONLY company still having major issues with low yields. Come their next financial quarter it will make for VERY interesting reading on how much they have blown as they will have to pay for the wafers now... not TSMC.

[kalniel]

Nv's yields are down to their inability to follow process technology. NV are the ONLY company still having major issues with low yields. Come their next financial quarter it will make for VERY interesting reading on how much they have blown as they will have to pay for the wafers now... not TSMC.

So conversely, if the financials are good, you'll be revising your opinion of them right?

[HalloweenJack]

a quick google search is showing both AMD and Nv have commented about tsmc and 20nm wafer costs.... I do suspect then neither will `make the jump` early ; maybe 1 reason AMD have already said HD8xxx will be refresh?

[watercooled]

I thought that was more about Sea Islands, which a lot of people assumed would be released under the HD8000 brand? AFAIK Nvidia also have a 28nm refresh to go but I thought they were both still aiming for their 20nm GPUs after the refreshes? But yeah they might want yields to stabilise before they make the jump...