Goes without saying to take a few bags of salt with this, but if it turns out to be true would this spell the end for AMD GPU's ? http://videocardz.com/51009/nvidia-preparing-four-maxwell-gm204-skus
Spell the end for AMD GPUs? Errr nope. Not sure how you figured that one. It'll mean is that nvidia gpus built on the new process should be nicely power efficient amongst a few other things. It'll also mean they are rare, and expensive vs the performance they offer, for the first 1-6 months after release, unless by some miracle TSMC gets good yields right out of the gate.
tsmc will ship 300k 20nm wafers this year - BULK has only really just started - they are set for 1.2 million wafers next years. APPLE is hovering most of those wafers. as for 16nm - they haven't even SAMPLED , let alone RISK production - so its just a pie in the sky story as BULK wont start till end 2016 start of 2016 - if theres no issues , and TSMC doesn't have a stella track record
Ohh, SemiAccurate as a reference. The hint's in the name. Heh, now you have to subscribe to see the typically inane ramblings of Charlie D? No thanks.
Well i did say take it with a few bags of salt. @Parge, Spell the end of AMD because AFAIK they are still on 28nm that uses more power and produce high temperatures, from what i understand they are only going to drop to 20nm next time. If Nvidia gets 16nm parts into peoples hands while AMD are still on 28nm there will be no competition, no ? Surely a 16nm part is going to be head and shoulders better than a 28nm part.
Looking into this some more it seems TSMC are saying... And going on an article on kitguru.net... So does this mean even AMD may skip 20nm ? Going from 28nm to 16nm.
AMD are going 28nm at GloFO for tonga and I stand by what I said - apple have hoovered the 20nm from tsmc this year , and your own link shows TSMC haven't sampled 16nm yet, it does depends if they can actually get out of the gate first....
Will all depend on thermal efficiency, a big die on a mature process has its benefits, look at Intels core chips, they have problems getting the heat out on lower geometries, despite lower power designs. TSMC have put a few things through over the past year on 16nm so they have a handle on it which is probably why Nvidia are confident to skip 20nm.
Apple has pretty much aquired as much space in 20nm area as it can get its hands on in preparation for Iphone 6 launch this year. Apples budget for just the one part of its phone is bigger than AMD and nvidia combined. If they went to 16nm by 2016 it would change things that's for sure as the mid range cards would be quick whilst sipping power. 28nm has run its course that's for sure, 20nm has benefits but probably not as big as most of the stories I've seen. Talks of 50% pure performance gains are nuts. It might gain 50% in power to performance ratio. If the rumours did turn out to be true it would mean 20nm is even worse performer than most think it will be. Not certain how much I'd want AMD gpus to fail either as once the Gpu side goes then AMD as a whole will follow it. A world with just intel and nvidia in it is expensive.
@Harlequin, When you say "TSMC haven't sampled 16nm yet", forgiven me but what do you mean by sample ? Do you mean sending 16nm parts for testing with the likes of AMD, Nvidia ? What happens when the demand Apple is putting on the 20nm parts subsides ? In the Kitguru link it says... When is the IPhone 6 supposed to launch ? Won't the demand for 20nm parts fall of once it does. Very true, it's why i asked if AMD may also make the jump to 16nm parts, we need them to stay competitive and i think them moving to 16nm would defiantly make that so.
Apple makes 200mil + phones a year then tablets and iPods will go on 20nm after. Demand will never drop enough as when it does they launch the next model. Apple is basically buying fab space, enough I still personally think intel will just do it eventually as its a multi billion dollar contract for whoever gets it. AMDs fab space is been took up by ps4 and Xbox one.
I know rumors are nothing concrete, but Semiacurate is not the only site reporting that TSMC is going to have 16nm parts ready at the start of 2015. http://hexus.net/business/news/components/71929-tsmc-accelerates-development-10nm-process/
The GPU market always goes like this though doesn't it? One side releases their new products, and then the other side has to price theirs appropriately until they can release something equivalent.
I hope that research and development are going well regarding smaller fabs, different materials ect, as those foundries are getting awful close to the limit of what is currently possible with silicon.
several commentators are saying that basically Intel went FinFET too early , and @ 22nm its not of benefit (pointing to the latest haswell as proof). TSMC and GloFo could well be better off not going finfet at 20nm and waiting - which I think glofo will do. most new articles are all pointing back to SA in regards to the nv story. personally I see this as Nv trying to force TSMC to `sort there **** out` - TSMC 20nm process has problems - they are only offering 1 solution (not multiple eg high-k , LP etc) sample>risk>bulk ; sample means the CAD file as been sent *tape out* , for the first article on the new process. risk or early production - once the final tape out (no more spins needed) then production can start , at a low rate checking for yields , as low yields cost money bulk or volume production - the yields of working IC`s is at an accepted standard so prosuction can move for low rate to high. would love to know what UMC are doing in all of this ; also @ 10nm your getting noticeable quantum tunnelling
