Polaris both on 16 and 14nm?

Stoly

Supreme [H]ardness
Joined
Jul 26, 2005
Messages
6,713
Polaris has just been announced and its quite interesting but something got my attention

From PCPERSPECTIVE

On slide 4 FinFet Excitement there is no mention of the manufacturing process compared to previous years.

polaris-7.jpg


On the next slide it reads FinFet Enabling the drive to 16nm and beyond.

polaris-10.jpg


The article states that the first Polaris chip was built on 14nm GloFo but the slide seems to indicate it's 16nm.

So maybe Polaris will be built on 2 different processes by 3 different manufacturers. TSMC, Samsung and GloFo.

Quite intriging.
 
According to Anandtech, some GPU's will be on 14nm GloFo, and some will be on 16nm TSMC. They said that they are not dual sourcing, but rather having some designs go to one and some designs go to the other,

Now, this means they have to lay out some of the common logic blocks twice, which ultimately will increase costs. It has been suggested that part of this is because of the wafer purchase agreements that AMD has with GloFo, and sending some GPU's their way will help AMD fulfill those agreements. I am thinking that this is probably a large part of why they are splitting manufacturing as such.
 
Right, since AMD couldn't hand-off Global Foundries without promising a certain amount of wafer production, and since their CPU volume has tanked, they either have to cough-up the penalty payments direct to GF, or they have to move GPU production over.

Guess it's cheaper to pay for the overhead of porting their logic to a second fab than it is the pay those penalties? :D

At least since last year GF has had a bulk 28nm silicon process available, so they can make the porting of designs much easier.
 
um, last i heard some will be on glofo, some will be on Samsung.
 
Hmmm ... top-end 14nm on GloFo and rest on 16nm TSMC/Samsung? Getting more curious by the day.
 
hmm top end at TSMC and mid and low end at Samsung and GF, since Samsung and GF are the same process and the the low end Polaris is on 14nm GF sounds more reasonable.
 
I just see a bunch of [H]ardocp users removing their heatsinks to figure out if they lucked out or not in the GPU lottery. :)
 
hmm top end at TSMC and mid and low end at Samsung and GF, since Samsung and GF are the same process and the the low end Polaris is on 14nm GF sounds more reasonable.

That's also a reasonable speculation; my take is that maybe AMD is tired of TSMC's delays, but who can say? Should make for a pretty interesting year in video cards come 2nd quarter w/NVIDIA in the mix ...
 
That's also a reasonable speculation; my take is that maybe AMD is tired of TSMC's delays, but who can say? Should make for a pretty interesting year in video cards come 2nd quarter w/NVIDIA in the mix ...

Glofo doesn't have a good record either, probably worse than TSMC. Hopefully the deal with samsung will help.

I'm willing to bet the high end will be on TSMC as they have more experience with large dies.

I don't think there's much difference if any between 16 and 14nm anyway
 
I just see a bunch of [H]ardocp users removing their heatsinks to figure out if they lucked out or not in the GPU lottery. :)

No, they are not dual sourcing. So each GPU will only be made at one fab. There won't be a lottery.
 
No, it is TSMC 16nm and GloFo 14nm (which is a licensed copy-exact of samsung's 14nm process)

What he is referring to is there was an article about AMD using BOTH GloFo and Samsung foundries. It wasn't mentioned using TSMC with that release.
 
I am looking forward to some half height SFF cards with this.
 
That's also a reasonable speculation; my take is that maybe AMD is tired of TSMC's delays, but who can say? Should make for a pretty interesting year in video cards come 2nd quarter w/NVIDIA in the mix ...


i'm kind of in this same sentiment as well.. AMD and nvidia have been getting screwed for years with TSMC promising things and never actually delivering on time or outright canceling process manufacturing at basically the last minute. luckily nvidia has the money and man power to rework their designs but AMD doesn't. biggest two i remember was the failure to deliver 40nm on time for the 5k series and canceling i believe 32nm(some one correct me if i'm wrong). for the 6k series.

i'm hopeful though with samsung and glofo in the picture it can only get better so we'll see. this will be the best test run to prove if they can compete against TSMC.
 
Yep HD 6000 was originally supposed to be on 32nm HKMG, but of course TSMC fucked it up.

Actually if you really start to dig deeper, TSMC hasn't delivered a single node on time or without issues starting with 40nm.
 
Glofo doesn't have a good record either, probably worse than TSMC. Hopefully the deal with samsung will help.

I'm willing to bet the high end will be on TSMC as they have more experience with large dies.

I don't think there's much difference if any between 16 and 14nm anyway

Samsung doesn't have any capability right now to deal with GPU, and their 14nm is stolen from TSMC's design, which they couldn't get the yield better than TSMC's 16nm process.
Apple's A9 chips is suffering from this problem too.

GF's 14nm is also the same as Samsung.
 
Samsung doesn't have any capability right now to deal with GPU, and their 14nm is stolen from TSMC's design, which they couldn't get the yield better than TSMC's 16nm process.
Apple's A9 chips is suffering from this problem too.

GF's 14nm is also the same as Samsung.

TSMC does't have and won't have a 14nm mode, how could Samsung steal it? :rolleyes::rolleyes:
 
Back
Top