AMD Going Straight to 14nm

Just thought I would share this with you guys.  What do you think about this?  If AMD can pull it off, they will be on the SAME NODE as INTEL.  Intel said that 10nm is still a long ways off and that would mean that both Intel and AMD will both be on the same node creating more of an even playing field.  The only thing that I am worried about competition wise is that Intel will have had a long time to mature their chips on 14nm, where AMD is going straight from 32/28nm all the way to 14nm.  One other thing I'm also interested in is the FM3 socket.  This socket will probably be the socket that supports project skybridge, so it would have to be compatible with x86 and ARM.  Now here is where I am concerned.  I honestly know just about nothing with actual chip engineering, but I would assume that all chips for this platform would have to be SoC's to stay pin compatible, correct?  How will that effect the performance of the more enthusiast chips?  Will they have to be bigger chips than current fx chips, more like the x99 ones, and if they are that seems like it would be a waste of space for the lower end chips that won't use near that much space?  Hopefully the more educated will be able to understand what they are planning to do.  One thing is certain here though, and that is that I am excited to see if they can reclaim some of the OG Athlon glory.

Article -

I hope AMD pulls back , the intel chips are the go to choice for gamers ATM.

Zen may prove to be problematic for AMD simply for the fact that they might be biting off more than they can chew.

Going straight to 14nm may prove to be very unwise, Intel had a lot of problems with the 14nm node and I doubt AMD's transition will anywhere close to smooth

I also want to babble about this unified socket thing. From what I understand its supposed to accept enterprise, enthusiast, mainstream (APUs), and ARM parts this is a monumental task that will most likely require a very large pin out (because of this and enterprise support expect FM3 to be LGA), a lot of very bizarre bus routing (and expect the chipset to be large as well), and expect the footprint of the processor to be around the same size 2011. 

I hope AMD can pull this off but I would keep the hype down until it gets closer to launch

1 Like

I'm wondering if we'll be seeing an 8 core 140-220w APU system, preferably at 14nm, AMD's HSA tech looks very good it'll be interesting to see how it pans out.

If they can somehow avoid intel's problems it could work for them. I may be planning a haswell e build but I still want amd to be strong. Heck if they had new cpus out this year I'd probably give those a look.

Well Zen won't come out until Q3 2016 so really it is going to have to be a wait and see. I think they should jump to 14nm if they can do it. AMD really needs to pull a rabbit out of a hat if they truly want to defeat Intel. as they been holding the crown for long enough. Just hope AMD doesn't go over their heads.

Glad they are going back to SMT rather than CMT that ruined bulldozer.

im very excited in what amd´s new zen architecture will bring to the table.

But indeed its a matter of waiting and waiting.

Amd doesn't have a choice the one area AMD has an advantage on Intel (integrate graphics) is getting closer and closer to losing the edge. It's either this or AMD dies.

Carrizo will likely be based on 14nm so they have time to work out any problems with zen chip. You have to remember that Intel isn't their only competition. They have to compete with nvidia also which is the point of moving to 14nm for the apu.

No. Carrizo will be 28nm. 

Thanks for the correction

What's actually interesting is when you compare the 32nm Vishera parts to their 32nm Sandybridge counterparts. Obviously Intel had the wins in single threading and lightly threaded workloads, but AMD won a lot of multithreaded tasks. There wasn't some huge performance gap between the two (Outside of some games which are very single threaded).

Also remember, AMD doesn't actually manufacture its parts anymore. So, they wouldn't have to come up with a manufacturing process to make all of these parts themselves like Intel had to. This is great because it helps them overcome Intel's historical manufacturing edge.

I have a feeling AMD will pull this off reasonably well, they've pulled rabbits out of their asses before, and sincerely hope they do for the sake of the entire market.

Maybe if we sacrifice half the world population to the gods they will have working graphene chips before IBM.

I would never bet against IBM, that's just stupid :P

Since they don't make consumer products anymore, a lot of people don't understand how freaking genius that company is.

I'm still getting by with my 2500k but I'd like to see something amazing from AMD, so I hope that the rumors are true. Knowing AMD it will probably be very affordable and I'll probably get a Zen in my next build then. A 14nm 95 TDP 8 core sounds nice!

Would indeed be nice, but i dont expect to see a "real" 8 core on the next Zen architecture. If the rumors are true, they are going with SMT. So its more likely we probably gonne see 4 core 8 threads cpu´s.

But yeah nobody can look in the future offcourse ☺

That might be a better idea, actually. We've seen what Intel can do with fewer but faster cores. I'm sure AMD can do it just as well. If not better then at least cheaper. Which is always good.

The thing with 1X nm node is that it will be the first FinFET node available to AMD. Both TSMC and Global Foundries have a 20 nm node, BUT it is planar, not FinFET. It's very dense yes, but have lots of problems for high performance stuff. So it is not really viable for big CPUs (and GPUs for that matter). So they are stuck at 32 nm PDSOI or 28 nm bulk.

Enter FinFETs. Intel have had that since their own 22 nm node, they call it Tri-gate. Sure there will be problems, FinFET is not the same as planar, but what is the alternative? Stay at 28 nm? Nope, not a good idea. Samsung to the rescue, it is their process that Global Foundries are licensing (GloFos own stuff didn't work so they scrapped it). TSMC has a fabled FinFET node coming, but delayed into 2016. If AMD waited for that their new stuff would probably be delayed into 2017, as you need significant time too make chips on a new process once it is available.

So they can't do much else than go "straight to 14 nm". They have no choice. They desperately need a stable FinFET process for their chips. And remember here that 14 nm FinFET is not much denser than 20 nm planar. The thing is that you actually can make big powerful stuff on FinFET that is not possible on 20 nm planar.

Interesting. Means that they'll basically spend 5 years developing this architecture. Has quite a bit of potential.

They picked a great name for the Project Skybridge. Halfway between Intel's Sky Lake and Sandy Bridge. Let's just hope that's not representative of the performance...

One thing that really worries me about this is that AMD will make the really dumb decision of trying to make the die as dense as possible which would make bulldozer look like a heated blanket in comparison. If my predictions about a large footprint chips is actually correct hopefully they will space out the transistors enough to combat mutual convection