AMD to Build 2nd Gen. Ryzen and Radeon Vega on GloFo 12nm

[btarunr]

Not to be held back by silicon fabrication process limitations like in the past, AMD will build its second-generation Ryzen CPUs and Radeon Vega GPUs on the new 12 nanometer LP (low power) FinFET process by GlobalFoundries. From the looks of it, "2nd generation Ryzen" doesn't seem to be the same as "Zen2" (a micro-architectural advancement due to be built on the 7 nm process), and is more likely an optical shrink of existing 14 nm IP to the 12 nm process, giving AMD the headroom to increase yields, and clock speeds across the board. The 12 nm switch allows AMD to roll out a new "generation" of Ryzen processors as early as the first half of 2018.

The "Vega 10" silicon could be another key piece of AMD IP on the receiving end of an optical shrink to 12 nm, which will give AMD much needed power savings, letting it increase clock speeds, and probably implement faster standards of HBM2 memory, such as 2.00 GT/s. AMD will likely label this shrunk down silicon "Vega 20." There's also the possibility of AMD building a bigger new GPUs altogether. In 2019, the company will give its CPU and GPU lineups major micro-architectural upgrades, and the switch to the 7 nm node. The new "Zen2" micro-architecture with IPC increases and new ISA instruction-sets, will be launched on the CPU side, and the new "Navi" graphics architecture will take center-stage.



View at TechPowerUp Main Site

 

[Rehmanpa]

So does this mean no zen 2 in 2018, it's 2019 now?

edit: Misread the post, I guess with 12nm AMD is releasing new ryzen processors in 2018 but a major advancement is in 2019? Sorry it's a bit confusing

 

[Jism]

Back in the days when AMD had their own fab(s), they would simply 'refresh' the current chip or node to a better proces, making them able to deliver chips that offer a higher clock, lower power and small fixes that the previous chips failed on.

I think AMD is continuing this refresh, but being independent of GloFo, they are going for 12NM rather then sticking onto the 14NM proces. This saves both partys some time as GloFo's best interest is to provide a silicon base which is able to clock higher with less power requirements.

I woud'nt say that LP Finfet is 'bad' but it explains why most Ryzens are hitting a wall beyond 4.2Ghz. The Vishera's for example where on a different proces and where able to clock much higher (5GHz on air).

 

[de.das.dude]

Back in the days when AMD had their own fab(s), they would simply 'refresh' the current chip or node to a better proces, making them able to deliver chips that offer a higher clock, lower power and small fixes that the previous chips failed on.

I think AMD is continuing this refresh, but being independent of GloFo, they are going for 12NM rather then sticking onto the 14NM proces. This saves both partys some time as GloFo's best interest is to provide a silicon base which is able to clock higher with less power requirements.

I woud'nt say that LP Finfet is 'bad' but it explains why most Ryzens are hitting a wall beyond 4.2Ghz. The Vishera's for example where on a different proces and where able to clock much higher (5GHz on air).

visheras had 2 generations of improvement so its natural they can hit 5GHz on air in a cool room. Mine could

 

[Jism]

They are just improving the chip itself on a smaller node (12nm vs 14nm finfet). I woud'nt say it will save huge amounts of power but it will surely bring the chips towards the desired clockspeeds AMD is looking for. The ryzen / threadripper are amazing chips. Half the price of intel's counterparts.

 

[Rehmanpa]

so will these newer 12nm chips be able to fit on current ryzen motherboards? I'm currently considering a threadripper build and I don't want to dish out a bunch of money on a mobo/cpu only to have it replaced in a few months.

 

[Jism]

AM4 is here to stay basicly.

 

[wizyy]

Why are they stating certain improvements over 16nm (TSMC)?
Why would they not measure improvements over their own 14nm?
Just curious...

 

[jigar2speed]

Why are they stating certain improvements over 16nm (TSMC)?
Why would they not measure improvements over their own 14nm?
Just curious...

Ever wonder what is being made at 16nm by TSMC and why AMD is comparing 12nm GF to 16nm TSMC ? hint - Nvidia GPU.

 

[sergionography]

Why are they stating certain improvements over 16nm (TSMC)?
Why would they not measure improvements over their own 14nm?
Just curious...

https://www.globalfoundries.com/new...-technology-for-high-performance-applications

According to this linke they say 10-15% compared to current 16/14nm FF solutions. All in all idk how i feel about this. Ryzen is super good but it definitly got its rough edges that need to be ironed out
For starters ryzen is very power efficient so a shrink isnt whats needed to help it perform as needed. It needs to be revised to clock higher, and to me it seems a simple die shrink wouldnt result in much progress there as the power efficiency gain will most likely be advantages below the 4ghz mark but not at peak. Amd need to be clocking ryzen the 5ghz mark to stay relevant against intel who seems to be taking advantage of their peak turbo clocks right now to differentiate their products, and with them moving to 6 and 8 core mainstream parts Amd cant simply throw more cores at the problem.

The main benefit i can see from this is that amd can improve their interconnect to scale better and have lower latencies all while producing even smaller chips; which will help in producing higher supply targeting lower pricepoints, and higher marketshare.

 

[TheGuruStud]

https://www.globalfoundries.com/new...-technology-for-high-performance-applications

According to this linke they say 10-15% compared to current 16/14nm FF solutions. All in all idk how i feel about this. Ryzen is super good but it definitly got its rough edges that need to be ironed out
For starters ryzen is very power efficient so a shrink isnt whats needed to help it perform as needed. It needs to be revised to clock higher, and to me it seems a simple die shrink wouldnt result in much progress there as the power efficiency gain will most likely be advantages below the 4ghz mark but not at peak. Amd need to be clocking ryzen the 5ghz mark to stay relevant against intel who seems to be taking advantage of their peak turbo clocks right now to differentiate their products, and with them moving to 6 and 8 core mainstream parts Amd cant simply throw more cores at the problem.

The main benefit i can see from this is that amd can improve their interconnect to scale better and have lower latencies all while producing even smaller chips; which will help in producing higher supply targeting lower pricepoints, and higher marketshare.

Clocks won't even matter if intel can't fix their 10nm problems LOL. That's why they're still not releasing performance desktop CPUs (if ANY desktop, maybe NUC?) in 2018 at all. After all the delays, they lost another 6 months at minimum.

At this rate, by the time 10nm actually comes, we'll have Zen 2 in full production. Ice Lake being another micro-increase over cannon...ouch.

 

[iO]

Sounds a lot like TSMCs 12nm process or Intels Hyperscaling. Same process with some tweaks but a new name.

Might be enough for 100-200Mhz more..

 

[Parn]

so will these newer 12nm chips be able to fit on current ryzen motherboards? I'm currently considering a threadripper build and I don't want to dish out a bunch of money on a mobo/cpu only to have it replaced in a few months.

According to AMD traditions, AM4 along with the current 300 series chipsets should be able to support everything up to at least Zen 2.

 

[efikkan]

I would appreciate if people stopped calling this a node shrink. GloFo "12nm" is just a refinement of GloFo "14nm", just like TSMC "12nm" is a refinement of TSMC "16nm". Intel does it too, with their "14+" and "14++" enhanced nodes. xx nm is just a marketing name at this point, just take a look:

(note: this is a logarithmic scale)

so will these newer 12nm chips be able to fit on current ryzen motherboards? I'm currently considering a threadripper build and I don't want to dish out a bunch of money on a mobo/cpu only to have it replaced in a few months.

There is a good chance it will, since it's just a new stepping.
But why would you buy one and then upgrade? That's like upgrading from Skylake to Kaby Lake.

Sounds a lot like TSMCs 12nm process or Intels Hyperscaling. Same process with some tweaks but a new name.

Might be enough for 100-200Mhz more..

Exactly.
Just a new stepping with some tweaks on a slightly refined node.

 

[Steevo]

Its a divergent process as the 3D FIN tech used by Intel is different than those used by other fabs. One of the other things to consider is how "clean" the process is in creation of the transistors, so a leaky 10nm might be worse than a 14nm equal part, due to how much extra impedance is created by dirty edges of the fin and capacitive buildup.

Does the process really matter as long as power consumption and clock is competitive? A full on example is Ryzen kicking around current Intel in power consumption, sure half of it is better power control, but the other part is lower leakage even if the transistor needs a little more switching power.

 

[Krzych]

One thing to give to AMD is that they are really good at maintaining hype and hopes, making people wait. There is always something on, even if something new flops there are instantly new rumors, announcements and promises. Not much is delivered in the end, those are always second grade products and the very peak of what AMD they is capable of is pricing those second grade products extremely low (Ryzen), but all of this "wait" nonsense is going on for years and years and only gets stronger. Mostly from people with no money who are looking to justify why they are not buying anything, they can now delude themselves and say that they are waiting instead of not being able to afford, and quality products from the competition are for the taking at any given time, so no serious customer is going to wait, but still, if AMD is good at something, this is it.

 

[sergionography]

Clocks won't even matter if intel can't fix their 10nm problems LOL. That's why they're still not releasing performance desktop CPUs (if ANY desktop, maybe NUC?) in 2018 at all. After all the delays, they lost another 6 months at minimum.

At this rate, by the time 10nm actually comes, we'll have Zen 2 in full production. Ice Lake being another micro-increase over cannon...ouch.

Well actually the main reason intel shouldnt bother with 10nm is because their 14nm is pretty darn good and after so many years of refinement are cloxking up to 5ghz. I doubt 10nm can clock that high atleast initially. And right now its not about fabrication process as both amd and intel are as close in process and ipc than ever but to be fair amd is still a bit behind there as ryzen ipc is about 5-10% behind and cannot clock beyond 4.2ghz or so. Intel on the other hand are releasing chips with turbo to like 4.7ghz or so, keeping them with a healthy over 20% better single threaded performance. The playing field right now is that amd has better multithreaded performance while intel has better single threaded performance, but that will change once intel release an 8 core mainstream chip in 2018

 

[R-T-B]

ryzen ipc is about 5-10% behind

It's actaully far closer than that. At least, barring Infinity Fabric issues due to slow ram or something. My Ryzen basically matches Kaby Lake IPC in CPU-Z, yes not the most scientific bench, but 10% behind? No way.

 

[sergionography]

It's actaully far closer than that. At least, barring Infinity Fabric issues due to slow ram or something. My Ryzen basically matches Kaby Lake IPC in CPU-Z, yes not the most scientific bench, but 10% behind? No way.

I think thats what i remember from the review here. But yes you are right for the most part as many applications match in ipc but still with some falling slightly behind probably due to optimization. Regardless though amd is behind on clockspeed and right now we are living good days again being able to compare performance between amd and intel by clockspeed and core numbers. Rocking a ryzen 7 1700 and am very pleased.

 

[jabbadap]

Sounds a lot like TSMCs 12nm process or Intels Hyperscaling. Same process with some tweaks but a new name.

Might be enough for 100-200Mhz more..

Yeah you are most probably right. Sounds like a redefined 14nm LPP. It would be super cool if it could have been redefined ibms 14HP(FinFets+SOI), but reading that Globalfoundries news it's obviously not.

And Risk Production in H1 2018!? Erhm so if all goes to bonkers again, we could see something new on that process in Q4/2018.