AMD Moving to 12nm Vega and Ryzen in 2018

#5473879 Posted on: 09/21/2017 06:26 AM

---

Seems like 12nm lpp is an extention of 14nm lpp, much like tsmc's 12nm ffn is to 16nm ff, appears the intent is to move stuff from 14nmlpp , it specifically mentions it minimizes redesign cost from 14nmlpp.

#5473887 Posted on: 09/21/2017 07:12 AM

---

Very nice I hope they can do more with Vega being on 12nm but I doubt it'll make any noticeable difference. I can't help but feel like they would have been better off using a core with around 3000 shaders and 2 stacks hbm2 just going on gamers nexus comparisons.

#5473892 Posted on: 09/21/2017 07:28 AM

---

Isn't Vega already using 2 stacks of HBM2? Or perhaps you meant 4 x2 instead of the current 2 x4 design to reach the 8 GB capacity the consumer version currently has.

Shaders are a interesting one too, similar to the Fury versus Fury X GPU's there doesn't seem to be a major difference between the 64 and 56 which from my understanding of it after having read around a little bit about the subject the cards are being bottlenecked in other areas instead thus the Vega 56 when bios modded to remove some of their other limits actually do really well.

If there was a tool to allow locking and unlocking these clusters or cores it would have been really interesting to see just how far down one could take the GPU before performance was seriously impacted and how much headroom this would in turn allow for overclocking due to thermals and power usage dropping down, perhaps if disabling 8 of them (64 -> 56) gives some headroom then more could allow for the GPU to compensate by clock speed instead and perhaps see greater gains from that, eh perhaps it would hit some other limitation instead and at one point reducing these cores would start affecting performance too but there might have been some room for fine tuning things. (Perhaps later on bios modding could have some interesting possibilities depending on how things go.)

#5473897 Posted on: 09/21/2017 07:46 AM

---

Waiting for Zen+ /Zen 2.

#5473903 Posted on: 09/21/2017 08:05 AM

---

WHY NOT 10NM OR 7 NM
SAMSUNG HAVE IN 2018 7 NM

#5473921 Posted on: 09/21/2017 08:59 AM

---

So 7nm may not be as ready as they wanted.

Also, the 14nm to 12nm will probably be very compatible with each other, which means a straight copy of the masks etc from one process to the other.

i suppose it will enable lower power consumption, and maybe, more performance, but I would mostly expect lower power consumption.
It might enable Zen refresh to clock a bit higher, but its more likely to enable better mobile chips, GPU's and CPU's alike.

#5473923 Posted on: 09/21/2017 09:04 AM

---

WHY NOT 10NM OR 7 NM
SAMSUNG HAVE IN 2018 7 NM
Chillax

#5473925 Posted on: 09/21/2017 09:05 AM

---

2018 might be the year for me then.

The second Ryzen Generation and my current CPU aged enough to justify an upgrade.

#5473932 Posted on: 09/21/2017 09:14 AM

---

@JonasBeckman i meant keeping the current configuration just with less shaders as they dont seem to be helping all that much going from Vega 56 to 64. Both cards gain alot more from memory bandwidth increases. So I feel like 3000 shaders would have made the chip more balanced and easier/cheaper to produce.

#5473945 Posted on: 09/21/2017 09:44 AM

---

Wanted to upgrade my R9 290 to Vega 56, but since the custom models are delayed, quote "indefinitely", might as well wait a year more for the refresh and die shrink. The first generation Vega is a bit of a mess, but I was willing to get over it - until the availability and pricing problems came up.

Hopefully, by the time the refresh happens:
- AMD works out the more serious kinks of the arch (power consumption and shader bottlenecks)
- HBM2 availability goes up and price down
- mining craze dies down so we get sane pricing and better card availability

#5473995 Posted on: 09/21/2017 12:14 PM

---

WHY NOT 10NM OR 7 NM
SAMSUNG HAVE IN 2018 7 NM
It's not that easy, even Intel have been delaying 10nm for years now.
Also, I think the number doesn't compare between manufacturers because there isn't a definition to what it means. I think I've read Nvidia 16nm is the equivalent to AMD 14nm. It's bad when a company doesn't move forward for so long (like Intel), but we can't compare the numbers between companies.
So 7nm may not be as ready as they wanted.
Roadmaps show 7nm for second half of 2018, maybe they want to introduce refreshed products before that?
Wanted to upgrade my R9 290 to Vega 56, but since the custom models are delayed, quote "indefinitely", might as well wait a year more for the refresh and die shrink. The first generation Vega is a bit of a mess, but I was willing to get over it - until the availability and pricing problems came up.
Availability and pricing problems were stamped from the beginning. Don't wait on them unless you want to pay a premium. Maybe, and I say maybe, Vega 11 will have more availability. I'm really curious to see what AMD will lunch to replace Polaris, as I had to buy a RX560 I still need an upgrade.

#5474080 Posted on: 09/21/2017 03:44 PM

---

Wasn't AMD planning on releasing Zen (1st gen) deskop APUs in 2018? I'm sure this is what they're referring to that will get the 12nm treatment. As far as I'm concerned, this doesn't suggest "Zen 2" (at least the non-APU parts) won't get 10nm.

#5474155 Posted on: 09/21/2017 06:14 PM

---

12nm will be 10% increase in clocks while using 10% less energy which will result in slightly less heat and power consumption. Jumping from 14/16nm to 7nm would be great but i believe they want to stretch it out as much as possible. Once you hit 7nm there's hardly any room left for them poor electrons, they will start jumping off like fleas.

#5474169 Posted on: 09/21/2017 06:36 PM

---

WHY NOT 10NM OR 7 NM
SAMSUNG HAVE IN 2018 7 NM

Samsung is planned to be able to fabricate chips on their 7nm node 2H 2018 so will Gloflo. If AMD is doing a 12nm refresh first part of 2018(by the way there will be no Samsung 7nm then) I could see them bumping 7nm back to early 2019. Samsung and GloFlo are on right around the same dates. We may seem some products from Samsung first as they are doing simpler SRAM on there first wave of 7nm production.

One final note is Glflo aka AMD's fab partner will very likely have most dense node on their 7nm as its predicted to be a tad denser than Intel's 10nm, Samsung's 7nm and TSMC's 7nm based off all the available info, for example:
https://www.semiwiki.com/forum/content/6879-exclusive-globalfoundries-discloses-7nm-process-detail.html

#5474302 Posted on: 09/21/2017 11:08 PM

---

I think it will be waay better by power consumption, thus higher gpu frequency bringing it to 1080ti territory.