New Zen microarchitecture details

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

zentan

Member
Jan 23, 2015
177
5
36
Samsung 14FF so far has only been proven in low-power mobile chips (i.e. A9, SD820, Exynos 7/8). There is no evidence thus far that it can deliver >4GHz, high perf/clock chips in anything close to the kind of power envelope that Intel's can.

Intel's processes are fine tuned for the very limited range of fairly high-performance PC processors that it builds, while Samsung/TSMC are basically gunning for the mobile guys first and foremost.

Intel also builds its own chips & doesn't have to hand over any foundry margin to Samsung/GloFo/TSMC, which certainly helps on the cost structure front.

BTW, don't you think it's just a teensy-weensy bit suspicious that AMD is able to fit 8 whole super duper Zen cores into a 95W power envelope on a process that is probably inferior for high performance CPUs than Intel's 14nm? What do you think those CPU cores are going to come clocked at? 4GHz at Skylake-like IPC? Very doubtful.

Perhaps Zen would have a significant clock-deficit to Intel or maybe they can try to make it as closer as possible but w/ taking a significant beating on power.This is all assuming that Samsung 14nm can't go head-to-head w/ Intel 14nm at higher clock-ranges. Also till now there's no actual tests or benches we have seen.Think,we should wait more.

I am more interested in Zen-based APUs,expecially for notebook market.
It is where AMD could have leveraged its more core advantage if it goes w/ it's traditional 4core options with or without smt enabled.It is where it could have taken well on the dual core i5s which are what most mid-range notebooks are especially in MT(this is assuming Zen is a pretty significant improvement in per core performance and close enough to ~SB or IB ipc).But it seems zen-based APUs would be launched closer to cannonlake on notebooks.Intel might be more equipped with replies by then.
 
Last edited:

Soulkeeper

Diamond Member
Nov 23, 2001
6,712
142
106
Bristol Ridge and AM4 motherboards release imminent ?
Seems like FM2+ supply is drying up, any NDA on this stuff ?
 

PPB

Golden Member
Jul 5, 2013
1,118
168
106
You can get today a 5820K for 320 bucks while the 5960X goes for 900. If someone is telling us that a competitive 8c/16t part with similar clocks than the 5930k but smaller die size (~19nm density+HDL+2CH DDR4IMC vs 22nm density+4CH DDR4 IMC + More I/O lanes) cant go just in the middle tending to 5820k's pricing, you are just delusional.

We can tell it is already an smaller core than HW/BW/SKL by design decisions, dismiss all the uncore stuff that needs to go in lga2011-3 and wont be in a future AM4 platform, add a node that is just a little smaller parameter wise compared to Intel's 22nm. You have an smaller die whenever you compare both designs. Then add some perks like HDL to aid the density and you have really a small die for 8c/16t.

Instead of being butthurt that AMD is probably bringing more cores to mainstream platforms, you all should cheer for Zen's success so we finally get over the dreaded 4 core mainstream trend. And no, that rumor of Cannonlake going 6/8 core is probably implying that it's Xeon-D variant should see some consumer action, that probably will happen but with Intel's juicy margins behind it.
 

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
Could you please provide proof that Samsung's 14nm is "more efficient" than Intel's? Thanks.

Additionally, if you could provide a statement from a "competent engineer" that'd be great, preferably one with experience with both processes.

EDIT: To be clear, Abwx, please provide a proof that makes very little assumptions on the part of the reader. If you have a legitimate proof, you should be able to explain it without saying things like "people without EE degrees don't understand" and "trust me, I'm a retired semiconductor professional." Take it from first principles, explain all terms along the way, and show your math. Try to make a valuable contribution.

For GF i made a simplified mathematical demonstration that could be understood by about whoever know the four basic opeations, so far it wasnt contradicted neither here nor at SA where i explained the things more extensively, the numbers i estimated besed on GFs enginers dedicated conference are confirmed by their most recent slides...

As for Intel underdelivering the hint was posted in a conversation between engineers at EE Times after one noticed that the device performances have been degraded when shrinking to 14nm from 22nm, this wasnt difficult to check with all available infos that have a physical relevance.

Anyway you ll have the occasion to check my sayings in a few months if GF s slides or Intel s current public numbers are not enough for you.
 
Mar 10, 2006
11,715
2,012
126
For GF i made a simplified mathematical demonstration that could be understood by about whoever know the four basic opeations

Doesn't help me if you don't present it here and explain it thoroughly. Please do so.

, so far it wasnt contradicted neither here nor at SA where i explained the things more extensively, the numbers i estimated besed on GFs enginers dedicated conference are confirmed by their most recent slides...

OK, great, but that's probably because you're not explaining it well enough for anybody to confirm/disprove it. You make too many assumptions and skip over a lot of steps. If you want people on internet forums to understand, please assume as little knowledge on the part of the reader as possible.

As for Intel underdelivering the hint was posted in a conversation between engineers at EE Times after one noticed that the device performances have been degraded when shrinking to 14nm from 22nm, this wasnt difficult to check with all available infos that have a physical relevance.

Oh, right, the comment some dude made that Nenni ripped off and used in a blog post, pretty sure this is from user "carop." I seriously doubt that xtor device performance degraded in going from 22nm -> 14nm; the performance/power levels achieved in real shipping 14nm product from Intel very likely could not have been achieved on Intel 22nm.

Anyway you ll have the occasion to check my sayings in a few months if GF s slides or Intel s current public numbers are not enough for you.

I am sure when Zen hits, Chipworks will do a detailed transistor characterization report. Too bad I don't think anybody on this forum is going to be willing to spend ~$10k on such a report.

Anyway, still waiting for proof. Remember, you have to prove it in a way that somebody w/ little EE experience can understand the gist of the argument. Thanks.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
~Broadwell IPC for Zen cores? Hmm,would need some actual proof of that. I just hope it doesn't turn out to be another hype-train till the launch.

IPC is useless if we don't know clocks. For all we know it could have skylake IPC at 2.5 GHz and hence be quiet useless.


Has there been any finfet CPU/SOC clocking at even 3 Ghz outside of Intel? Not that I'm aware of. Heck only CPUs outside Intel reaching 3 Ghz were from AMD. So the capabilities of process are for sure a big ?.
 

IllogicalGlory

Senior member
Mar 8, 2013
934
346
136
I'm really hoping that Zen shakes up the low end market. I'm tired of Athlons, Pentiums and i3s. None of them are good enough anymore. Give us a real quad core <$140!

Offering a 2C/4T Zen chip at $70+ would be great too.
 

jhu

Lifer
Oct 10, 1999
11,918
9
81
Has there been any finfet CPU/SOC clocking at even 3 Ghz outside of Intel? Not that I'm aware of. Heck only CPUs outside Intel reaching 3 Ghz were from AMD. So the capabilities of process are for sure a big ?.

Incorrect. IBM POWER 6 already made it to 5 GHz back in 2007 on a 65 nm process.
 

deasd

Senior member
Dec 31, 2013
512
724
136
I'm surprised and glad to see Samsung to manufacture CPUs for AMD. It would have no serious capital problem like initial GloFo manufacturing FX and Llano.
 

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
I'm really hoping that Zen shakes up the low end market. I'm tired of Athlons, Pentiums and i3s. None of them are good enough anymore. Give us a real quad core <$140!

Offering a 2C/4T Zen chip at $70+ would be great too.

FX 8 core chips are less than $140. FX 4 & 6 core chips are less than $100.

Been at those prices for a while.
 

IllogicalGlory

Senior member
Mar 8, 2013
934
346
136
FX 8 core chips are less than $140. FX 4 & 6 core chips are less than $100.

Been at those prices for a while.
Yes but those are mostly useless. i3s perform as well or better in games. Not interested. I should have specified CPUs with a decent uarch. The only way to make use of those requires overclocking and that's an additional expense.
 
Last edited:

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Could you please provide proof that Samsung's 14nm is "more efficient" than Intel's? Thanks.

Additionally, if you could provide a statement from a "competent engineer" that'd be great, preferably one with experience with both processes.

EDIT: To be clear, Abwx, please provide a proof that makes very little assumptions on the part of the reader. If you have a legitimate proof, you should be able to explain it without saying things like "people without EE degrees don't understand" and "trust me, I'm a retired semiconductor professional." Take it from first principles, explain all terms along the way, and show your math. Try to make a valuable contribution.

On that note could you tell us how much it costs for Intel to have their own foundries?
 

Erenhardt

Diamond Member
Dec 1, 2012
3,251
105
101

cbn

Lifer
Mar 27, 2009
12,968
221
106
FX 8 core chips are less than $140. FX 4 & 6 core chips are less than $100.

Been at those prices for a while.

You can get FX8320E and USB 3.1 mobo for $125 at microcenter. Good for 4,5 Ghz Overclocks.
http://www.eteknix.com/msi-970a-sli-krait-edition-motherboard-review/4/
http://www.anandtech.com/show/8864/amd-fx-8320e-cpu-review-the-other-95w-vishera/2

I don't think Zen can touch perf/$ Vishera gives*.

*in multithreaded scenario.

FX8300 is $120 shipped at Newegg (with free Hitman 2016 game code) for folks that don't live near a Microcenter.
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
If ZEN is what im expecting, at the time of release with more DX-12 games and new 14/16nm GPUs already on the market, i can see them selling at high price for the first 6-12 months.

Also i would like to talk about the die size, the 4C 8T GT2 Skylake only has 8MB of L3 Cache. If AMD will only use a single die for both Server and Desktop SKUs, then im expecting a huge L3 or even L4 Cache. That will increase the die size of a 8C 16T to 160-200mm2.
 

cbn

Lifer
Mar 27, 2009
12,968
221
106
@AtenRA

Regarding Zen APU, what are you thoughts on other chips AMD could use on the interposer besides HBM?

Maybe for a mobile chip other ICs could be justified to make the laptop board more compact?

More compact laptop board = more room for cooling and thus thinner and lighter 35W mobile device?
 

myocardia

Diamond Member
Jun 21, 2003
9,291
30
91
FX8300 is $120 shipped at Newegg (with free Hitman 2016 game code) for folks that don't live near a Microcenter.

I'm not sure when you checked that price, but right now it's $110 @ newegg, and still has the free game.
 

myocardia

Diamond Member
Jun 21, 2003
9,291
30
91
Server SKUs are rumored to be 16C/32T and even 32C/64T, along with APU designs.

They will be making only 8C/16T dies first, and the 4C/8T APUs after. Their server SKUs will be using some form of the interconnect fabric that they inherited when they bought Seamicro, to join the 8C/16T dies together.
 

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
They will be making only 8C/16T dies first, and the 4C/8T APUs after. Their server SKUs will be using some form of the interconnect fabric that they inherited when they bought Seamicro, to join the 8C/16T dies together.

And how, precisely, are they going to do cache coherence over Seamicro fabric? o_O
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
Server SKUs are rumored to be 16C/32T and even 32C/64T, along with APU designs.

I dont know if and when they will make 16/32C dies but im only talking now for the 8C 16T dies they will release later this year. Those dies could be the same for the Desktop and Server.
If that is the case then they could have a huge L3 and/or L4 cashe.
Well now that i thing about it, they could aways play the same game as Intel and release a 4C 8T small die at sub 100mm2 for the desktop and entry Server market. It could be great to have a 4C 8T ZEN at $200 to compete against the Core i5. For the same price you could have more throughput but less SingleThread performnce, but with DX-12 games already in the market at the end of 2016 this sub 100mm2 die could be the next big splash for gamers.

And with this small die, they would actually be able to sell a 2C 4T CPU for sub $100 as well.