bitsandchipsNo Dual and Four Cores Zen CPUs, at least initially

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Feb 19, 2009
10,458
0
76
#51
Samsung claims defect rate of 0.2/cm2, 14nm FF wafers are ~$4,500.

A chip like Zen, is a little bigger than Skylake 4/8 currently due to ~50% of its die as the iGPU. No more than 160mm2 for Zen, IMO. Yields will be excellent if what Samsung claims is true.

Most of the chips will be fully functional, the rest can be harvested for 4/6 core SKUs.

Cost per chip is essentially ~$30.

For real. Do you think AMD can't be flexible and price it very aggressive to get sales?

Do the maths for Intel's Skylake, ~120mm2, and it's clear why they made over 14B profit last year on ~55B revenue.
 
Mar 10, 2006
11,696
84
126
#52
Do the maths for Intel's Skylake, ~120mm2, and it's clear why they made over 14B profit last year on ~55B revenue.
$14 billion on $55 billion in revenue isn't that great compared to a TSMC. They did ~$10 billion in operating profit on $26.4 billion in revenue.

Can't wait to hear about how TSMC is ripping its customers off :)
 
Mar 10, 2006
11,696
84
126
#53
A chip like Zen, is a little bigger than Skylake 4/8 currently due to ~50% of its die as the iGPU. No more than 160mm2 for Zen, IMO. Yields will be excellent if what Samsung claims is true.
.
Samsung only gave defect density numbers, but the yield issues that Samsung had (and is actually said to still be having) on 14nm is one of parametrics. That is, hitting desired clocks at desired power consumption level. This has nothing to do with defect density.

The reason that a 6700K Skylake is $350 isn't because Intel is making obscene profits on a 120mm^2 die (they're making good margins sure but nothing out of this world for a well-run semico, fabless or not), it's because of the chips that come off the line, only the very best get the honor of being binned to run at 4GHz base/4.2GHz turbo at a given voltage.

Stop using die size as the be-all, end-all metric of "value" delivered to the customer. It's misleading.
 
Jan 23, 2015
177
0
36
#54
$14 billion on $55 billion in revenue isn't that great compared to a TSMC. They did ~$10 billion in operating profit on $26.4 billion in revenue.

Can't wait to hear about how TSMC is ripping its customers off
:)


Samsung only gave defect density numbers, but the yield issues that Samsung had (and is actually said to still be having) on 14nm is one of parametrics. That is, hitting desired clocks at desired power consumption level. This has nothing to do with defect density.

The reason that a 6700K Skylake is $350 isn't because Intel is making obscene profits on a 120mm^2 die (they're making good margins sure but nothing out of this world for a well-run semico, fabless or not), it's because of the chips that come off the line, only the very best get the honor of being binned to run at 4GHz base/4.2GHz turbo at a given voltage.

Stop using die size as the be-all, end-all metric of "value" delivered to the customer. It's misleading.
Well said, making trans/mm^2 as the one and only metric is very short-sighted. Some would say TSMC and Intel are ripping us off as TSMC and Intel doesn't need any of those profits and R&D going forward.
 

rtsurfer

Senior member
Oct 14, 2013
733
0
76
#55
Most likely they will initially only release the full parts and put the defective parts aside for now. After they have sufficient amounts of the defective, but salvageable parts they will release them as lower end variants.

Potential harvested parts:

8C/16T /w 8MB L3
6C/12T /w 16MB L3
6C/12T /w 8MB L3
4C/8T /w 8MB L3
4C/8T /wo L3

Why have the 6-core with bigger cache than 8-core..? :\
 
May 1, 2014
74
0
0
#56
Why have the 6-core with bigger cache than 8-core..? :\
He just lists theoretical SKUs. The 8core with 8MB cache has part of it disabled while th 6core has the full cache but deactivated cores.
 
Mar 10, 2004
27,986
107
126
#57
AFAIK there will be only a single Zen CPU die, which will be used for all of the products (regardless of the segment). Raven Ridge (Zen APU) obviously is a different design entirely.
That makes the most sense to me.
 
Mar 10, 2004
27,986
107
126
#58
I feel just the opposite, as evidenced by my continual haranguing for a mainstream hex core from intel. At least the "top end system" should be on the most current process node and architecture. I suppose you could make a special platform for it with more features and so forth, but it is ridiculous for the "top end system" to be one (or currently 2 until BW-E) generations behind the mainstream products.
Well, I want AMD to push Intel with Zen, so if it results in a 6 core 1151 chip, I won't be unhappy about that. :D

But if top end Zen is HEDT only, in fact or in price, then it doesn't seem likely. D:
 
Feb 2, 2009
12,765
71
126
#59
Well, I want AMD to push Intel with Zen, so if it results in a 6 core 1151 chip, I won't be unhappy about that. :D

But if top end Zen is HEDT only, in fact or in price, then it doesn't seem likely. D:
I believe top ZEN will be between Quad Core Kabylake and Core i7 6-8 Core BD-E.
But AM4 is not a HEDT platform like 2011-V3.
 

myocardia

Diamond Member
Jun 21, 2003
9,223
0
71
#60
...
 
Last edited:

jaydee

Diamond Member
May 6, 2000
4,487
0
81
#62
Isn't this something that sort of bit them in the rear a few years back? They were taking healthy 4-core Phenom IIs to cover demand for 2/3-Core SKUs?

There was even a viral marketing pitch of buying 2/3-Core chips and unlocking the 3/4 cores. Hell, even I partook and got the GF a Phenom II X2-something and unlocked it to X4 with no issues.

EDIT:
Here is the ATF Article on it (which is most likely where I first heard of it):
http://www.anandtech.com/show/2927
I'm still running with that config!
 
Feb 19, 2009
10,458
0
76
#63
Is that a fact? :sneaky:
No, that's a good guess.

Skylake is ~120mm2. Half of it is the useless iGPU which Zen will not have.

4c/8t SMT is ~60mm2.

Obviously it won't scale perfect and Intel's node may be slightly better, but I don't expect 8c/16t Zen to be a very large chip.
 

jaydee

Diamond Member
May 6, 2000
4,487
0
81
#64
We could have the same situation that we had with AM3(+). Same socket, but some boards did not support the top end chip power requirements. Some barely supported the 125W chips, I think?

I think it makes some sense to have HEDT separated from the rest. There is a market for top end systems, and people pay more for it. People like to be elite. :)

Mfgs can make a little more money with the HEDT boards, too.
Was that an AMD design thing? I always kind of chalked it up to MB makers cheapening out in some way. My AM3 board (granted mini-ITX) was only rated to 95W I think.
 

deasd

Senior member
Dec 31, 2013
201
0
51
#65
Obviously it won't scale perfect and Intel's node may be slightly better, but I don't expect 8c/16t Zen to be a very large chip.
I bet in same process node we won't see any significant difference between vendors like intel/samsung/glofo/tsmc.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
64
106
#66
I bet in same process node we won't see any significant difference between vendors like intel/samsung/glofo/tsmc.
Isn't Samsungs 14nm ~21% larger than Intels 14nm? Effectively a ~17nm node?
 
Mar 10, 2004
27,986
107
126
#67
Was that an AMD design thing? I always kind of chalked it up to MB makers cheapening out in some way. My AM3 board (granted mini-ITX) was only rated to 95W I think.
I'm not sure, but I think we can count on mobo mfgs cheapening out if no one is regulating them. :)

So far, it seems like Intel keeps a tighter rein than AMD.
 

deasd

Senior member
Dec 31, 2013
201
0
51
#68
Isn't Samsungs 14nm ~21% larger than Intels 14nm? Effectively a ~17nm node?
LOL, in this world there's no 17nm node. And all of those numbers are just can be read on paper but not real world case......
 
Feb 2, 2009
12,765
71
126
#69
Isn't Samsungs 14nm ~21% larger than Intels 14nm? Effectively a ~17nm node?
What is known is the following Gate Pitch x Metal Pitch

Intel 14nm = 70nm x 52nm

Samsung 14nm LPE = 78nm x 64nm

But, I also believe that 8-Core ZEN die should be smaller than 200mm2. It could be close to 160mm2 as Silverforce11 estimates.
 
Last edited:
Mar 13, 2006
10,089
47
126
#70
No, that's a good guess.

Skylake is ~120mm2. Half of it is the useless iGPU which Zen will not have.

4c/8t SMT is ~60mm2.

Obviously it won't scale perfect and Intel's node may be slightly better, but I don't expect 8c/16t Zen to be a very large chip.
You think GloFo is getting anywhere near Intel's transistor density?
 
Feb 2, 2009
12,765
71
126
#71
You think GloFo is getting anywhere near Intel's transistor density?
At 14nm they are as close as ever. Intel still has higher density but the lead is very small this time.
 

iGigaflop

Junior Member
Apr 20, 2016
16
0
0
#72
I was trying to wait on zen but i went out and picked up a 5820k and a asus x99 pro and 32gb 3000mhz ddr4, i had a 3770k that im gonna put in a small media center build and use my cm storm stryker case for my 5820k. I hope zen comes out and does great just dont want it to outperform intels e series chips. I mean the fx 8 core chips might be old but are great for a gamer on a budget.
 
Oct 10, 1999
11,915
0
76
#73
I was trying to wait on zen but i went out and picked up a 5820k and a asus x99 pro and 32gb 3000mhz ddr4, i had a 3770k that im gonna put in a small media center build and use my cm storm stryker case for my 5820k. I hope zen comes out and does great just dont want it to outperform intels e series chips. I mean the fx 8 core chips might be old but are great for a gamer on a budget.
Same here except I opted for those dual Xeon E5 2670s. There's no way Zen (or any single socket CPU for that mattrr) can outperform this thing in rendering.
 
Mar 13, 2006
10,089
47
126
#74
At 14nm they are as close as ever. Intel still has higher density but the lead is very small this time.
Define what you think is very small and we'll see.
 
Feb 19, 2009
10,458
0
76
#75
You think GloFo is getting anywhere near Intel's transistor density?
No, not that close, but it doesn't need to be because the difference, ultimately, is not going to result in a much bigger chip.

Think about this and let it sink in. 4C/8T SMT of Skylake quality is ~60mm2.

Read that again, maybe you understand, these chips are small.

The cost is mostly R&D at this point, because manufacturing these small chips are cheap!
 

Similar threads



ASK THE COMMUNITY