[Sweclockers] AMD Zen coming in Q3 2016, will be on 14 nm

[witeken]

Not sure what sm625 meant. But he said skylake, not cannonlake, which is 10nm. Skylake on 14nm should be out sometime this year, and by 2017 we should be close to another shrink. But as someone else said, I am not putting much faith in any roadmaps these days.

Cannonlake is mainly a die shrink from SKL (Tick). If AMD will release a desktop CPU in Q3, it will probably compete for most of its life time againt CLK.

 

[frozentundra123456]

Cannonlake is mainly a die shrink from SKL (Tick). If AMD will release a desktop CPU in Q3, it will probably compete for most of its life time againt CLK.

yea, you are right, at least in the old tick/tock cadence. but who knows, 10nm could come first to mobile like Broadwell.

 

[witeken]

yea, you are right, at least in the old tick/tock cadence. but who knows, 10nm could come first to mobile like Broadwell.

There's no old or new Tick-Tock cadence. Still just simply Intel approaching the end of Moore's Law as fast as possible.

 

[NTMBK]

No way they would go with quad channel. If they really need massive memory bandwidth for e.g. integrated graphics, I would expect a HBM cache backed up by dual channel DDR4.

8 cores sounds a little worrying. That makes it sound like Zen is a Jaguar successor, not a Skylake competitor...

 

[cbn]

Maybe to provide sufficient bandwidth for the iGPU, if they won't be using HBM?

If the memory is going to be DDR4, then that would mean it would take at least 16GB of RAM to populate all four channels.

 

[cbn]

8 cores sounds a little worrying. That makes it sound like Zen is a Jaguar successor, not a Skylake competitor...

Maybe that 95 watts was arrived via a small iGPU? If that is the case the cores wouldn't be too low power, although it could also be that AMD got TDP up that high by pushing the cpu towards the end of the design specs. (Example: think of how FX-9590 was pushed to the end of its frequency/voltage curve in order to achieve a high TDP, and maybe apply a less extreme version of that to this scenario).

 

[mrmt]

Also from the SweClockers article:

Finally 8 cores om mainstream desktop then. In that case 95 W TDP does not sound too bad.

8 cores on the desktop are so 2011. Bulldozer is dead, long live Bulldozer!

 

[geoxile]

Up to 8 core on 95W. If true it sounds like Zen will in fact be "small" core.

 

[Abwx]

Up to 8 core on 95W. If true it sounds like Zen will in fact be "small" core.

So 14nm has transistors as big as 28nm and density will be the same.?.

 

[jpiniero]

Up to 8 core on 95W. If true it sounds like Zen will in fact be "small" core.

K12 and Zen have always been described as something closer to Jaguar. It's going to be optimized for MT over ST.

 

[Abwx]

K12 and Zen have always been described as something closer to Jaguar. It's going to be optimized for MT over ST.

You are making a confusion with the two Skybridges cores, Zen and K12 are two high performance parent cores for X86 and ARM respectively.

 

[cbn]

So 14nm has transistors as big as 28nm and density will be the same.?.

^^^ Abwx makes a good point. (Remember the processor is on 14nm FinFET so 95 watts goes farther than what it would on 28nm/32nm planar.)

With that mentioned I am concerned this will be a design aimed most for MT.

Furthermore, what do we know of the Samsung 14nm process tech that GF is licensing? Is there a high drive current/high leakage variant of the process or it what most people would call a "SOC process tech"? (I am thinking mostly the later will be true so I wonder how high the frequency will be able to extend and at what cost to power consumption)

 

[Fjodor2001]

Cannonlake is mainly a die shrink from SKL (Tick). If AMD will release a desktop CPU in Q3, it will probably compete for most of its life time againt CLK.

AMD Zen (14 nm): 2016Q3
Cannonlake (10nm): Early 2017 (Q1/Q2)?

That means Zen will be competing against Broadwell 14 nm for about 6 months before going up against Cannonlake.

Also, remember that Zen is a completely new architecture. So it has potential to provide significant performance gains compared to previous generations.

The Intel Core architecture base is from 2006, and will not see any similar revamp in the foreseeable future from what has been communicated by Intel so far.

 

[R0H1T]

AMD Zen (14 nm): 2016Q3
Cannonlake (10nm): Early 2017 (Q1/Q2)?

That means Zen will be competing against Broadwell 14 nm for about 6 months before going up against Cannonlake.

Also, remember that Zen is a completely new architecture. So it has potential to provide significant performance gains compared to previous generations.

The Intel Core architecture base is from 2006, and will not see any similar revamp in the foreseeable future from what has been communicated by Intel so far.

You mean skylake :hmm:

I think we shouldn't expect too much from AMD, the only way anyone loyal to them won't be disappointed in the long run. The fact is with the engineering resources they have ($$) it's a miracle they're still alive in the desktop arena & for them to increase their market share their best bet IMO is to make an efficient design (like Jaguar) & slowly add more cores whilst tweaking it for better desktop/server usage.

 

[bullzz]

@Fjodor2001 - Broadwell is a 2015 product. Zen will compete against Skylake 4+4e in laptops, Skylake-K in mid range and Broadwell-E in high end. Considering 14nm would have matured in a year Intel would be ready for price and performance fight

 

[Idontcare]

^^^ Abwx makes a good point. (Remember the processor is on 14nm FinFET so 95 watts goes farther than what it would on 28nm/32nm planar.)

With that mentioned I am concerned this will be a design aimed most for MT.

Furthermore, what do we know of the Samsung 14nm process tech that GF is licensing? Is there a high drive current/high leakage variant of the process or it what most people would call a "SOC process tech"? (I am thinking mostly the later will be true so I wonder how high the frequency will be able to extend and at what cost to power consumption)

Samsung 14nm is high powered enough for Apple's and Samsung's needs, and that is about it. Still much better than what GloFo was originally intending to put into production with 14XM. But not powerful enough for the MPU and GPU chips, those will probably stay at TSMC for the fabless customers who are free to make that choice (not AMD).

 

[mrmt]

8 cores sounds a little worrying. That makes it sound like Zen is a Jaguar successor, not a Skylake competitor...

Would make sense, no? Their brave new world is the embedded market, which means low priority to raw performance. They didn't win the console contracts with derpdozer @ 5Ghz, they won with the >>cost-effective<< Jaguar. If they are going further down in the embedded rabbit hole, they will need an even more cost-effective part for this market, which means more IP-friendly design and low power consumption. The low raw performance is almost of a consequence.

In their current situation, if they could get more embedded designs, they could actually get rid of their consumer division for good. That would improve their balance sheet since the CPU division is the one draining their money. That would, of course, require a change in the R&D framework, but they don't have much of a choice.

 

[scannall]

Is the new architecture for the Zen something Jim Keller's return to AMD brought about? It should be interesting to see what comes out of this.

 

[ShintaiDK]

Is the new architecture for the Zen something Jim Keller's return to AMD brought about? It should be interesting to see what comes out of this.

He is about the ARM K12. Not that a single person changes much, if anything.

 

[AtenRa]

Would make sense, no? Their brave new world is the embedded market, which means low priority to raw performance. They didn't win the console contracts with derpdozer @ 5Ghz, they won with the >>cost-effective<< Jaguar. If they are going further down in the embedded rabbit hole, they will need an even more cost-effective part for this market, which means more IP-friendly design and low power consumption. The low raw performance is almost of a consequence.

In their current situation, if they could get more embedded designs, they could actually get rid of their consumer division for good. That would improve their balance sheet since the CPU division is the one draining their money. That would, of course, require a change in the R&D framework, but they don't have much of a choice.

You do realize that 14nm offers 60% less energy than 32/28nm. A 95-125W TDP 14nm FF 8-Core would have enormous performance by today's standards even if that was an 8-core Excavator design.

Just so people will understand what 14nm can bring, a direct port of the current Kaveri APU to 14nm and you would have the performance of desktop A10-7850K + 30-50% higher iGPU performance from the HBM at ~30W TDP with more than half the die size. And that is without any mArchitectural performance/power efficiency increases.

 

[stuff_me_good]

He is about the ARM K12. Not that a single person changes much, if anything.

I call that utter bullshit. In this world there are too many "I know better than you" people and so that is the reason why one single person makes a lot difference.

 

[Fjodor2001]

You mean skylake :hmm:

Yup, my bad. Skylake it is.

 

[Unoid]

Let me guess, they got 14nm(unknown product) from a transscript and then fabricated the rest. Sites sure go a long way for clickbaits these days.

I remember reading about how the current consoles would never have APU's in them from another online tech enthusiast. That never panned out either!

 

[ctsoth]

I'm not a big fan of 'what if' arguments, but I think these two possibilities are worth consideration.

1. 8 core modal lacks iGPU and is a desktop port of a server class chip, hopefully with aggressive turbo frequencies.

2. By 8 core they mean 8 thread :/

Finally, I don't think the following statement has a high probability, but I think it is possible that the brand new Zen (which they say will take the best of bulldozer uarch and jaguar) will also see some heavy inspiration from stars. I would be quite surprised if we get something 'entirely new.' I don't think taking inspiration from stars would be a bad thing either. I think if the stars program was never dropped AMD would have more competitive products on the market than they do today.

At this point though pretty much everything is speculation, and I am hesitant to invest too much mental energy contemplating the 'what ifs' of an almost completely unknown product.

I will say, that no matter what I am very happy to see a new AMD cpu coming to market. Maybe it will flop hard, maybe it will be competitive, at this point I can only assign probabilities. I am very excited though... CPU hardware has been really boring for a while.

 

[Fjodor2001]

Within 95 W TDP I think an APU with 8C/16T and iGPU it totally feasible on 14 nm, with a completely new uArch.