[Techpowerup] AMD "Zen" CPU Prototypes Tested, "Meet all Expectations"

[Edrick]

Broadwell-E was pushed to Q2-Q3 2016. Making a Skylake-E H1 2017 release highly questionable.

Why? They released Skylake just a few short months after Broadwell. Intel wants the Purley platform released as it will offer higher revenue for them over Broadwell-E.

 

[Fjodor2001]

I bet it will, and it will put AMD in its place again once it launches.

In that case so should Haswell-E and Broadwell-E too, because Skylake-E won't bring much IPC increase over those anyway.

But they are too expensive anyway. $1000 for an 8 core.

 

[Fjodor2001]

Why? They released Skylake just a few short months after Broadwell.

On desktop, yes. But that was just because desktop Broadwell only got the tiny niche SKUs 5675/5775C. I.e. desktop Broadwell was more or less skipped.

On mobile, Skylake lauched much later than Broadwell. Because there we got a full Broadwell lineup.

 

[Sweepr]

In that case so should Haswell-E and Broadwell-E too, because Skylake-E won't bring much IPC increase over those anyway.

But they are too expensive anyway. $1000 for an 8 core.

Source please. You don't know nothing about Skylake-E. It could very well offer more than 8 cores, servers will have up to 28 cores. It might even pack upgraded Skylake cores, we already know it's not the same as current mainstream chips.

 

[jpiniero]

I'm sure Intel would love to get Purley out ASAP, 3DXP DIMMs should be a game-changer and I'm sure there's great interest from their cloud customers to check it out.

Come to think of it, 3DXP DIMM being Intel-exclusive is a big deal. It's obviously going to make it that much harder for AMD on Zen Server to get any traction.

 

[dark zero]

I'm sure Intel would love to get Purley out ASAP, 3DXP DIMMs should be a game-changer and I'm sure there's great interest from their cloud customers to check it out.

Come to think of it, 3DXP DIMM being Intel-exclusive is a big deal. It's obviously going to make it that much harder for AMD on Zen Server to get any traction.

AMD and nVIDIA can counter Intel with HBM.
A memory wars will start in that generation.
The most interesting thing now is that all the chips from 2016 on beyond will be at 14 nm and that seems to include Bristol Ridge and Stoney Ridge.

Also... If they move to LGA, they might have a chance

 

[frozentundra123456]

Well, even if Skylake E comes out mid 2017, I think Zen, in decent volumes, could easily be delayed till this time frame, especially for the desktop. I mean, all you have to do is look at Intel 14nm to see the difference between stated availability and true availability in the consumer channel. Zen is already optimistically supposed to arrive Q4 2016, and I would expect early allocation to go to servers.

I think it could easily be just token availability and quite a bit later for full availability in the retail channels. Actually, this has pretty much been the case for a lot of AMD "launches" (eg Carizzo) for even derivatives of existing products. I cant imagine a totally new architecture on a new process will magically appear in volume on the "release date".

 

[jpiniero]

Zen is already optimistically supposed to arrive Q4 2016, and I would expect early allocation to go to servers.

I'm sure it will be for enthusiasts (gamers) first, since they have at least a decent shot of selling what can be made. Yeah the volume will be tiny but the enthusiasts market is tiny so they should be able to handle the small volume. You're talking about a process that would be basically 18+ months old by the end of 2016. Granted it's GloFo's approximation of it.

 

[fourdegrees11]

So is there a chance that Zen not having IGP can give it room to come in a lot better than some might expect clock for clock?

 

[looncraz]

So is there a chance that Zen not having IGP can give it room to come in a lot better than some might expect clock for clock?

No, it just effects die size and power draw.

Zen will probably be a ~4GHz part, the question is if there will be any room left for overclocking.

There are some who are concerned about being able to even reach 4GHz, but I think AMD knows the tricks to reaching high clocks :thumbsup:

The process itself is, technically, more amenable to high clocks (transistors switch quite a bit faster, and have lower leakage). The problem is getting current to where you need it with the smaller traces and higher line impedance so that you can keep transistors switching fast enough to meet the time target (*just* before the next clock pulse).

If you have that, clock speeds are simply determined by the time it takes the longest-running stage of the CPU to complete its operations and have the result on the output wires with a usable signal strength.

 

[frozentundra123456]

I seriously doubt they will reach 4ghz at 95 watts on an 8 core part.

Maybe on a quad core, but even on 14nm intel quads are 90+ watts at 4ghz.

 

[Arachnotronic]

No it wasn't. Q2 maybe, Q3 comes from your imagination.

Broadwell-EP is a Q1 2016 affair per all leaks and available information. This would also fit the buying patterns of big data center customers.

Q2'16 release or later is just wishful thinking among the AMD fans, IMO.

 

[AtenRa]

Broadwell-EP is a Q1 2016 affair per all leaks and available information. This would also fit the buying patterns of big data center customers.

Q2'16 release or later is just wishful thinking among the AMD fans, IMO.

Broadwell-E production was pushed back to Feb-March 2016. So the release is going to be med to late Q2, early Q3 at the latest.

 

[myocardia]

Trust the saleswoman?

Really?

Read what you quoted.

If Zen actually averages anywhere near 25% higher IPC than Excavator, with anywhere near the efficiency that it should have,

Seriously, dude...do you even READ what you quote?

 

[myocardia]

We simply have no idea about Zen at this point.

Of course we don't. Why exactly did you think I used the word "if"?

 

[Vesku]

Based on GF just now doing a press release about AMD products starting validation testing on their 14nm LPP and taking into account the typical foundry enthusiasm I think AMD will be very lucky to get any Zen to retail before 1H2017.

 

[Phynaz]

Broadwell-E was pushed to Q2-Q3 2016. Making a Skylake-E H1 2017 release highly questionable.

The Skylake schdule has been shown to be unafected by Broadwell.

 

[myocardia]

I seriously doubt they will reach 4ghz at 95 watts on an 8 core part.

Maybe on a quad core, but even on 14nm intel quads are 90+ watts at 4ghz.

No doubt about that. Anyone saying a 14nm 8 core/16 thread part from either vendor will be 4 Ghz and also under 100 watts is delirious. The idea of AMD being able to do it, on their first attempt at SMT, is nothing other than hilarious.

 

[AtenRa]

The Skylake schdule has been shown to be unafected by Broadwell.

Even if Broadwell-E delay will not push back Skylake-E, Purley platform was originally scheduled for H2 2017 by Intel.

 

[Sweepr]

Even if Broadwell-E delay will not push back Skylake-E, Purley platform was originally scheduled for H2 2017 by Intel.

Unfortunately for AMD, it wasn't.

Xeon E5-2600 v5 (Purley) is H1-2017.

 

[AtenRa]

Unfortunately for AMD, it wasn't.

Xeon E5-2600 v5 (Purley) is H1-2017.

That is not H1 2017

 

[Sweepr]

That is not H1 2017

Your roadmap is not only older but also doesn't mention H2-2017 anywhere.

Nice try though.

 

[AtenRa]

Your roadmap is not only older but also doesn't mention H2-2017 anywhere.

Nice try though.

So where is the new roadmap having Purley at H1 2017 ??

 

[Sweepr]

So where is the new roadmap having Purley at H1 2017 ??

Above, Purley systems in H1-2017. Still waiting for your H2-2017 imaginary roadmap. Don't make me laugh with that May roadmap divided by years.

 

[Fjodor2001]

It might even pack upgraded Skylake cores

From where have you heard that Intel would develop a new Skylake core for Skylake-E that significantly differs from the one in 1151 desktop Skylake? They've not done that before on the E series.

And do you really think it would have substantially higher IPC than 1151 Skylake anyway? Intel have been struggling to just get 5% annual performance increase between CPU generations, and that's when going from one generation to the next. So how could then a tweaked Skylake core for Skylake-E really all of a sudden provide a significant jump in IPC over that?