Fudzilla: New AMD Zen APU boasts up to 16 cores (plus Greenland GPU with HBM)

[Arachnotronic]

I guess I'll have to dig up some of the review sites then, but since it's not official PR from Intel it'd just be a waste of time. So now how did anyone come to the conclusion that Intel's 14nm is better than the competition, other than Intel proving that beyond contention?

You can paint all the electrical parameters, from the competition, in bad light but since the underlying Silicon is different (to the point of them being designed for various markets as well as the underlying uarch & ISA) you cannot decisively award victory to one or the other unless you have removed every possible anomaly or difference that makes each chip unique.

With all due respect, AMD was talking about how they are running test chips on FinFET nodes today and that they're going to launch these 14/16nm parts beginning in 2016 (I'd say late 2016 for the FX chip; sometime in 2017 for the APUs).

Intel is shipping 14nm processors today by the truckload in tablets (Cherry Trail), low cost PCs (Braswell), mainstream notebooks (Broadwell-U 2+2 and 2+3), and even microserver (Broadwell-DE).

I'd bet you within a month or so, Intel will formally launch Broadwell-H (4 + 3e), and then in the August/September timeframe, you'll probably see Skylake-S (4+2, 2+2). Then, by Q1 2016, expect Broadwell-E/EP, then by Q2/Q3 Broadwell-EX. I also wouldn't be surprised if we saw Skylake-DE for microservers in early 2016 as well.

The notion that some people try to paint that AMD is just "catching up" to Intel on process and that Intel is just going to idly twiddle its thumbs while it "cedes" all sorts of "crowns" to AMD just doesn't pass the common sense test -- at least not for me.

 

[R0H1T]

With all due respect, AMD was talking about how they are running test chips on FinFET nodes today and that they're going to launch these 14/16nm parts beginning in 2016 (I'd say late 2016 for the FX chip; sometime in 2017 for the APUs).

Intel is shipping 14nm processors today by the truckload in tablets (Cherry Trail), low cost PCs (Braswell), mainstream notebooks (Broadwell-U 2+2 and 2+3), and even microserver (Broadwell-DE).

I'd bet you within a month or so, Intel will formally launch Broadwell-H (4 + 3e), and then in the August/September timeframe, you'll probably see Skylake-S (4+2, 2+2). Then, by Q1 2016, expect Broadwell-E/EP, then by Q2/Q3 Broadwell-EX. I also wouldn't be surprised if we saw Skylake-DE for microservers in early 2016 as well.

The notion that some people try to paint that AMD is just "catching up" to Intel on process and that Intel is just going to idly twiddle its thumbs while it "cedes" all sorts of "crowns" to AMD just doesn't pass the common sense test -- at least not for me.

Conversely the single biggest disadvantage that was drowning AMD, besides BD of course, is now gone so expect competitiveness to return.

Intel is a moving target but AMD isn't standing still either, they've taken one giant leap of faith with Zen & 14nm but that's all there's to it. Everything else is pretty much like shooting in the dark at best or selling bridges at worst

 

[Abwx]

With all due respect, AMD was talking about how they are running test chips on FinFET nodes today and that they're going to launch these 14/16nm parts beginning in 2016 (I'd say late 2016 for the FX chip; sometime in 2017 for the APUs).

For sane analysis capability one should make abstraction of his stock holder quality or eventual financial advising activities, otherwise posts will be tainted with what are some kind of wishfull thoughts.

 

[Arachnotronic]

Conversely the single biggest disadvantage that was drowning AMD, besides BD of course, is now gone so expect competitiveness to return.
P

What disadvantage do you refer to here? Process technology?

You do realize that the Zen APUs will very likely compete against Cannonlake on 10nm. This will be a very difficult performance/watt barrier to break through.

In the late 2016 timeframe, AMD will be able to pit an 8 core/16 thread Zen against a 14nm Intel Skylake-E, which should be a "fairer" fight. That said, I still remain unconvinced (not saying they won't for sure -- I don't have a crystal ball) that AMD's Zen will bring this huge IPC uplift, deliver similar clocks, and similar power consumption to a many-core Skylake chip launched in that same timeframe.

I personally think Intel's "14nm" should be able to deliver noticably better performance/watt than Samsung/GloFo "14nm," especially for high-performance devices.

We'll see, though. I look forward to the reviews.

 

[R0H1T]

What disadvantage do you refer to here? Process technology?

You do realize that the Zen APUs will very likely compete against Cannonlake on 10nm. This will be a very difficult performance/watt barrier to break through.

In the late 2016 timeframe, AMD will be able to pit an 8 core/16 thread Zen against a 14nm Intel Skylake-E, which should be a "fairer" fight. That said, I still remain unconvinced (not saying they won't for sure -- I don't have a crystal ball) that AMD's Zen will bring this huge IPC uplift, deliver similar clocks, and similar power consumption to a many-core Skylake chip launched in that same timeframe.

I personally think Intel's "14nm" should be able to deliver noticably better performance/watt than Samsung/GloFo "14nm," especially for high-performance devices.

We'll see, though. I look forward to the reviews.

No, as you pointed out later, Zen will compete against Skylake at 14nm for a good 6~12 months since I do expect Cannonlake to be delayed, just like Broadwell was. This is the window AMD needs to target & get Zen anywhere & everywhere possible even at wafer thin margins, hey can't be worse than BD, then build on that momentum to further improve Zen, K12 & of course add more cores.

A 32c/64t monster chip, based on some of the earlier leaks, is going to get them some high end Xeon buyers, the question is if they can release it in time & if it's efficient enough for the target market. Regardless of the earlier debate I've had over this issue I do expect a competent cooling solution to handle something that's ~10% less efficient, especially since the load will not be 100% all the time (24/7) unless anyone else can prove otherwise.

 

[ShintaiDK]

Zens core count sure goes up fast. 8 cores, 16 cores, 32 cores.

I think someone lost realism along the way.

 

[Arachnotronic]

No, as you pointed out later, Zen will compete against Skylake at 14nm for a good 6~12 months since I do expect Cannonlake to be delayed, just like Broadwell was. This is the window AMD needs to target & get Zen anywhere & everywhere possible even at wafer thin margins, hey can't be worse than BD, then build on that momentum to further improve Zen, K12 & of course add more cores.

Cannonlake was originally aimed at Q2 2016. I am assuming a delay of a broad roll-out into early 2017, with perhaps the initial Core M models showing up in late Q3/early Q4 2016.

 

[R0H1T]

Zens core count sure goes up fast. 8 cores, 16 cores, 32 cores.

I think someone lost realism along the way.

8 cores was desktop FX wasn't it, what's the max core count in servers? Didn't see yesterday's presentation so not sure if it was revealed.

Cannonlake was originally aimed at Q2 2016. I am assuming a delay of a broad roll-out into early 2017, with perhaps the initial Core M models showing up in late Q3/early Q4 2016. [/IMG]

Don't the server chips follow the PC client rollout, more testing & validation, lagging by 6~12 months ?

 

[MrTeal]

Cannonlake was originally aimed at Q2 2016. I am assuming a delay of a broad roll-out into early 2017, with perhaps the initial Core M models showing up in late Q3/early Q4 2016.

Where did you find that roadmap?

 

[Arachnotronic]

8 cores was desktop FX wasn't it, what's the max core count in servers? Didn't see yesterday's presentation so not sure if it was revealed.Don't the server chips follow the PC client rollout, more testing & validation, by 6~12 months ?

Don't expect Cannonlake for servers until late 2017; like I said in my post, I think the 8 core FX will compete against Skylake-E.

 

[Arachnotronic]

Where did you find that roadmap?

AnandTech user III-V posted it on another thread a while ago.

 

[R0H1T]

Don't expect Cannonlake for servers until late 2017; like I said in my post, I think the 8 core FX will compete against Skylake-E.

I though we were talking servers, at least I was, anyway the timeline overlaps so the FX will also have it's shot at the top end desktop market.

Just to add to this, a true eight core high performance chip will sway many buyers towards AMD especially since they haven't done much recently in the desktop arena either; if priced right.

AMD is just launching its Seattle chip in 2H 2015. Do you honestly think Zen will show up in a 2P (or greater) server configuration in late 2016?

I would argue that a Cannonlake-based Xeon D could easily come in early 2017; the additional complexities associated with validating a multiway CPU configuration don't apply to this class of product, nor do these kinds of chips require the kind of RAS features that a Xeon E5/E7 would.

Depends on where AMD's priority lie, if Zen was launched specifically to enter or target that market then no one's stopping AMD from selling the high end SKU's starting late 2016.

 

[Arachnotronic]

I though we were talking servers, at least I was anyway the timeline overlaps so the FX will also have it's shot at the top end desktop market.

AMD is just launching its Seattle chip in 2H 2015. Do you honestly think Zen will show up in a 2P (or greater) server configuration in late 2016?

I would argue that a Cannonlake-based Xeon D could easily come in early 2017; the additional complexities associated with validating a multiway CPU configuration don't apply to this class of product, nor do these kinds of chips require the kind of RAS features that a Xeon E5/E7 would.

 

[Fjodor2001]

Cannonlake was originally aimed at Q2 2016. I am assuming a delay of a broad roll-out into early 2017, with perhaps the initial Core M models showing up in late Q3/early Q4 2016.

Interesting that there's a Skylake-Refresh on the roadmap for Desktop. Does it mean there will be no full desktop line of Cannonlake for Desktop, just like on Broadwell? And no full desktop replacement until 2018 or so when the successor to Cannonlake arrives?

If so Zen will only have to go up against Skylake until that. Even the Zen based APUs arriving in 2017 only have to compete with Skylake.

 

[Sweepr]

Interesting that there's a Skylake-Refresh on the roadmap for Desktop. Does it mean there will be no full desktop line of Cannonlake for Desktop, just like on Broadwell? And no full desktop replacement until 2018 or so when the successor to Cannonlake arrives?

If so Zen will only have to go up against Skylake until that. Even the Zen based APUs arriving in 2017 only have to compete with Skylake.

You wish, just like your 2016/2017 launch predictions for Skylake-S. What this means is that they might cancel desktop Cannonlake or release a limited number of SKUs just like they did with Broadwell. That doesn't mean desktop Icelake got delayed.

That is, assuming that this Roadmap which doesn't even include Braswell and Broadwell-E isn't fake.

 

[DrMrLordX]

Zens core count sure goes up fast. 8 cores, 16 cores, 32 cores.

I think someone lost realism along the way.

What do you mean? The first Zen leak was for 8C/16T, 95W. Later on, there was a semiaccurate poster reporting that an AMD rep in Japan mentioned a 16C/32T chip, and then the server APU. Later on there was talk that the 16C chip would be MCM. Nothing AMD has said to date contradicts any of that, other than the fact that the MCM chip probably won't hit until 2017, or at least not at the same time that they launch the 8C chip.

Everything from that series of leaks going back to March of this year has been borne out by AMD slides. AMD has confirmed the "big" server APU (for 2017). The only thing that has proven untrue were recent fabrications/exaggerations of AMD's Zen timeline in smaller CPUs, all of which contradicted earlier leaks that made no such claims. It looks like the original Bristol Ridge leaks were more accurate.

16C/32T doesn't look unrealistic at all.

 

[Fjodor2001]

You wish, just like your 2016/2017 launch predictions for Skylake-S. What this means is that they might cancel desktop Cannonlake or release a limited number of SKUs just like they did with Broadwell. That doesn't mean desktop Icelake got delayed.

That is, assuming that this Roadmap which doesn't even include Braswell and Broadwell-E isn't fake.

2016 for Skylake was not that far off. It'll be 2015H2 with some models in 2016.

I'm not assuming Icelake desktop will be delayed, I'm just saying it's likely to arrive in 2018 or so. Cannonlake is expected 2017, but there won't be a full desktop lineup. So AMD Zen will effectively compete with Skylake until 2018 or so. By that tile these will likely even be a Zen successor around.

 

[Sweepr]

2016 for Skylake was not that far off. It'll be 2015H2 with some models in 2016.

I'm not assuming Icelake desktop will be delayed, I'm just saying it's likely to arrive in 2018 or so. Cannonlake is expected 2017, but there won't be a full desktop lineup. So AMD Zen will effectively compete with Skylake until 2018 or so. By that tile these will likely even be a Zen successor around.

August 2015 is quite far from your H2-2016/2017 predictions for desktop Skylake. So you not only think there won't be a full desktop lineup with Cannonlake but desktop Icelake also won't arrive till 2018? In your predictions Zen dominates the mainstream desktops next year while it will take 3 years for Intel to release Skylake-S successor? Lol.

 

[Fjodor2001]

August 2015 is quite far from your H2-2016/2017 predictions for desktop Skylake. So you not only think there won't be a full desktop lineup with Cannonlake but desktop Icelake also won't arrive till 2018? In your predictions Zen dominates the mainstream desktops next year while it will take 3 years for Intel to release Skylake-S successor? Lol.

I said likely 2016, and it'll be 2015H2. So 1H off, i.e. very close to what I predicted. Also, not all models will arrive in 2015H2, some will arrive in 2016.

Then on Cannonlake, according to Wikipedia:

"Cannonlake (formerly Skymont) is Intel's codename for the 10 nanometer die shrink of Intel's Skylake microarchitecture, expected to be released in 2017."

If Cannonlake is expected to arrive in 2017, do you really think Icelake will arrive at the same time? I'd say 2018 is more likely. But who knows for sure...

 

[ShintaiDK]

What do you base those assumptions on? If we go by the latest track record from SB->B-K, desktop S will be released ~16 months after B-K, so that will be around 2016Q2/Q3.

Thats more than 1H off...

And you even talked about summer 2017.

 

[Sweepr]

"Cannonlake (formerly Skymont) is Intel's codename for the 10 nanometer die shrink of Intel's Skylake microarchitecture, expected to be released in 2017."

If Cannonlake is expected to arrive in 2017, do you really think Icelake will arrive at the same time? I'd say 2018 is more likely. But who knows for sure...

Most likely early 2017 for mobile Cannonlake, and yes, I totally expect them to introduce desktop Icelake sometime in mid-2017. I think it will arrive less than a year after the first Zen 8C parts and not too much time after the first Zen-based APUs.

Remmember you denied till the last minute the possibility of a limited Broadwell desktop launch less than a quarter away from Skylake.

 

[Fjodor2001]

Thats more than 1H off...

And you even talked about summer 2017.

2016-2017 includes January 2016 (and I said most likely 2016). So 1Q to 1H off. And the complete Skylake lineup will not be released until 2016 anyway. So the prediction was very close.

Maybe you should think over your own predictions:

* Intel's process tech lead will expand.
* There will be no APU in the PS4/XBONE.
* AMD will be a Via type company in 2015/2016.
* AMD will exit the server market.
* Zen will be a cat core replacement.
* AMD will exit the high performance desktop segment.

The list goes on...

 

[Fjodor2001]

Most likely early 2017 for mobile Cannonlake, and yes, I totally expect them to introduce desktop Icelake sometime in mid-2017.

What do you base that assumption on?

Remmember you denied till the last minute the possibility of a limited Broadwell desktop launch less than a quarter away from Skylake.

It's not a "real" full desktop release. It's a forced extremely limited specialized SKU release, only to keep the promise of a Broadwell desktop release.

Also, it's funny you now consider it a desktop release. When estimating Skylake desktop performance increase, you keep comparing to Haswell only to avoid mentioning the embarrassing 5% yearly Intel performance increase...

Compare that to AMD Zen's 40% IPC increase. And then we're not even including any frequency increase going from 28 -> 14 nm! That's like close to all the aggregated performance increase from Nahalem->Skylake, and the Intel crowd calls it a failure. How sane is that!

 

[Arachnotronic]

What do you base that assumption on?

It's about two years after Skylake desktop. Seems like a very sensible prediction to me.

 

[Sweepr]

What do you base that assumption on?

The fact that Broadwell delays didn't affect Skylake's desktop launch.
What do you base your 2018 launch prediction on?
In the last few pages you predicted AMD will demolish Intel's mainstream offerings, care to share what do you base that assumption on and if you seriously think Intel will stand still if that happens?