[Anandtech] Intel's Architecture Day 2018

[PeterScott]

Intel's Architecture Day 2018: The Future of Core, Intel GPUs, 10nm, and Hybrid x86

I know some bits of this were getting some mention in other threads, but it seems like it was big enough to warrant it's own discussion.

Highlights for me were a new architecture (finally) Sunny cove, that looks to have potential (4-5 wide allocation) to deliver Intels first real IPC improvement in MANY years, and also what should be a nice improvement in the the IGP, and a look at Intels multichip solutions(Foveros) and Intel Big-little.

The Q&A was also somewhat interesting, like where they admit being node-locked on their designs really messed them up, and they will be more node agnostic in the future.

Overall, more actual news than we have seen from Intel in some time.

 

[epsilon84]

They sure af care about TCO, and Intel will need ~2 years at least to beat Rome in it.

2 years since Zen versus 4 (well actually 5) years since SKL.
I sure af expect Intel to bring MORE, like, pull something from TGL or anything.

Skylake came out August 2015, so H2 2019 would make it 4 years. Regardless, we are long overdue for a new architecture.

I'm sure Intel would want more too, but I'll take 10 - 15% at this stage. That's also 10 - 15% higher than Zen 2, assuming Zen 2 achieves Skylake level IPC.

 

[rainy]

Skylake came out August 2015, so H2 2019 would make it 4 years. Regardless, we are long overdue for a new architecture.

I'm sure Intel would want more too, but I'll take 10 - 15% at this stage. That's also 10 - 15% higher than Zen 2, assuming Zen 2 achieves Skylake level IPC.

I presume he's sceptical, that Intel "glorious" 10nm will show up next year (even very late).

Btw, what if Sunny Cove IPC would be 5 - 10 percent higher than Skylake?
Hardly compelling after over 4 years.

 

[epsilon84]

I presume he's sceptical, that Intel "glorious" 10nm will show up next year (even very late).

Btw, what if Sunny Cove IPC would be 5 - 10 percent higher than Skylake?
Hardly compelling after over 4 years.

5 - 10% would indeed be disappointing, but not the end of the world. I would still expect Sunny Cove to be ahead of Zen 2 IPC wise in any case. It's a matter of how much. 5% would be a disappointment. 10% would maintain the current status quo. 15% would actually exceed the current IPC advantage between CFL and Zen+.

Don't get me wrong, increased IPC is great, but the power efficiency improvements are probably more important in the grand scheme of things. Intel needs the thermal headroom to increase core count and remain relevant.

If 10nm was to be delayed further that would probably be more disappointing than a 5-10% uplift.

 

[Carfax83]

In terms of IPC increase, I'm not so sure at all. Today, Skylake execution core is often wider than what the frontend and retire can support. It could be that most of the potential IPC increase of Skylake is basically gated on not being able to rename/retire more than 4 ops per clock. Widening those paths could alone mean a lot of extra IPC for tight loops that have great ILP. Two stores per clock is also a great boost, and one that I would not have expected. +50% L1d is also big.

I certainly expect substantially more IPC gain than from Skylake. To temper that somewhat, the widened rename and the increased L1d size both spell out "relaxed clock targets" to me.

Your posts are always informative! I hope you stick around on these forums

 

[PotatoWithEarsOnSide]

The big question marks appear to be over clocks. Both are expecting similar IPC improvements, though AMD are rumoured to also be upping clocks and core count, whereas Intel rumours suggests lower clocks. When combined, that's where you get Intel standing still, and AMD making a giant leap forward, likely into the lead.
It's not about being a fanboy; it's about combining all of the factors and not solely focusing on one improvement.
Are we really supposed to view an IPC uplift with increased clocks and increased cores in the same way as increased IPC with reduced clocks and same core count?
Sure, it's all still speculative at this point...on all quarters, but the strength of rumours surrounding Zen 2 is such that you kind of do have to sit up and take notice. There surely can't be any non-fanboys that don't want to see AMD make that leap forward...

 

[JoeRambo]

Are we really supposed to view an IPC uplift with increased clocks and increased cores in the same way as increased IPC with reduced clocks and same core count?

We are supposed to look at performance only. In the past AMD executives were very relaxed before C2D with "reduced clocks and same core counts" has completely demolished A64 X2. And today we have Apple Axx chips that have outright amazing performance even if clocks are low by desktop chip standards.

 

[DrMrLordX]

I would still expect Sunny Cove to be ahead of Zen 2 IPC wise in any case.

Too bad Sunny Cove will face Zen3 on the desktop and in the server room.

 

[epsilon84]

The big question marks appear to be over clocks. Both are expecting similar IPC improvements, though AMD are rumoured to also be upping clocks and core count, whereas Intel rumours suggests lower clocks. When combined, that's where you get Intel standing still, and AMD making a giant leap forward, likely into the lead.
It's not about being a fanboy; it's about combining all of the factors and not solely focusing on one improvement.
Are we really supposed to view an IPC uplift with increased clocks and increased cores in the same way as increased IPC with reduced clocks and same core count?
Sure, it's all still speculative at this point...on all quarters, but the strength of rumours surrounding Zen 2 is such that you kind of do have to sit up and take notice. There surely can't be any non-fanboys that don't want to see AMD make that leap forward...

What Intel rumours? Lower clocks? You mean this slide from over a year ago showing 10nm behind 14nm++? It's been 16 months since that slide, we're not even sure if it's the 'same' 10nm process anymore, since according to that slide 10nm (and even 10nm+!) should have been released by now. What 'version' of 10nm will we get? No one knows, though that won't stop people speculating of course.

When asked specifically if the first gen 10nm parts would achieve the same frequencies as 14nm++, the answer was "we're not going to release a product that's worse" so make of that what you will. It could mean 'not worse' in terms of frequency, or simply 'not worse' in terms of overall performance, which could mean a hypothetical 10nm chip @ 4.5GHz with +10% IPC would outperform a 9900K @ 4.7GHz, for example.

WR to core count, nowhere was core count mentioned, but what makes you think Intel will be stuck at 8 cores at 10nm? The power/thermal headroom will be there for additional core, unless you expect 10nm to offer worse performance/watt than 14nm++, which would be a disaster for Intel and would be 'Prescott V2'. I don't think that is going to happen though.

You're right that there is a LOT of speculation here, from both camps, especially with the estimated IPC improvements. I'm still not sure how Inteluser2000 came up with his 5 - 10% estimate from those slides alone.

Then with AMD, a 13% improvement on an unknown scientific workload (from a twitter post no less) was enough to send AMD fans into a frenzy. Then things went into overdrive with the 5GHz 16 core Zen 2 'leaks' which I personally think is unrealistic but AdoredTV maintains is plausible. Time will tell, as always.

 

[epsilon84]

Too bad Sunny Cove will face Zen3 on the desktop and in the server room.

I wasn't aware Sunny Cove was a 2020 product...

https://www.anandtech.com/show/13699/intel-architecture-day-2018-core-future-hybrid-x86

"Sunny Cove, built on 10nm, will come to market in 2019 and offer increased single-threaded performance, new instructions, and ‘improved scalability’. Intel went into more detail about the Sunny Cove microarchitecture, which is in the next part of this article. To avoid doubt, Sunny Cove will have AVX-512. We believe that these cores, when paired with Gen11 graphics, will be called Ice Lake.

Willow Cove looks like it will be a 2020 core design, most likely also on 10nm. Intel lists the highlights here as a cache redesign (which might mean L1/L2 adjustments), new transistor optimizations (manufacturing based), and additional security features, likely referring to further enhancements from new classes of side-channel attacks."

 

[NTMBK]

I wasn't aware Sunny Cove was a 2020 product...

https://www.anandtech.com/show/13699/intel-architecture-day-2018-core-future-hybrid-x86

"Sunny Cove, built on 10nm, will come to market in 2019 and offer increased single-threaded performance, new instructions, and ‘improved scalability’. Intel went into more detail about the Sunny Cove microarchitecture, which is in the next part of this article. To avoid doubt, Sunny Cove will have AVX-512. We believe that these cores, when paired with Gen11 graphics, will be called Ice Lake.

Willow Cove looks like it will be a 2020 core design, most likely also on 10nm. Intel lists the highlights here as a cache redesign (which might mean L1/L2 adjustments), new transistor optimizations (manufacturing based), and additional security features, likely referring to further enhancements from new classes of side-channel attacks."

You missed the "on the desktop and in the server room" bit of those quotes Everything Intel demo-ed was U-series or lower. I think it's still going to be a while before we see big 10nm chips.

 

[epsilon84]

You missed the "on the desktop and in the server room" bit of those quotes Everything Intel demo-ed was U-series or lower. I think it's still going to be a while before we see big 10nm chips.

That may very well be, but its still speculation at this point. We could also see desktop 10nm in late 2019. Did Intel state that the initial Sunny Cove release was mobile only? Is a tech demo on a U class chip on a node that is up to 12 months from launch an all encompassing view of Intel's plans with 10nm? I don't think so.

Of course, that won't stop people pretending they have a crystal ball...

 

[jpiniero]

That may very well be, but its still speculation at this point. We could also see desktop 10nm in late 2019. Did Intel state that the initial Sunny Cove release was mobile only? Is a tech demo on a U class chip on a node that is up to 12 months from launch an all encompassing view of Intel's plans with 10nm? I don't think so.

See the Comet Lake thread. That is what is coming in 2019 for desktop.

 

[Spartak]

People here still didnt get the memo. For high frequency 1st gen 10nm will perform worse compared to 14nm. That's why the whole effort to decouple Sunny Cove from Icelake took place.

Now that they have the tools, nothing is stopping 'the latest products to come to market on the best process technologies available at the time'. A literal quote since all you are sleeping.

The whole point of Sunny Cove is to not regress on clocks on the high end by using 14nm++(+).

The 2nd gen 10nm+ will have higher clocks, this is what we will see in 2020.

 

[Spartak]

First we have SemiAccurate anouncing 10nm dead. It wasnt dead according to Intel. Maybe it wasnt dead for mobile but it was indeed dead for desktop next year?

What if they retooled their 10nm fabs poised for ICL desktop to 14nm Sunny Cove? Hmmm...did they announce some 14nm tooling investments along those lines..?

Third, they literally spell out they succeeded in making ICL process independent as of now.

 

[DrMrLordX]

Third, they literally spell out they succeeded in making ICL process independent as of now.

. . . or maybe they spelled out how they made a mistake making Sunny Cove process dependent, and they're not going to make that mistake again . . . I mean really, what uarch did they claim was irrevocably process dependent? Cannonlake?

 

[HurleyBird]

Where on earth is this talk of 14nm Sunny Cove coming from? Pure speculation, or did I miss something substantial recently?

 

[jpiniero]

Where on earth is this talk of 14nm Sunny Cove coming from? Pure speculation, or did I miss something substantial recently?

At the architecture day Intel said they decoupled the process from the designs. Now it is very unlikely that Sunny Cove is part of Comet Lake, but it is possible that you could see Willow Cove for instance on some other node that's not Intel's 10 nm.

 

[DarkComplex]

You missed the "on the desktop and in the server room" bit of those quotes Everything Intel demo-ed was U-series or lower. I think it's still going to be a while before we see big 10nm chips.

Wrong, they had an Icelake-SP demo, see https://www.servethehome.com/first-pictures-of-intel-ice-lake-xeon-server-chips/

 

[jpiniero]

Wrong, they had an Icelake-SP demo, see https://www.servethehome.com/first-pictures-of-intel-ice-lake-xeon-server-chips/

Technically they aparently are still going to release it, but it's going to be a paper/FUD launch I figure. Trying to confuse the enterprise market while the hyperscalers load up on Rome.

The Icelake Client is different, you might be able to actually buy it even if it's only a token amount in the scheme of things compared to Comet and Whiskey.

 

[Dayman1225]

Technically they aparently are still going to release it, but it's going to be a paper/FUD launch I figure. Trying to confuse the enterprise market while the hyperscalers load up on Rome.

The Icelake Client is different, you might be able to actually buy it even if it's only a token amount in the scheme of things compared to Comet and Whiskey.

Even Charlie Dem from SemiAccurate says that Icelake SP will be out in volume in Q3 2020 (intel says 2020 for Icelake Server just in case any one is wondering)

 

[jpiniero]

Even Charlie Dem from SemiAccurate says that Icelake SP will be out in volume in Q3 2020 (intel says 2020 for Icelake Server just in case any one is wondering)

Intel would have to legitimately fix 10 nm for that to be the case though, given how big the dies must be. You can get away with it to an extent with small dies and some creative binning but not on a big die.

 

[PotatoWithEarsOnSide]

So to be clear, in 2019 Intel will be using new cores but on an old process...? i.e. stil on 14nm. They're doing this a) because they've decoupled the uarch from the node, and b) because 10nm is still going to be an inferior process to 14nm for them...?
Implication being that the IPC gain can be coupled with the 14nm frequencies, but we won't be seeing any power improvements until 10nm. Intel are saying they can offer the best of both worlds, minus improved power efficiency.

 

[NTMBK]

Wrong, they had an Icelake-SP demo, see https://www.servethehome.com/first-pictures-of-intel-ice-lake-xeon-server-chips/

That's not a demo, that's waving around a non-functional chip.

 

[coercitiv]

People here still didnt get the memo.

Now that they have the tools, nothing is stopping 'the latest products to come to market on the best process technologies available at the time'. A literal quote since all you are sleeping.

Is there a quote where they literally say they will offer Sunny Cove on 14nm, that they have this at hand?

So far, all the info I saw on decoupling arch from process was on future tense, a promise of safer practice:

Q: A lot of the CPU microarchitecture at Intel has been hamstrung by delays on process node technology. What went wrong, and what steps have been made to make sure it doesn't happen again?

R/J: Our products will be decoupled from our transistor capability. We have incredible IP at Intel, but it was all sitting in the 10nm process node. If we had had it on 14nm then we would have better performance on 14nm. We have a new method inside the company to decouple IP from the process technology. You must remember that customers buy the product, not a transistor family. It’s the same transformation AMD had to go through to change the design methodology when they were struggling. At Apple it was called the ‘bus’ method.

M: This is a function of how we as a company used to think about process node technologies. It was a frame tick (limiting factor) for how the company moved forward. We've learned a lot about how this worked with 14nm. We now have to make sure that our IP is not node-locked. The ability to have portability of IP across multiple nodes is great for contingency planning. We will continue to take aggressive risks in our designs, but we also will have contingency. We need to have as much of a seamless roadmap as possible in case those contingencies are needed, and need to make sure they are executed on ASAP if needed to keep the customer expectations in line. You will see future node technologies, such as 10/7, have much more overlap than before to keep the designs fluid. Our product portfolio on 14nm could have been much better if our product designs were not node-locked to 10nm.

R: In the future there will be no transistor left behind, no customer left behind, and no IP left behind.

All this points towards future designs and what they're announcing here is a management/planning method, not a port of Sunny Cove to 14nm+*/-+

 

[exquisitechar]

AMD gets a 10 - 15% uplift and people are hanging from the chandeliers. Intel gets a 10 - 15% uplift and they are doomed.

People expect Zen 2 to have a decent clockspeed increase too.

Also, what Bondrewd said.