[Anandtech] Intel's Architecture Day 2018

[DrMrLordX]

Because Intel bad.

Don't be so simplistic. It's been years since Intel managed more than 5-10% (variable) on top of a last-gen core. I would not be at all surprised if Sunny Cove turns out to be 5-10% over Cannonlake.

 

[Arachnotronic]

Don't be so simplistic. It's been years since Intel managed more than 5-10% (variable) on top of a last-gen core. I would not be at all surprised if Sunny Cove turns out to be 5-10% over Cannonlake.

Just look at the changes in the Sunny Cove architecture.

Also, by this logic, Zen should’ve been a failure because Piledriver, Steamroller, And Excavator were all underwhelming.

Again, there’s nothing in this architectural disclosure that should warrant your pessimism.

 

[Markfw]

Just look at the changes in the Sunny Cove architecture.

Also, by this logic, Zen should’ve been a failure because Piledriver, Steamroller, And Excavator were all underwhelming.

Again, there’s nothing in this architectural disclosure that should warrant your pessimism.

There is also nothing to make me optimistic until I see a benchmark

 

[jpiniero]

Don't be so simplistic. It's been years since Intel managed more than 5-10% (variable) on top of a last-gen core. I would not be at all surprised if Sunny Cove turns out to be 5-10% over Cannonlake.

The thing you have to remember is that Sunny Cove isn't the core after Cannonlake; it's the one after that. So it's essentially 3 generations removed from Skylake. 10-15% doesn't seem unrealistic.

 

[maddie]

Just look at the changes in the Sunny Cove architecture.

Also, by this logic, Zen should’ve been a failure because Piledriver, Steamroller, And Excavator were all underwhelming.

Again, there’s nothing in this architectural disclosure that should warrant your pessimism.

5%-10% improvement isn't pessimism. Your optimism is a faith based argument until facts present themselves.

 

[Arachnotronic]

5%-10% improvement isn't pessimism. Your optimism is a faith based argument until facts present themselves.

Facts have presented themselves. There's been a significant architecture disclosure, there have been some GB4 scores that point to a pretty healthy IPC gain.

 

[Tuna-Fish]

More to the point, you are advocating for Intel to do a radical redesign to beat Apple, when Apple designs are converging on the exact same design ideas that Intel uses.

Bollocks. People keep saying, "wait until Apple slows down. Wait until they have picked all the low-hanging fruit". They've been saying that since the A8. Hasn't happened yet. The A12x beats my glorious Ryzen 1800x @ 4.0 GHz in Geekbench single-thread.

Sorry, that was not what I meant at all. What I meant is: "The specific techniques Apple is using to improve their CPUs are the same that Intel has done." This should not be taken as derogatory of Apple -- being the second guy to implement a really good branch predictor, or L1 cache, or really fast register rename, is not substantially easier than being the first guy. However, Apple is not doing anything radically new, they are just doing the same things Intel has done and executing better on their design teams, and using a better process. The thing Intel needs to do to compete is not to figure out something radical -- that is typically how they lose, just think of Pentium 4, IA64 and iAPX-432. What they need to do to compete better is start executing better on refining the tech they have, and move to a better process.

For proof that Apple is fundamentally going for the same solution space as Intel, just look at any microarchitectural deep dive of an Apple core.

Don't be so simplistic. It's been years since Intel managed more than 5-10% (variable) on top of a last-gen core. I would not be at all surprised if Sunny Cove turns out to be 5-10% over Cannonlake.

Are you or are you not acknowledging that Intel execs are on the record (in this very article we are commenting on!) stating that over the past few years they have developed new microarchitectural IP, but which has not been shipped to customers because they designed it all for 10nm, and since 10nm got so delayed they just shipped the same cores over and over? Please just give a straight yes or no answer.

Because if you acknowledge that, it should be very easy to understand why they have not shipped any kind of real improvements for years, and yet why there is reason to expect that they might ship that real improvement with Sunny Cove.

 

[Brunnis]

The thing you have to remember is that Sunny Cove isn't the core after Cannonlake; it's the one after that. So it's essentially 3 generations removed from Skylake. 10-15% doesn't seem unrealistic.

Where did you see that? I was under the impression that Sunny Cove is what was always supposed to go into Ice Lake, which would make it two generations newer than Skylake, not three.

 

[jpiniero]

Where did you see that? I was under the impression that Sunny Cove is what was always supposed to go into Ice Lake, which would make it two generations newer than Skylake, not three.

The rumors. Essentially the rumor is that they pulled in the changes from Tiger Lake's core into Icelake because they were able to complete the design due to all the additional time given because of the 10 nm delays. They are calling that core Sunny Cove.

 

[mikk]

Here is a interview by Gamers Nexus concerning Sunny Cove with Intel's Chief Core Architect, Ronak Singhal. Not a particularly insightful interview for industry professionals or engineers, but for non techies, it's quite useful

This interview is new to me.

2:10: "With Sunny Cove the first products will be on 10nm"
8:30: "Remember that the first products with Sunny Cove will be on 10nm, so you get the advantage of the shrink factor by moving to the new process technology"

So he says first Sunny Cove products will be out on 10nm (probably ICL-U), interestingly he suggested that Intel will also release Sunny Cove on an older process technology some time later. I wonder if the rumoured CML-S on 14nm is based on Sunny Cove cores.

 

[Brunnis]

The rumors. Essentially the rumor is that they pulled in the changes from Tiger Lake's core into Icelake because they were able to complete the design due to all the additional time given because of the 10 nm delays. They are calling that core Sunny Cove.

Interesting, and not entirely implausible, of course. Got any links?

 

[jpiniero]

Interesting, and not entirely implausible, of course. Got any links?

Nothing specific, but this 2017 Geekbench result I suspect is using the old core:
https://browser.geekbench.com/v4/cpu/2400363

Now it's flagged as inaccurate because of the outrageous memory copy and memory latency score but I suspect that's because it was a test sample using the on package memory (which Lakefield is using) and that broke those two tests.

 

[Arachnotronic]

Nothing specific, but this 2017 Geekbench result I suspect is using the old core:
https://browser.geekbench.com/v4/cpu/2400363

Now it's flagged as inaccurate because of the outrageous memory copy and memory latency score but I suspect that's because it was a test sample using the on package memory (which Lakefield is using) and that broke those two tests.

Or it could've just had half the L2$ disabled. Or the benchmark could've been reading it wrong. There are a lot of possibilities here beyond that Sunny Cove was dramatically modified.

 

[DrMrLordX]

For proof that Apple is fundamentally going for the same solution space as Intel, just look at any microarchitectural deep dive of an Apple core.

And that folds back into what I was saying in the first place: Intel is not doing very well "going for the same solution". For whatever reason, they aren't getting it done. So they need to either change their course or give up. Hoping for them to spontaneously get better after years of failure (and I'm more pointing at Sandy -> Skylake here, so stop trying to obfuscate) is a fools errand.

Are you or are you not acknowledging that Intel execs are on the record (in this very article we are commenting on!) stating that over the past few years they have developed new microarchitectural IP, but which has not been shipped to customers because they designed it all for 10nm, and since 10nm got so delayed they just shipped the same cores over and over? Please just give a straight yes or no answer.

Because if you acknowledge that, it should be very easy to understand why they have not shipped any kind of real improvements for years, and yet why there is reason to expect that they might ship that real improvement with Sunny Cove.

I am fully aware of that. I'm also aware of their recent developments with Core before they hit a stagnation point. So yes, I'm aware. Again, stop trying to obfuscate. They were already stagnant before they got stuck on Skylake. That "real improvement" will probably be more of what we got before they were stuck on process. Consider me underwhelmed.

Where did you see that? I was under the impression that Sunny Cove is what was always supposed to go into Ice Lake, which would make it two generations newer than Skylake, not three.

There seems to be some confusion about what would have been the orderly transition away from Skylake had 10nm not failed so massively. Generally, it was considered that there would be Cannonlake, followed by IceLake, and then TigerLake. Three generations.

We've already seen Cannonlake, and it was . . . to the best of my knowledge, maybe 3% faster clock-per-clock than Skylake? It is not known what improvements TigerLake would have offered over IceLake, but it is generally prudent to assume that IceLake would have been 5-10% faster than Cannonlake, albeit (maybe) at lower clockspeeds than Skylake/Kabylake due to process upheaval (think of the transition from Sandy to Haswell).

If you really want to drink the Intel KoolAid/Wylers (heh), maybe there would have been a 10% uplift from Cannon-Ice-Tiger, producing a TigerLake core that is 33% faster clock-per-clock than Skylake. Cannonlake already broke that chain showing less-than-thrilling gains on the few rare commercial products that used it. So best case is maybe a 24% IPC uplift from Skylake/Kabylake/Cannonlake to whatever finally uses Sunny Cove. And I'm sorry to say that, looking at Intel's release schedule, that 24% may not be all that great if it comes at reduced clocks (which it probably will). I mean, it's better than more Skylake but . . .

 

[PeterScott]

Sorry, that was not what I meant at all. What I meant is: "The specific techniques Apple is using to improve their CPUs are the same that Intel has done." This should not be taken as derogatory of Apple -- being the second guy to implement a really good branch predictor, or L1 cache, or really fast register rename, is not substantially easier than being the first guy. However, Apple is not doing anything radically new, they are just doing the same things Intel has done and executing better on their design teams, and using a better process. The thing Intel needs to do to compete is not to figure out something radical -- that is typically how they lose, just think of Pentium 4, IA64 and iAPX-432. What they need to do to compete better is start executing better on refining the tech they have, and move to a better process.

For proof that Apple is fundamentally going for the same solution space as Intel, just look at any microarchitectural deep dive of an Apple core.



Are you or are you not acknowledging that Intel execs are on the record (in this very article we are commenting on!) stating that over the past few years they have developed new microarchitectural IP, but which has not been shipped to customers because they designed it all for 10nm, and since 10nm got so delayed they just shipped the same cores over and over? Please just give a straight yes or no answer.

Because if you acknowledge that, it should be very easy to understand why they have not shipped any kind of real improvements for years, and yet why there is reason to expect that they might ship that real improvement with Sunny Cove.

I disagree. Being the first to develop/prove viability of a lot of new techniques is usually significantly harder than doing the same after someone already blazed the trail.

 

[jpiniero]

The reduced max clocks isn't likely to be an issue since Icelake is mobile U/Y only.

Problem with a theoretical Sunny Cove on 14 nm is of course that the core would be much bigger and draw more power; and throwing in 2 core cores would be too much. 28W Comet Lake U by the way is I believe supposed to be released before anything Icelake.

 

[Nothingness]

They're nice but that's it.

So what is a great core?

Nothing.
x86 will still rule the world.

I guess you mean the server/desktop/laptop world (aka the old world of computing devices). Because the rest of the computing world, such as tablets and phones, is ruled by ARM, and their number dwarfs x86.
(EDIT: just saw you indeed were talking about servers)

 

[Nothingness]

Nice to finally see a strong counterbalance to the "Apple CPUs are godlike" mantra that's been on the CPU forums since the A12 series made its debut!

What technical counterbalance did we see exactly? I only read messages after messages of people talking about an unreleased x86 chip. It's hard to swallow but currently Apple micro arch is better than Intel for a huge part of the market.

 

[Nothingness]

Yeah, when I saw the Sunny Cove stuff, my immediate thinking was, "Wow, these changes look really solid and represent the biggest 'Tock' since Sandy Bridge."

Interestingly I was disappointed. Don't get me wrong these are solid and needed changes that will likely make these chips shine. But I was expecting something more radical

 

[Spartak]

5%-10% improvement isn't pessimism. Your optimism is a faith based argument until facts present themselves.

Flat earth society is looking for new members.

Facts have presented themselves already but something tells me even after benchmarks you and mark and others will maintain your position.

 

[Nothingness]

Facts have presented themselves already but something tells me even after benchmarks you and mark and others will maintain your position.

What facts have we seen? I don't consider marketing slides or dubious leaked Geekbench results as a proof of performance. Do I miss some information?

 

[coercitiv]

What facts have we seen? I don't consider marketing slides or dubious leaked Geekbench results as a proof of performance. Do I miss some information?

https://fuse.wikichip.org/news/1941/intel-reveals-10nm-sunny-cove-core-a-new-core-roadmap-and-teases-ice-lake-chips/2/

Sunny cove is wider design, with changes throughout the core. These are not marketing slides.

 

[Nothingness]

https://fuse.wikichip.org/news/1941/intel-reveals-10nm-sunny-cove-core-a-new-core-roadmap-and-teases-ice-lake-chips/2/

Sunny cove is wider design, with changes throughout the core. These are not marketing slides.

These don't provide any performance improvement figure. At least none that can be labelled as facts. Did Intel give any performance figure? Because "Intel believes Sunny Cove delivers a significant IPC increase." is not a fact and in my book marketing speech.

My point is that Spartak is talking of "flat earth society" though he has nothing concrete to disprove Maddie's claim. At least nothing that can be called a "fact". If anyone is able to give accurate performance improvements only looking at micro arch slides, they should send me a resume, I have open positions

 

[jpiniero]

These don't provide any performance improvement figure. At least none that can be labelled as facts. Did Intel give any performance figure? Because "Intel believes Sunny Cove delivers a significant IPC increase." is not a fact and in my book marketing speech.

If you want accurate numbers you'll just have to wait until a product with Sunny Cove ships. I'd say the bigger problem is just how widespread Icelake Mobile is going to be.

 

[Nothingness]

If you want accurate numbers you'll just have to wait until a product with Sunny Cove ships

Exactly! So accusing others of believing the earth is flat because there are "facts" that prove them wrong when there are none is just amusingly ridiculous