Intel Announces Sunny Cove Brings 18% IPC Improvement

[Dan_D]

That would be amazing Dan. I think Haswell, Skylake or Kabylake, would be a good sampling of the older architectures. I can see it would be a bit of work though, as you would need both old and new Bios's for the older boards, plus a hardrive with older install of windows to ensure it was pre-patch. Summer of 2017 build should be old enough. Pretty sure that by Fall/winter 2017 Microsoft had already slipped the first round of patches in, even before the vulnerabilities were made public.

Well, allot of that isn't as hard as it sounds. The biggest issue would be the time it takes to setup, configure, come up with the tests, methodology and then perform it that many times.

 

[N4CR]

Sunny Cove was designed to compete with Skylake- and since AMD still hasn't caught up to Skylake, we really should be thanking Intel for getting their 10nm foundry sorted.

You couldn't sell me land in waterworld

 

[IdiotInCharge]

My contribution was: It is hard to nail down the 18% improvement due to intel's instruction segmentation. You took it off topic and continue to derail it.

The topic is still Sunny Cove.

The only factor that can alter IPC externally is communication (ex. edram). Otherwise IPC should be identical. Unless one artificially segments their instruction set. Oh wait, now we've come full circle. Maybe reread my original post and stop derailing it.

It's a new architecture. It's not Skylake.

 

[schmide]

The topic is still Sunny Cove.
It's a new architecture. It's not Skylake.

Duh you can't be that daft? The reference point is 18% from skylake. If you didn't derail things so much, you might get my original point. Then again maybe your posts are just here to confuse?

 

[ChadD]

Why are we still even debating this junk.

*may not reflect all publicly available security updates*

End of story.

Their previous generation gave up 18% in performance once security updates are turned on... and they can't even be bothered to say mitigations are on in this infographic. But they mention the "MAY NOT" have applied any of them or all of them... you guess.

So I will guess. They applied them to their previous generation as well as the new, so best case they are saying their new generation will have hardware mitigation so they "Gain 18%". Or worse case they simply turned them off on their new chips. I would like to believe they turned them on for all chips tested... and their new chips are hardware hardened vs all the side channel crap... BUT if that where true they would have said so. Frankly people that care about 18% IPC... would have been just as happy to hear we have updated all security and given up nothing.

Its crap PR like this that makes it really easy to hate Intel. As a big Linux booster I really want to like Intel... clear Linux, unifying open source drivers, 1000s of kernel fixes every year. So much to love... and yet they stoop to this type of crap when they could likely just sit on their hands and hold a solid majoirty of the market for years even if AMD is making the better desktop chips.

 

[IdiotInCharge]

Frankly people that care about 18% IPC... would have been just as happy to hear we have updated all security and given up nothing.

If they've been working on the 'mitigations' since Skylake, why wouldn't they be in Ice Lake and thus Sunny Cove?

 

[Derangel]

If they've been working on the 'mitigations' since Skylake, why wouldn't they be in Ice Lake and thus Sunny Cove?

If they were Intel would be shouting it from the rooftops.

 

[ChadD]

If they've been working on the 'mitigations' since Skylake, why wouldn't they be in Ice Lake and thus Sunny Cove?

Sort of my point...

Its quite possible that Intel... just fixed at least most of their side channel issues in hardware. Then ran some tests on systems with them. Then claimed an 18% uptick. Which there may well be if you lost 16% or so in the previous generation to mitigations.

So its quite possible this chip has 2-3% improved IPC just like every Intel generation for 8 or 9 years now.

Unless Intel releases a bunch of tests with very exact list of what specific OS / patches where used or not used for each chip. I call BS.
I don't know I hear they have at least one person working for them these days that has plenty of experience running fair and honest tests. lol

 

[DTN107]

Did Intel get these numbers from Principled Technologies?

 

[IdiotInCharge]

Did Intel get these numbers from Principled Technologies?

Alright, that was funny

 

[techguymaxc]

Not that I give a crap about the opinion of rando fanboys on teh interwebs, but if you think the massive architectural enhancements present in Sunny Cove account for a 2-3% IPC uplift you have zero business discussing this subject.

 

[Grimlaking]

Massive architectural enhancements. I guess when you're talking 10nm to 7nm, 10nm does look massive.

 

[IdiotInCharge]

Massive architectural enhancements. I guess when you're talking 10nm to 7nm, 10nm does look massive.

They're about the same size, comparing Intel 10nm to TSMC 7nm.

But of course it's like comparing firearm cartidges using only caliber: there's a huge difference between .50 caliber cartidges, to use one example. A single dimensional measurement at some point of a three dimensional device, that's not even standardized when it comes to microprocessor fabrication, just does not make it easy to make direct comparisons across companies.

With respect to actual architectual changes, we'll likely have to wait for desktop parts to ship to see what's improved and what's been fixed between Skylake and Ice Lake. Sunny Cove is going to be very difficult to benchmark given that it will come in smaller laptops.


[if they put it in a NUC, that'll probably be the best comparison that we can get, until the desktop Ice Lake parts arrive]

 

[ChadD]

Not that I give a crap about the opinion of rando fanboys on teh interwebs, but if you think the massive architectural enhancements present in Sunny Cove account for a 2-3% IPC uplift you have zero business discussing this subject.

Intel gave away the game in their own infographics when they couldn't simply detail patches on or off. So I'll assume they where doing whatever makes them look best... so to account of the 10-20% loss from mitigations they would have had to be claiming a 30% uplift to see an actual double digit improvement.

 

[ChadD]

They're about the same size, comparing Intel 10nm to TSMC 7nm.

They are technically very much the same sure. Accept that one works and so far the other doesn't.

At this point the only chips Intel has shipped have been very broken. I hope they sort it... TSMC has already completed their 5nm design work and companies are designing 5nm stuff right now. Granted it doesn't sound like 5nm is going to offer a ton of extra performance, as the claimed gains are much lower then the jump to 7 was. Having said that they are cliaming 45% reduced die sizes vs 7nm. I'm sure AMD will use 7nm for the zen2+ update... but after that a Zen 3 5nms die reductions would make it much easier to slot more chiplets be they cpu gpu or even custom stuffs.

Anyway bottom line Intel needs to get their 10nm fabs cooking for real.

 

[clockdogg]

Sort of my point...

Its quite possible that Intel... just fixed at least most of their side channel issues in hardware. Then ran some tests on systems with them. Then claimed an 18% uptick. Which there may well be if you lost 16% or so in the previous generation to mitigations.

So its quite possible this chip has 2-3% improved IPC just like every Intel generation for 8 or 9 years now.

Unless Intel releases a bunch of tests with very exact list of what specific OS / patches where used or not used for each chip. I call BS.
I don't know I hear they have at least one person working for them these days that has plenty of experience running fair and honest tests. lol

Not anymore.

Oh well... I'm sure that Intel is sure that Sunny Cove will be an unmitigated success.

 

[notarat]

And this is also highly dependent on workload. Ultrabook customers are far less likely to be running over-provisioned mission critical VMs .


[I run VMs on both my Windows and Linux installs all the time, but I'm weird...]

j/k

 

[Tsumi]

Sort of my point...

Its quite possible that Intel... just fixed at least most of their side channel issues in hardware. Then ran some tests on systems with them. Then claimed an 18% uptick. Which there may well be if you lost 16% or so in the previous generation to mitigations.

So its quite possible this chip has 2-3% improved IPC just like every Intel generation for 8 or 9 years now.

Unless Intel releases a bunch of tests with very exact list of what specific OS / patches where used or not used for each chip. I call BS.
I don't know I hear they have at least one person working for them these days that has plenty of experience running fair and honest tests. lol

Well, it's not as if a hardware fix for these security holes magically come with no performance penalties, especially for ones like Spectre and Meltdown. Any performance penalty due to security loophole fixes will have to be mitigated with architectural improvements, so you cannot say "Hey, hardware patch is all they needed to do to claim performance improvement."

 

[ChadD]

Well, it's not as if a hardware fix for these security holes magically come with no performance penalties, especially for ones like Spectre and Meltdown. Any performance penalty due to security loophole fixes will have to be mitigated with architectural improvements, so you cannot say "Hey, hardware patch is all they needed to do to claim performance improvement."

But it looks like that is exactly what they have done. Mitigating side channel attacks in software is slower then having the hardware do it. This infographic stinks to high heaven.

If they run a patched OS.... on an older chip, its a 100% software solution and kills performance. On a newer chip that has at least most of the mitigations in hardware performance is less effected. Its possible to take the exact same Intel chip change nothing at all but add the security fixes... and you could say look it runns much faster. Which is true it does sif your going to run it with security fully enabled. (which is good news don't get me wrong) Its just sort of complete BS to then claim that FIX in marketing materials as an evolutionary leap which it isn't at all.

In these testes they clearly say... right there on the bottom in fine print that the security patchs may have been applied or not. They don't even say one way or the other. Which right off the top is shady. Makes me question their claim. I would say they are completely not lying at all... there really is likely an 18% increase when measured against the previous generation with all security turned on. (moving the mitigations to hardware would account for the majority of that number)

Its just BS PR.... as the previous generation is 16% faster on average with mitigations off. (which they where on day one of those chips sales)

 

[IdiotInCharge]

In these testes they clearly say... right there on the bottom in fine print that the security patchs may have been applied or not.

I haven't chimed in on this yet, since it's obvious that Intel is being inspecific- but it's important to note that they're saying 'at the time'. More and more extreme hacks are being found on a regular basis (nearly all of them operationally unexploitable, for now), so a disclaimer saying basically that they benchmarked against what was current at the time but may not be everything, does make sense.

Of course, if they're not providing detailed specifics, their numbers can't be checked until Sunny Cover parts are available to reviewers- and I should repeat that even that is going to be tough given the platform.

I realize that most in this thread are more interested in the Ice Lake cores on the Sunny Cove SoCs than Sunny Cove itself. I get that. Unfortunately we're all going to have to wait for that, and in the meanwhile, Sunny Cove is actually pretty interesting on its own.

 

[Tsumi]

But it looks like that is exactly what they have done. Mitigating side channel attacks in software is slower then having the hardware do it. This infographic stinks to high heaven.

If they run a patched OS.... on an older chip, its a 100% software solution and kills performance. On a newer chip that has at least most of the mitigations in hardware performance is less effected. Its possible to take the exact same Intel chip change nothing at all but add the security fixes... and you could say look it runns much faster. Which is true it does sif your going to run it with security fully enabled. (which is good news don't get me wrong) Its just sort of complete BS to then claim that FIX in marketing materials as an evolutionary leap which it isn't at all.

In these testes they clearly say... right there on the bottom in fine print that the security patchs may have been applied or not. They don't even say one way or the other. Which right off the top is shady. Makes me question their claim. I would say they are completely not lying at all... there really is likely an 18% increase when measured against the previous generation with all security turned on. (moving the mitigations to hardware would account for the majority of that number)

Its just BS PR.... as the previous generation is 16% faster on average with mitigations off. (which they where on day one of those chips sales)

You and others have been claiming Intel used these insecure pathways to be able to make their processors fast. Securing said pathways means speed must be made up elsewhere in order to actually be faster. If Intel had found a way to make the pathways secure without performance penalties, they would have done so already, like they did on the latest steppings for the Hyperthreading-based attacks.

 

[IdiotInCharge]

If Intel had found a way to make the pathways secure without performance penalties, they would have done so already

Seriously, they've been using the same uarch for over half a decade. Already done so?

Ice Lake is their first new uarch, and as they've known about these vulnerabilities for years, they've had time to work fixes into hardware. Their 10nm rollout may be a shitshow- and it may never really happen in earnest- but that doesn't mean they stopped designing CPUs or working on related technologies including security.

 

[chithanh]

If they've been working on the 'mitigations' since Skylake, why wouldn't they be in Ice Lake and thus Sunny Cove?

The situation is complicated. Since the Zombieload/RDS class of vulnerabilities was revealed, this situation became even more complicated, as Meltdown hardware mitigations caused some perturbation here (fixed some, made others even worse).

Newer Intel CPUs with partial hardware mitigations suffer from performance losses not as hard as the older ones. While an older laptop from Broadwell generation sees 18% drop (25% if you disable HT), for newer Cascade Lake chips this can be as low as 4% in server-oriented workloads (where Epyc drops by around 1%). Skylake on desktop sees 16% drop, compared to AMD 3%.

But this means:
A large part of the performance increase could come just from fixing hardware security flaws (making fewer operating system mitigations necessary), and not from architectural improvements. How much exactly we do not know, because Intel hides this behind:

*may not reflect all publicly available security updates*

 

[IdiotInCharge]

But this means:
A large part of the performance increase could come just from fixing hardware security flaws (making fewer operating system mitigations necessary), and not from architectural improvements. How much exactly we do not know, because Intel hides this behind:

I can't disagree. Their inspecificity has created the vacuum.

So it does depend on the points of comparison.

Ice Lake vs. Skylake both pre-patch? Both post?

The comparison I'd want to see is Skylake pre-patch vs. Ice Lake post-patch, as they'd both be used in their time.

 

[ChadD]

I haven't chimed in on this yet, since it's obvious that Intel is being inspecific- but it's important to note that they're saying 'at the time'. More and more extreme hacks are being found on a regular basis (nearly all of them operationally unexploitable, for now), so a disclaimer saying basically that they benchmarked against what was current at the time but may not be everything, does make sense.

Of course, if they're not providing detailed specifics, their numbers can't be checked until Sunny Cover parts are available to reviewers- and I should repeat that even that is going to be tough given the platform.

I realize that most in this thread are more interested in the Ice Lake cores on the Sunny Cove SoCs than Sunny Cove itself. I get that. Unfortunately we're all going to have to wait for that, and in the meanwhile, Sunny Cove is actually pretty interesting on its own.

That is fair there is always a legal dance to play.

Having said that... if all these chips where tested fully patched right now. Of course the new chip shows an 18% imporvement. Which is the point... if that is the case I'm very glad they fixed most of the side channel stuff in hardware.

Having said that the 18% uptick... is really 2% cause they dropped 16% on the previous chip due to secuirty flaws.

 

[ChadD]

You and others have been claiming Intel used these insecure pathways to be able to make their processors fast. Securing said pathways means speed must be made up elsewhere in order to actually be faster. If Intel had found a way to make the pathways secure without performance penalties, they would have done so already, like they did on the latest steppings for the Hyperthreading-based attacks.

If you understand how the side channel exploits work... it takes more then a simple stepping fix. Yes they can protect against a handful of side channel stuffs by changing a few gates as is possible with a stepping. However for things like Spector and Meltdown specificlly the fix requires an actual completre redisgn.

Is it fair to say Intel was cheating for performance. Yes, yes it is. Because they where.

There is no way in hell anyone can convince me that an intelligent chip designer of any stature doesn't understand that you can't be allowing software to check kernel level cache space BEFORE checking to see if said software has permission. It was done because the secuirty check takes actual time. I will grant that they likely thought "this is a 32k L1 cache space... we can just check for permission when its clear those bits are actually being requested." as for sure most of the bits in the L1 cache are placed their predictavely and likely half or more simply get over written never being used.

Still ARM, AMD, IBM, Alpha... and every other chip maker I know of DOES actually check permissions first.

That is the most grevious of a handful of cheats implemented into Intels chips to speed their predictive speculation engine.

As I have said though and I'm serious... cause I really don't hate Intel. If they have gone and corrected those issues... and found a way to tweak their cores so they are still gaining even 2-3% with hardware mitigations to Side channel stuff. That is FREAKING increadabley great news.

I wish their infographic reflected that and simply said. 100% fully hardware secured vs all current known side channel attacks.... with no performance pentalty vs no mitigations. If they could still show 2-3% great. I mean its not like AMD, ARM and Power ect are 100% clean their implmentations need microcode fixes as well. Its just they wheren't not cheating quite so hard and only drop 2-3% on average with microcode and software fixes.

 

[IdiotInCharge]

Having said that the 18% uptick... is really 2% cause they dropped 16% on the previous chip due to secuirty flaws.

This is basically what we're going to have to wait to see confirmed or refuted. On the one hand, rumors had a 10% to 15% before mitigations; on the other, if the mitigations were done in hardware and didn't tank performance, and the IPC increase holds, I agree that we should be looking for more than an ~18% post-mitigation jump.

My bet is that it's something in the middle.

 

[MMitch]

I can't disagree. Their inspecificity has created the vacuum.

So it does depend on the points of comparison.

Ice Lake vs. Skylake both pre-patch? Both post?

The comparison I'd want to see is Skylake pre-patch vs. Ice Lake post-patch, as they'd both be used in their time.

Gotta hand it to you, this reply seems rational. Dan seemed to hint that MAYBE he'll test a few generation with & without mitigation so we shall see.

 

[IdiotInCharge]

Gotta hand it to you, this reply seems rational. Dan seemed to hint that MAYBE he'll test a few generation with & without mitigation so we shall see.

I expect them to be ripped apart on release

 

[Dan_D]

And the pentium 4 hit over 5Ghz stock...right?? According to Intel it was supposed to.

Like everyone else is saying, I'm waiting for real benchmarks.

They were supposed to hit 5GHz and beyond. They got close with a 3.8GHz stock part. That didn't even last long and they dropped to 3.73GHz. I actually have one of the 3.73GHz Extreme Editions somewhere around here.

 

[SIS]

Hardware mitigations do not automatically bring back lost performance. All they do is reduce or remove the flaw/exploit from being able to happen on the hardware instead of having software fix the problem with the hardware still being vulnerable. There is nothing which guarantees that any performance penalty will be reduced or eliminated.

Additionally, there is no guarantee that new vulnerabilities haven't been baked into the new hardware, unintentionally, intentionally, or both.

There is a conflict of interest when it comes to secure design. Given that there is a lack of adequate competition (duopoly doesn't really cut it and AMD only took out its PSP black box for China — not the rest of us), Intel knows it can sell chips to people by telling them the ones they have are insecure and the new ones fix the publicly known vulnerabilities. As long as it's profitable to build in vulnerabilities it can't be discounted as a possible corporate move. (The vulnerability as planned obsolescence.)

There is also the issue of government demands for backdoors. People may be placated by the exposure of vulnerabilities because it makes them feel that someone is watching out for them. However, who can say that replacement vulnerabilities haven't been substituted? If government A has spyware feature X installed in a CPU and uses it for years, then discovers that opposition government B is using it, then suddenly government A will allow a corporation to let the public know about it. Meanwhile, it has cooked up spyware feature Y for the next generation of chips.

The FDA, for many decades, allowed arsenic to be added to animal feed to combat parasites due to filthy overcrowded conditions. When the public finally became aware of the practice, the FDA and one or more drug companies held a dog and pony show. The FDA banned a bunch of the chemicals and the drug company(ies) voluntarily withdrew some from production. What the public wasn't meant to think about was the remaining arsenicals which were quietly continuing to be used and which were hardly so significantly different from the banned/withdrawn ones. Media reports, unsurprisingly, generally made people think that arsenic had been taken out of the food supply. This is a typical strategy. Give people the false impression that a problem has been solved or that they have the ability to solve it (like privacy control switches that don't cover all of the territory and/or don't stick).

 

[Staples]

Until I see “real” benchmarks, AMD and Intel are just pissing on walls asking me to look at how high it went.

My thoughts exactly. I am amazed how much excitement has been stired up over AMD and Intel claims yet we've seen zero benchmarks. Both these companies have a track record (AMD being the worse offender) of inflating their performance numbers.

 

[Hagrid]

Wikichip has an article on the latest information about Intel Sunny Cove core optimizations. Intel Chief Core Architect Ronak described the 18% IPC increase across various workloads in this manner. “The last time we have had something in this range was during the Merom era,” Ronak, Intel Chief Core Architect, said. He added, "I have had some of you ask me, “can you actually push the IPC further anymore or are we running out of steam doing this?” And the answer is: No. And Sunny Cove is one example of this. The other microarchitectures we are working on back at home will go even beyond this. Sunny Cove is the foundation for our next wave of innovation."

Make sure to read the article's beginning to understand how Intel improved various aspects of Skylake to create Ice Lake and then changed the game with new technology in the Sunny Cove cores.

View attachment 163760

View attachment 163761

I guess they did not want the 9th gen in there.

 

[Jandor]

Massive architectural enhancements. I guess when you're talking 10nm to 7nm, 10nm does look massive.

Sunny Cove just has more cache but this is good. AMD makes the same move with Zen 2.
However Intel has a huge problem if they are stuck to 4c/8t on 10nm for a while. AMD will deliver à 12 core/24 threads in July and are restraining themselves to launch a 16 core because Intel is so much late with what was expected. I never thought seeing that kind of situation, when AMD doesn't deliver even if they could because they have no more competition.
Now mind that AMD may have the ability to launch sooner than expected an APU with 2 chiplets, one of 8 c/16t and one with a Navi graphic chip that will completely surpass Intel Gen 11.
It looks like Intel will have hard time catching AMD in a couple of years if AMD is on standby. This looks really bad for Intel. It's not even close to the situation they had with Pentium IV vs Athlon 64.
Not sure Intel will survive this couple of years. Intel may become a second class company or split or sold by pieces. In a couple of years if a miracle doesn't happen, Intel will be on par with low tech China who will compete with their own Zen 1 chips (AMD licensed them only for China).
For instance in 2021/2022 Samsung will be producing 3nm EUV with new kind of nano-tube gates and TSMC will be mass-producing for sure 5nm EUV for everybody, like AMD. At that time Intel will supposedly, if they deliver as expected, start small tests on 7nm EUV, like they have on 10nm with Canon Lake more than 2 years ago (2.5 years to go from 2c/4t to 4c/8t...).
Not even sure AMD will not have hard time in a couple of years against Japanese tech ARM on 3nm Samsung.
Intel is so much out of the game now, only promising for unexciting things, that may happen while they always failed to deliver for the last 10 years. I don't even bet on anything on their GPU promised side. They tried for so long on the GPU side and always completely failed. With the huge money Intel already invested in GPU, AMD will be skyrocketting against Nvidia.
It's sad, but it's game over for Intel and I'm not even betting on the future of X86 at AMD. Apple will very soon go the ARM route on MacOS for sure. HiSilicon ARM CPU cores made by Huawey on 7nm TSMC are already better than Zen 2 cores. Trump is a bit late at trying to stop Huawei and US chip foundries are already lower tech than Korean, Taiwanese, Japanese and on the verge of being caught by the Chinese next year. Not even discussing Europe who is completely out of the game now.

 

[IdiotInCharge]

However Intel has a huge problem if they are stuck to 4c/8t on 10nm for a while. AMD will deliver à 12 core/24 threads in July

This is true, but also a false comparison; Intel will be stuck on quad-core for 10nm for a while, but these parts will not be competing with larger AMD parts. In reality, AMD doesn't have a part that is competitive with Intel's 14nm ultrabook parts, let alone their upcoming 10nm Sunny Cove SoC.

 

[OrangeKhrush]

R

My thoughts exactly. I am amazed how much excitement has been stired up over AMD and Intel claims yet we've seen zero benchmarks. Both these companies have a track record (AMD being the worse offender) of inflating their performance numbers.

That must be the weakest objectivity I have ever seen. I mean Intel said the 9900k was 2 times faster than the 2700 used smear benches from a bogus company doing gimp tests only to find when real results came they were a long way short? Or the MCE cheap trick debacle with coffee lake?

At least AMD had cinebench and PUBG while Intel has slides and the worst set of disclaimers I have ever seen in my 12 year legal career

 

[IdiotInCharge]

At least AMD had cinebench and PUBG while Intel has slides and the worst set of disclaimers I have ever seen in my 12 year legal career

If you're attempting to refute one set of vendor supplied benchmarks with another set of vendor supplied benchmarks, you're wrong.

 

[OrangeKhrush]

If you're attempting to refute one set of vendor supplied benchmarks with another set of vendor supplied benchmarks, you're wrong.

that is basically intel, we test our stuff with our own custom set of configs and intel only stuff, results may vary in the real world but in our own pre customed tests we are heman.

 

[IdiotInCharge]

that is basically intel, we test our stuff with our own custom set of configs and intel only stuff, results may vary in the real world but in our own pre customed tests we are heman.

And there's your naked bias.

 

[OrangeKhrush]

And there's your naked bias.

It says so in their disclaimer, the only refute is a refute and slide without anything substantial. If anything the lies and deception have come from intel.

Snap remembered the third one, when Intel claimed the 28 core Xeon hit 5ghz but forgot to mention the Industrial sub zero chiller, when they were busted their alibi was "oh damn, how did that get there".