Intel Announces Sunny Cove Brings 18% IPC Improvement

18% is a little below what I had hoped for (I predicted 20%), but still substantial when taken in-context. We are talking about an 18% jump over what was already the fastest x86 architecture ever released, afterall. It's not like there was easy low-hanging fruit like AMD had with Zen/Zen+ -> Zen2 (double FPU width).
 
Intel's of course talking about an 18% improvement over a 4 year old CPU architecture. It occurred to me that Intel has always pulled crap like this when talking about IPC improvements. They are used to making claims based on best case scenarios with specific workloads which are hardly representational of overall performance. Intel virtually always compares its CPU's against its own CPU's rather than the competition. It's done that even when it had massive advantages in performance. Remember the old iComp index? Intel CPU's were only ever compared against other Intel CPU's. So this isn't surprising. In this case, Intel's marketing team simply found a way to claim an IPC improvement that sounds better than Ryzen 2's 15% improvement.

A claim which we haven't even validated as being true either. In Intel's defense, we've had very little in the way of IPC improvements since Skylake. Kaby Lake and Coffee Lake increased turbo frequencies and a couple other changes but nothing really that increased IPC.
 
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.
 
18% is a little below what I had hoped for (I predicted 20%), but still substantial when taken in-context. We are talking about an 18% jump over what was already the fastest x86 architecture ever released, afterall. It's not like there was easy low-hanging fruit like AMD had with Zen/Zen+ -> Zen2 (double FPU width).

18% over Skylake. So probably 3-5% (at most) over mobile Coffee Lake refresh CPUs. The 15w single-thread chart shows it barely above Wiskey Lake and that chart is trying to compare vs Broadwell.
 
18% is a little below what I had hoped for (I predicted 20%), but still substantial when taken in-context. We are talking about an 18% jump over what was already the fastest x86 architecture ever released, afterall. It's not like there was easy low-hanging fruit like AMD had with Zen/Zen+ -> Zen2 (double FPU width).

Depends how they are measuring that. If you run a mitigated OS... the newest steppings of Intels chips on the market right now are 10-18% faster then the same chip without the hardware fixes.

So I would take anything Intel has to say one day after their only real competition one upped them with a huge grain of salt.
 
So I would take anything Intel has to say one day after their only real competition one upped them with a huge grain of salt.
I'll take anything either company has to say with a grain of salt, since AMD also likes to cherrypick best case scenarios when presenting it's specs and hardware. As things stand now though I am absolutely buying a Ryzen 3 assuming there are no gotchas with the need for expensive B-Die ram again to get the most out of the CPU.

The real-world benches will be what's interesting. Until then we're all just paddling pork here in the comments.
 
18% over Skylake. So probably 3-5% (at most) over mobile Coffee Lake refresh CPUs. The 15w single-thread chart shows it barely above Wiskey Lake and that chart is trying to compare vs Broadwell.

Uh, no.

Coffee Lake (and Refresh) has identical IPC to Skylake. They're the same architecture. Same goes for Kaby Lake.
 
18% over Skylake. So probably 3-5% (at most) over mobile Coffee Lake refresh CPUs. The 15w single-thread chart shows it barely above Wiskey Lake and that chart is trying to compare vs Broadwell.

With respect to their mobile parts, IPC increases will help- but what really helps is the ability to maintain boost clocks over time without hitting thermal saturation as well as being able to hit higher base clocks without burning through battery to keep the system responsive.

From the 6000-series to the 8000-series, Intel has come a long way, just by massaging Skylake on 14nm. A new uarch on 10nm might be a pretty big jump across the board, with slightly higher boost performance, but much longer endurance at those speeds within common thermal envelopes, not to mention that they're doubling IGP performance.

Supposing AMD has an APU update coming for mobile, they're going to have some heavy competition waiting.
 
Depends how they are measuring that. If you run a mitigated OS... the newest steppings of Intels chips on the market right now are 10-18% faster then the same chip without the hardware fixes.

So I would take anything Intel has to say one day after their only real competition one upped them with a huge grain of salt.

This is simply untrue. If it were true, you would be able to cite a source with before and after performance testing.
 
Uh, no.

Coffee Lake (and Refresh) has identical IPC to Skylake. They're the same architecture. Same goes for Kaby Lake.

Could have sworn the refreshes have had IPC improvements (even if minor) over time.

With respect to their mobile parts, IPC increases will help- but what really helps is the ability to maintain boost clocks over time without hitting thermal saturation as well as being able to hit higher base clocks without burning through battery to keep the system responsive.

From the 6000-series to the 8000-series, Intel has come a long way, just by massaging Skylake on 14nm. A new uarch on 10nm might be a pretty big jump across the board, with slightly higher boost performance, but much longer endurance at those speeds within common thermal envelopes, not to mention that they're doubling IGP performance.

Supposing AMD has an APU update coming for mobile, they're going to have some heavy competition waiting.

Yeah, if both the CPU and iGPU performance jumps work out like promised it could be very good for the mobile sector.

Considering AMD is presumably doing a custom Zen 2/Navi APU for Sony (and MS, I'd assume) I'd be surprised if they don't have something for mobile down the line. Maybe not any time soon, but I could see it happening eventually.
 
This is simply untrue. If it were true, you would be able to cite a source with before and after performance testing.

I'm betting that Phoronix has the benches, that dude lives chained to his benchmark rack.

But I'd personally like to see a consumer app survey from say GamersNexus showing before and after of common uarchs too.
 
Considering AMD is presumably doing a custom Zen 2/Navi APU for Sony (and MS, I'd assume) I'd be surprised if they don't have something for mobile down the line. Maybe not any time soon, but I could see it happening eventually.

Absolutely. I assume that it's a bit of a chicken vs. egg thing for AMD; they have to build some mindshare before they can invest in truly competitive mobile parts and have OEMs actually ship them in more premium systems.

Would love to see an AMD APU in a 13" 2-in-1 like an XPS13. Give me an IPS 1080p120 panel with VRR, and I'd be happy.
 
Uh, no.

Coffee Lake (and Refresh) has identical IPC to Skylake. They're the same architecture. Same goes for Kaby Lake.

He's talking about efficiency improvements, not IPC.

intel-sunny-cove-ipc-claim2-png.png


Kaby lake is 20% more efficient than Skylake at the same power limit, which results in about 15-20% higher performance. The additional Coffee Lake efficiency improvements was why they were able to bump to 6 cores at the same TDP.

At 15w TDP, Intel's quad-cores are still power-limited, so any efficiency gains result in performance increases (higher sustained clocks).

You will notice the IPC gains of Ice Lake more on 45w parts or higher, which are frequency-limited.
 
Last edited:
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.

Remember when Intel was having problems with the PIII competing with the Athlon, and they pushed the PIII to 1Ghz, and ended up having to recall the CPU because it would introduce errors at that speed?
 
I hope there are no knock-outs from either AMD or Intel in this skirmish, so that they can keep leapfrogging each other and we get the same kind of advancement and pricing we saw in the Athlon vs. PIII / P4 era.

I've been saying this all along though. AMD has an advantage right now due to process size. This won't last forever. They need to REALLY get their act together and improve the underlying architecture so they can remain competitive at the very least when Intel's 7nm process hits in 2021, if not sooner. It is unclear if 10nm will ever really provide anything of value, despite it being on their roadmap.

AMD needs a knock out though, for money reasons. Intel will obviously respond to that. Still this better than the past 4-5 years prior to Ryzen.
 
Remember when Intel was having problems with the PIII competing with the Athlon, and they pushed the PIII to 1Ghz, and ended up having to recall the CPU because it would introduce errors at that speed?

They recalled the 1.13GHz PIII, not the 1.0GHz


--------------


Isn't this the architecture that Intel was supposed to have released three or four years ago?
 
Uh, no.

Coffee Lake (and Refresh) has identical IPC to Skylake. They're the same architecture. Same goes for Kaby Lake.

clock for clock yes you're correct they're identical, the way they were able to claim IPC gains was through brute forcing higher boost clocks with each iteration of process improvements after skylake. 6th gen was 4.3Ghz, 7th 4.5Ghz, 8th 4.7Ghz, 9th 5.0Ghz boost. there's your 5% each generation. now if they're able to get 18% running at say 4.2Ghz then ok that's pretty damn nice gain within the architecture it's self but if it's just 18% because it's running at 5Ghz compared to the 4.2Ghz of skylake then it's the same damn shit just in a chiplet design.
 
clock for clock yes you're correct they're identical, the way they were able to claim IPC gains was through brute forcing higher boost clocks with each iteration of process improvements after skylake. 6th gen was 4.3Ghz, 7th 4.5Ghz, 8th 4.7Ghz, 9th 5.0Ghz boost. there's your 5% each generation. now if they're able to get 18% running at say 4.2Ghz then ok that's pretty damn nice gain within the architecture it's self but if it's just 18% because it's running at 5Ghz compared to the 4.2Ghz of skylake then it's the same damn shit just in a chiplet design.

Uhh, IPC is independent of clock speed. They cant claim an IPC increase just by bumping the clock speed up. IPS is what you've described.
 
clock for clock yes you're correct they're identical, the way they were able to claim IPC gains was through brute forcing higher boost clocks with each iteration of process improvements after skylake. 6th gen was 4.3Ghz, 7th 4.5Ghz, 8th 4.7Ghz, 9th 5.0Ghz boost. there's your 5% each generation. now if they're able to get 18% running at say 4.2Ghz then ok that's pretty damn nice gain within the architecture it's self but if it's just 18% because it's running at 5Ghz compared to the 4.2Ghz of skylake then it's the same damn shit just in a chiplet design.

IPC = Instructions Per Clock

It's an independent measurement of improvements in the architecture that having nothing to do with clock speed. Typical IPC measurements would clock 2 cpus at identical speeds, say 4.2Ghz, and run the normal gamut of benchmarks. The improvements show how the architecture is better.

I believe that graph posted earlier was about power efficiency improvements, not IPC or overall performance. Gotta be careful and keep it straight on what they are saying is improved, and keep in mind they will be saying/showing anything that puts the product in the best possible light. (Job of all marketing dept's).
 
IPC = Instructions Per Clock

It's an independent measurement of improvements in the architecture that having nothing to do with clock speed. Typical IPC measurements would clock 2 cpus at identical speeds, say 4.2Ghz, and run the normal gamut of benchmarks. The improvements show how the architecture is better.

I believe that graph posted earlier was about power efficiency improvements, not IPC or overall performance. Gotta be careful and keep it straight on what they are saying is improved, and keep in mind they will be saying/showing anything that puts the product in the best possible light. (Job of all marketing dept's).

True. Intel is pretty good at obfuscating what's truly going on by reporting things in an odd way. Additionally, they've made a big deal about performance per watt improvements from one generation to the next. In fact, Skylake to Kaby Lake was an example of that. It meant nothing on the desktop and was really only a mobile / server oriented improvement.
 
It's the compact notebook part would be my guess. Don't really need more cores or even more boost clock there, but better sustained clock and longer battery time.
Yield. Same reason the 10nm quad in leonovo is cut to a dual and no igpu.
They'd have them on desktop otherwise...
 
18% is a little below what I had hoped for (I predicted 20%), but still substantial when taken in-context. We are talking about an 18% jump over what was already the fastest x86 architecture ever released, afterall. It's not like there was easy low-hanging fruit like AMD had with Zen/Zen+ -> Zen2 (double FPU width).

This is why intel throws up numbers. Clearly it worked on you lol. Its not 18% over current gen lol.
 
So... either 18% better IPC at 10% lower clocks or 18% better "ipc" when clocked 10% higher (akin to the KBL improvements).
 
Take a long hard look at who DIDN'T post the Intel news article here. That right there should tell you there is no real foundation for these numbers. IF he was asked he didn't post them without being able to verify. Chances are though they never even bothered to ask him. Which was a mistake.

Saying this is the foundation for the NEXT chip that will take things further means... they realized they could make this chip and eeek out some more dollars before releasing the next one making them look like they are iterating when they already have the new deal in hand and ready to fab.

Why would I say that? Remember CDrom drives.. when 1x, then 2x then 3 x and so on were being released about every six months? Know why? Because they had warehouses of them when they were already making 40x and 100x cdrom drives.
 
Take a long hard look at who DIDN'T post the Intel news article here. That right there should tell you there is no real foundation for these numbers. IF he was asked he didn't post them without being able to verify. Chances are though they never even bothered to ask him. Which was a mistake.

Saying this is the foundation for the NEXT chip that will take things further means... they realized they could make this chip and eeek out some more dollars before releasing the next one making them look like they are iterating when they already have the new deal in hand and ready to fab.

Why would I say that? Remember CDrom drives.. when 1x, then 2x then 3 x and so on were being released about every six months? Know why? Because they had warehouses of them when they were already making 40x and 100x cdrom drives.

This announcement has nothing to do with the enthusiast sector or enthusiast events/outreach. Posting random PR fluff on forums would be outside of his job duties most likely.
 
Yield. Same reason the 10nm quad in leonovo is cut to a dual and no igpu.
They'd have them on desktop otherwise...

I would agree that they're likely not producing enough to switch the desktop line over, and since the desktop line won't benefit much from a new quad-core, I can see why they'd focus on mobile.
 
so another year for this to hit Desktop?...meanwhile AMD is releasing 7nm Ryzen chips this July?...sounds like Intel is in real trouble...plus AMD is not susceptible to any of those Meltdown/Spectre issues...I might have to get my first new AMD chip since Thunderbird
 
I would agree that they're likely not producing enough to switch the desktop line over
But mobile is bigger than desktop. Surely if they make enough for mobile, then desktop would not be a problem. Especially given that AMD is taking away marketshare in desktop more than in mobile.
since the desktop line won't benefit much from a new quad-core
I think the retooling of fabs from 10nm back to 14nm (ie. the intentional reduction of production capacity) means that Intel expected not much demand for 10nm. And certainly Intel could have made an 8-core Ice Lake if they wanted to, but the lower clocks are unsuitable for desktop.
 
But mobile is bigger than desktop. Surely if they make enough for mobile, then desktop would not be a problem. Especially given that AMD is taking away marketshare in desktop more than in mobile.

The problem with mobile is its a 100% oem game. That doesn't tend to trickle... when mobile sales shift they shift hard. AMD may be eating desktop cpu sales today... but with mobile sales anyone is just 2 or 3 major oem contracts away from changing the game.

If mobile version of Zen2 on 7nm are attractive enough to get the right contracts things could look very different very quickly. Still no doubt Intel has had a pretty hard lock on the mobile oems for a long time... question really becomes how long with the OEMs protect Intel and continue buying up stuff like broken 10nm chips ect.
 
Intel: Look at me! Look at me!
Why even bother, they mega dominate the industry ATM.
Google and Amazon have a greater chance of burying Intel than AMD ever will...
 
These graphs are weird. The second one (hardly an improvement over whiskey lake at 15W) seems to contradict the first slide, unless 10nm is truly terrible in terms of efficiency as well as just not clocking very high, or Sunny Cove's IPC increases have come at the expense of perf/W.

And then there's the spectre (heh) of the various speculative mitigations. Are those a part of the claimed IPC increase? If so, then things make more sense, even if it's horribly deceptive. And Intel's history of being over-optimistic about IPC increases in the past, combined with providing a single synthetic benchmark, does little to convince anyone this isn't still the case.

That said, I expect Sunny Cove to probably edge out Zen 2 for IPC, and generally be a good architecture, but the 10nm process probably means it won't be able to compete in the enthusiast space.
 
so another year for this to hit Desktop?...meanwhile AMD is releasing 7nm Ryzen chips this July?...sounds like Intel is in real trouble...plus AMD is not susceptible to any of those Meltdown/Spectre issues...I might have to get my first new AMD chip since Thunderbird

That's a bit of an overstatement. AMD CPU's have some exposure here, but it is MUCH less.
 
Ha... HAHAHAHA.. Wheeeze...

upload_2019-5-29_7-51-53.png


COME ON INTEL. 'Performance results are based on testing as of dates shown in configuration and MAY not reflect all publicly available security updates.'

Emphasis on may is mine.

Yea.. they MIGHT not be there... assholes... they should have released the CVE's that WERE applied to all systems.
 
Back
Top