Discussion Intel current and future Lakes & Rapids thread

[uzzi38]

More stupid than spending millions of dollars porting Tigerlake back to 14nm, for a small market of enthusiasts?

Not to mention the fact that it'd barely be able to outperform the 'SKL-X on ringbus' idea too. Maybe an extra 5%, if you're incredibly, incredibly lucky.

 

[Richie Rich]

Its a good way to get AVX-512 support across their whole lineup (finally), and would help encourage developers.

There is something like 20 different instruction subsets under AVX512. Every Intel CPU lineup supports different subset group. Even IceLake doesn't support all of them (14 out of 20). https://en.wikipedia.org/wiki/AVX-512#CPUs_with_AVX-512
AVX512 is messy and that's why there's so slow adoption from SW developers and AMD (and partly from Intel too). Developers will be encouraged after CPUs will support all subsets IMHO.

 

[jpiniero]

More stupid than spending millions of dollars porting Tigerlake back to 14nm, for a small market of enthusiasts?

The purpose would be to maintain single thread superiority in mobile and desktop. Skylake's just not going to cut it.

 

[uzzi38]

The purpose would be to maintain single thread superiority in mobile and desktop. Skylake's just not going to cut it.

Rocket Lake won't either.

 

[NTMBK]

The purpose would be to maintain single thread superiority in mobile and desktop. Skylake's just not going to cut it.

So they should spend millions to backport and still lose?

 

[jpiniero]

So they should spend millions to backport and still lose?

I wouldn't say it'll work. Clearly it would have to be a combination of additional efficiency gains and still blowing out the power consumption to make Willow Cove work on 14 nm.

 

[mikk]

More stupid than spending millions of dollars porting Tigerlake back to 14nm, for a small market of enthusiasts?

Yes of course, there is a healthy IPC increase with Willow Cove, it makes much more sense, really strange question. Also the desktop is not a small market for Intel.

 

[SAAA]

For anyone not running at 5GHz the jump to a new architecture would be fabulous. Think like… uh 80% of the CPUs they sell, not some high end i7-i9 or unlocked parts.

Also the fact that finally we could get new instructions and advancements rather than a 5 years old zombie. Man if Piednoel really did suggest for backporting Ice Lake a couple years ago and they didn't on some positive 10nm outlook that really got to hurt now, especially with Zen 3 incoming.

AMD kept running on Bulldozer for desktop because they had limited time and money so rather than 8 core Steamroller or Excavator we had ever increasing clocks on outdated parts, while they solely focused on Zen.
I find hilarious a company with 10 times as many resources such as Intel can't push all the levers and just implement every new trick they have, rather than have them dusting in their portfolio.

IMHO: we should have a tiger successor coming out, not some infinity lake. It's impossible they aren't ready arch wise, it's only a node issue.
If anything even newer is boiling inside I sure hope it's at least 50% better core/core, or totally disruptive to the CPU world like reverse multithreading actually working.

 

[IntelUser2000]

So they should spend millions to backport and still lose?

The alternative is doing nothing. Also, it'll cost millions making a ringbus version of SKL-X, the AVX-512 part of which will be insignificant for majority of consumers. I also can't imagine it'll reach the clocks 9900K variants will reach so it'll lose anyway.

If Willow Cove gets 7-8% perf/clock increase, then it'll end up being 25-30% faster than Skylake per clock. Then even at 4GHz clocks, it'll be a win against 5GHz Skylake parts(when you consider imperfect scaling, and being much easier to hold the top clocks).

Maybe it won't have the fastest part, but bulk of the chips lower than that willl benefit in a big way, since most chips aren't pushing 5GHz. Just like @SAAA rightly points out.

Intel shouldn't have pushed for 5GHz. But that happened because they had no other choice. If we had Willow Cove now and a 4GHz chip, then I bet the chip would have had some overclocking headroom, making it a better chip overall.

Frankly, I think its amazing they are pushing 5GHz stock chips, but its nothing more than flexing their muscle to prove a point, rather than a valid strategy. We use to need water cooling to get older chips to that frequency. Now that water cooling is becoming more common, and the chip can easily reach 150W or more so nothing has changed. Coffeelake and 14nm++ isn't superior to any methods before it, its just that they had the same chip for 4 years and got to refine it, slowly, and using clever binning methods.

 

[DrMrLordX]

So they should spend millions to backport and still lose?

The alternative is doing nothing.

That's the crux of it. Rocket Lake-S is going to lose on IPC and core count, and might not do much better in clockspeed either, but at least Intel will have a new product to replace Comet Lake-S that has some improvements. Rocket Lake-S could have been out by now had Intel created a backup plan for 10nm failing as badly as it has. If it were on the market right now, it wouldn't be so bad. The only thing wrong with the product is that it's delayed until 2021.

 

[IntelUser2000]

@DrMrLordX If Rocketlake is early 2021, then they'll still have the perf/clock lead.

Rocketlake-S with Willow Cove only needs about 4GHz to outperform Coffeelake in single threaded workloads, because 5GHz is an unrealistic frequency target, and the point where the power and thermal levels run out of control. They somehow made a Netburst chip again with CFL.

If you go through NBC data, you'll see 1065G7 reaching its 3.9GHz max clock readily, while Cometlake falls 100-200MHz short of the 4.7-4.8GHz top clock.

At ultra low power levels, Amberlake even fails to reach a much more conservative 4.2GHz level. Amberlake with 600MHz advantage fails to outperform Kabylake-Y in Cinebench ST! HWInfo shows that it can reach 4.2GHz, but benchmark results tell us its extremely easy to dislodge off its peak clock, and by a lot. So much for 14nm's clock prowess!

At the -Y levels, 10nm chips will easily outperform 14nm chips, because all it has to do is reach the same clock level, and uarch will do the rest.

Ultimately though, Intel should go back down to the 4GHz level, and AMD should continue to do what they are doing, and not reach 5GHz as some are suggesting. Intel had to because they had no choice. They deserved it, because they put in the work. But its a bad way to go about things. 3 years for 300MHz. A single architecture advance of 10% would have obliterated all that.

 

[A///]

For anyone not running at 5GHz the jump to a new architecture would be fabulous. Think like… uh 80% of the CPUs they sell, not some high end i7-i9 or unlocked parts.

Also the fact that finally we could get new instructions and advancements rather than a 5 years old zombie. Man if Piednoel really did suggest for backporting Ice Lake a couple years ago and they didn't on some positive 10nm outlook that really got to hurt now, especially with Zen 3 incoming.

AMD kept running on Bulldozer for desktop because they had limited time and money so rather than 8 core Steamroller or Excavator we had ever increasing clocks on outdated parts, while they solely focused on Zen.
I find hilarious a company with 10 times as many resources such as Intel can't push all the levers and just implement every new trick they have, rather than have them dusting in their portfolio.

IMHO: we should have a tiger successor coming out, not some infinity lake. It's impossible they aren't ready arch wise, it's only a node issue.
If anything even newer is boiling inside I sure hope it's at least 50% better core/core, or totally disruptive to the CPU world like reverse multithreading actually working.

You give him far too much credit. Netburst was such a failure that Intel did go back into their portfolio, dusted off blueprints for older uarchs and worked off of that to deliver Core. I wouldn't necessarily expect any major upset like they did with Core, especially not with their corporate offices becoming a revolving door of people coming in and leaving, or rehashed products over and over again pushed to their limits.

 

[DrMrLordX]

@DrMrLordX If Rocketlake is early 2021, then they'll still have the perf/clock lead.

Doubtful. Willow Cove isn't supposed to come with much of an IPC increase over Sunny, if I recall correctly. I think most of the TGL vs ICL leaks we've seen thus far are skewed by bad boost clocks on the ICL parts.

 

[mikk]

Doubtful. Willow Cove isn't supposed to come with much of an IPC increase over Sunny, if I recall correctly.

The last rumor claimed there is a 5-8% IPC increase over Sunny Cove. AMD needs a 20+% IPC increase with Zen 3 to match it which I doubt.

 

[DrMrLordX]

The last rumor claimed there is a 5-8% IPC increase over Sunny Cove. AMD needs a 20+% IPC increase with Zen 3 to match it which I doubt.

Zen3 allegedly has a 17% increase over Zen2 in aggregate (12% int, "up to" 50% fp), so we'll see.

 

[mikk]

Zen3 allegedly has a 17% increase over Zen2 in aggregate (12% int, "up to" 50% fp), so we'll see.

Skylake and Zen 2 have almost the same IPC, Skylake has a slightly better gaming IPC and Zen 2 a slightly better application IPC. So even the 17% increase won't be enough to match Willow Cove IPC, it would likely on par with Icelake IPC. But neither the 17% nor the 5-8% are confirmed.

 

[lobz]

The last rumor claimed there is a 5-8% IPC increase over Sunny Cove. AMD needs a 20+% IPC increase with Zen 3 to match it which I doubt.

Keep on dreaming

 

[Richie Rich]

Also the fact that finally we could get new instructions and advancements rather than a 5 years old zombie.

Exactly. Intel is stuck on 14nm longer then they expected and Intel has finished new uarch Willow Cove. So why not backport it for 14nm instead 5 years old Skylake? There is another big advantage: WillowCove@14nm will probably reach clock around 5 GHz. Together with +20% IPC over SkyLake it's highly probable that RocketLake will keep gaming crown on Intel side. Will RocketLake be power hungry? Yes, as hell. However for desktop/gaming this could be awesome CPU. IMHO more powerfull than Zen3 (at same core-count).

 

[uzzi38]

There is another big advantage: WillowCove@14nm will probably reach clock around 5 GHz.

Wrong.

 

[ondma]

Wrong.

Why not?

 

[uzzi38]

Why not?

Wonder how many times I'll have to say it at this point.
Willow Cove in it's entirety backported to 14nm would draw meme-levels of power to sustain even OG Skylake clocks. You can very much forget being able to even touch 5GHz.

Ice Lake-U is barely more efficient than Comet Lake-U once whatever issue it has at low power (<25W) is gone. And I mean barely. https://cdn.discordapp.com/attachments/476511857310564393/638725335306731541/unknown-7.png

Why is everyone suddenly assuming that Intel would be able to extract 20% more performance at the same power out of Willow Cove on 14nm compared to Comet Lake and, by extension, also 20% more power efficient than Sunny Cove on 10nm?

Are you people actually insinuating that the same architecture on both 10nm and 14nm would be just as power efficient?

God, I feel like I'm going mad the number of people that keep on going on about this 5GHz 14nm Willow Cove claim. For just two seconds, will someone just sit down and think about what they're suggesting? Because you're suggesting that 14nm and 10nm are as power efficient as one another.

 

[jpiniero]

10 nm is a broken node. They are jacking up the voltage to get as many chips as possible.

 

[uzzi38]

10 nm is a broken node. They are jacking up the voltage to get as many chips as possible.

Nah, undervolting headroom on ICL-U laptops tends to be really thin compared to 14nm chips. Intel are binning quite hard to get ICL worthy chips, and the ones that can pull off the clocks they get can barely do so.

It's just simply a fact that wider architectures are less power efficient.

 

[Zucker2k]

It's just simply a fact that wider architectures are less power efficient.

That's just basic logic, or science. On the other hand, Intel has done the math, and are apparently pleased with the result, contrary to the doom and gloom in here. I'm intrigued to see what this product brings to the x86 landscape.

 

[Richie Rich]

Why is everyone suddenly assuming that Intel would be able to extract 20% more performance at the same power out of Willow Cove on 14nm compared to Comet Lake and, by extension, also 20% more power efficient than Sunny Cove on 10nm?

Why do you think only about power efficiency? Desktop is not limited by TDP as high as 15W laptops. Yes, RocketLake @ 5GHz will consume +20% more power than Skylake, probably around 200W. But who cares? We talk about same kind of product like 9900KS, don't we?