Discussion Intel current and future Lakes & Rapids thread

[B-Riz]

AMD is hardly beating Intel by any significant margin

Uh, AMD did not plan on Zen3 selling out it's first run, and, it did. How is that not "beating" Intel right now?

AMD has a lot more positive vibes around its Zen3 products than the uninspiring 11 series accounting paper work CPU release. "Look Ma, we put something "new" out in 2021!"

 

[Gideon]

One red flag with this laptop is the term "IPS-level" being thrown around. That implies it is a TN panel. Either that or poor wording on their part. Outside of that, I will look into it more. Thanks for the link!

EDIT: Looks like it is NOT IPS or TN. It is AHVA.

Why not Zephrys G14 or G15? About the same with an RTX 3080 even if needed (the weight diff might be compensated by the power brick) and also 300 hz 1080p screen if required.

I would wait for reviews regarding all core loads, temps and noise for the Tiger Lake H design before pulling the trigger, JIC

And btw, I bought a hp x360 specter with 1065G7, optane cache 4k OLED display for my wife because it was the best fit at the time

My own rig is 3700X desktop. I use a 2018 Coffee Lake macbook pro (15") and i'm currently evaluating a m1 mac mini at work (and will probably buy 14" M2 macbook pro when it comes out in november).

So I don't consider myself a blind AMD fanboy

 

[uzzi38]

Agreed. AMD has executed flawlessly since 2017. Quite some of their execution consistency can be attributed to TSMC's excellent 7nm process, but AMD took the right steps in architecture as well.

So here we have it, 2019-20 ZEN3 architecture versus ~2015-16 IceLake core with cache redesign and doubled ring. When people post those Cinebench/power comparisons I am smiling, if anything Intel is doing damn great with the resources they have on chip.

I think Alder Lake will offer a first core designed after ZEN1 hit, and we will see what Intel built to fight resurgent AMD. On process side they can't expect to have any advantages in next 3-4 years if ever.

I am also going to also point out that Zen 4 could potentially mark the first product from AMD that's heavily affected by their financial success of Zen 2.

After all, Zen 2 launched in mid 2019. Zen 4 is coming 3 years afterwards.

2022 should be a very interesting year.

 

[Racan]

I don't know your full list of needs, but the Asus Zephyrus M16 is similar to the Razer Blade Advanced 15 but with 11% more vertical screen resolution will function as a much more productive workstation. It still has the gaming capabilities too.

ROG Zephyrus M16 | Laptops | ROG Global

rog.asus.com

Only HDMI 2.0b? Disappointing, do you know of any designs with HDMI 2.1?

 

[dullard]

Only HDMI 2.0b? Disappointing, do you know of any designs with HDMI 2.1?

Unfortunately, that is beyond my knowledge. Razor Blades tend to have HDMI 2.1, but the Razer Blade Advanced 15 posted above doesn't specify the HDMI version number.

 

[jpiniero]

Only HDMI 2.0b? Disappointing, do you know of any designs with HDMI 2.1?

Gen12 only supports 2.0b so for 2.1 you would need a design that isn't using Optimus.

 

[ondma]

I question the accuracy of these slides.

1. They mention perf/TDP for measuring “power consumption”. TDP is not power usage and the TDP for Intel and AMD parts both are 45W. Some Intel parts have a 65W TDP up mode and some AMD parts have a 54W TDP up mode.
2. I happen to know that the 11800H gets a CBr20 ST score that is higher than what is mentioned on the slide.
3. Higher power usage is fine as long as the task is completed quicker so the chip can go back to sleep. Simplified, fictional example: If chip 1 briefly consumes 100W for a task, but takes 2 seconds to complete it, while chip 2 only consumes 75W, but takes 6 seconds to complete it, which is the more efficient chip? Chip 1.

EDIT: Forgot to add, I'm not defending Intel here, but those slides do not appear to represent the product. A certain TGL-H chip I am aware of (NOT the fastest chip) scores 603 in CBr20. Wait for the reviews...

We certainly need better reviews. That second slide graphing TDP vs performance is pure garbage. Everybody knows TDP is not the same as power consumption (as you already said). Then the third slide, which is some kind of summary, concludes "Higer (sic) Performance but Higer Power", but no power numbers are shown. In fact TL appears very competitive in all the metrics except encoding, as far as performance goes. But they dont show power numbers in any of the slides, and also make it confusing by switching among different cpu models, for both AMD and Intel.

That said, I do expect Zen to be more (maybe by a lot) efficient. I mean how can it not be. It is starting out with a much newer architecture, and a state of the art process. TL, OTOH, is a refinement of an old core design, and isnt even on Intel's most efficient 10nm process, while it should have been on a refined 7nm, or maybe even 5nm.

 

[Hulk]

Ian is collecting questions for an interview with Keller, ask him!

My question: If you could take 3 things you know today but didn't know 3 years ago back in time 3 years, how would this knowledge have changed or not have changed the direction of Intel as a company?

 

[eek2121]

Why not Zephrys G14 or G15? About the same with an RTX 3080 even if needed (the weight diff might be compensated by the power brick) and also 300 hz 1080p screen if required.

I would wait for reviews regarding all core loads, temps and noise for the Tiger Lake H design before pulling the trigger, JIC

And btw, I bought a hp x360 specter with 1065G7, optane cache 4k OLED display for my wife because it was the best fit at the time

My own rig is 3700X desktop. I use a 2018 Coffee Lake macbook pro (15") and i'm currently evaluating a m1 mac mini at work (and will probably buy 14" M2 macbook pro when it comes out in november).

So I don't consider myself a blind AMD fanboy

I would prefer to go with an RTX 3060 or higher. The only ones I saw on their side were 3050ti models. I am going to wait a bit for reviews, obviously, but right now the new Razer laptop leads the pack. The 16" ASUS model is tempting, however.

Sigh, I know that, I've even posted it ad nauseaum here. Yet Intel cancelled Cannon Lake and added multiple Coffee Lake dies and Comet Lake (limited design effort but still layout costs, packaging improvements that neede to be developed, validation costs) and Rocket Lake. Thry also cancelled Rocket Lake mobile btw.

If Golden Cove would have been mostly "done" in 2017/2018 (other than layout specifics and validation) it would definitely been easier to redesign Tiger Lake than to develop Rocket Lake.

Maybe Intel just didn't feel threatened enough to bring it forward? I'm more inclined to think, it woukd just not have been ready.

Which brings me to the point ... Alder lake comes out just as soon as (a post 2018) Intel possibly could pull off. They don't have a backlog of chips anymore

When I say cancellations, I mean COMPLETE cancellations. as in Rocket Lake getting completely canceled. If only a partial launch happens, Intel can usually still make bucket loads.

1. Cannon lake was launched: Intel's 10nm Cannon Lake and Core i3-8121U Deep Dive Review (anandtech.com) They lost money on this, but it allowed them to get their 10nm process whipped into shape.
2. Rocket Lake was launched, it was just mobile that was withheld. Intel will make more than enough money from Rocket Lake Desktop (and there are Rocket Lake Xeons btw) to cover that cost. Rocket Lake will be incredibly profitable for Intel, even if it is a crap product.

From what we have seen in Leaks, Golden Cove is a power hungry chip, but it's fast. A 19-20% rumor is floating around, but I'm wondering if it may be faster than rumors indicate. Current "leaks" have the highest end Sapphire Rapids SKUs pegged with a 350W+ TDP. Zen 4, coincidentally, ALSO appears to be a power hungry chip. Their next gen EPYC SKUs are also rumored to have a 350W+ TDP.

It is worth nothing also that Intel IS still innovating in the PC space. They will launch the first CPU with more than 8 cores in a laptop configuration, for example. They also have some of the first hybrid/big.little designs in the desktop PC space.

I actually question AMD's ability to stay ahead when ADL-S rolls out. Especially when rumors say that Zen 4 isn't coming until the end of 2022. I do not expect AMD to pull ahead at all for the next 2 years unless Zen 4 is an absolute monster of a chip or they have a secret product in between. That is a good thing, however. Having performance parity between the two companies means that prices will mostly remain in check.

 

[Exist50]

Zen 4, coincidentally, ALSO appears to be a power hungry chip. Their next gen EPYC SKUs are also rumored to have a 350W+ TDP

Almost certainly because of higher core counts, rather than anything particular to Zen 4.

 

[uzzi38]

I would prefer to go with an RTX 3060 or higher. The only ones I saw on their side were 3050ti models. I am going to wait a bit for reviews, obviously, but right now the new Razer laptop leads the pack. The 16" ASUS model is tempting, however.



When I say cancellations, I mean COMPLETE cancellations. as in Rocket Lake getting completely canceled. If only a partial launch happens, Intel can usually still make bucket loads.

1. Cannon lake was launched: Intel's 10nm Cannon Lake and Core i3-8121U Deep Dive Review (anandtech.com) They lost money on this, but it allowed them to get their 10nm process whipped into shape.
2. Rocket Lake was launched, it was just mobile that was withheld. Intel will make more than enough money from Rocket Lake Desktop (and there are Rocket Lake Xeons btw) to cover that cost. Rocket Lake will be incredibly profitable for Intel, even if it is a crap product.

From what we have seen in Leaks, Golden Cove is a power hungry chip, but it's fast. A 19-20% rumor is floating around, but I'm wondering if it may be faster than rumors indicate. Current "leaks" have the highest end Sapphire Rapids SKUs pegged with a 350W+ TDP. Zen 4, coincidentally, ALSO appears to be a power hungry chip. Their next gen EPYC SKUs are also rumored to have a 350W+ TDP.

It is worth nothing also that Intel IS still innovating in the PC space. They will launch the first CPU with more than 8 cores in a laptop configuration, for example. They also have some of the first hybrid/big.little designs in the desktop PC space.

I actually question AMD's ability to stay ahead when ADL-S rolls out. Especially when rumors say that Zen 4 isn't coming until the end of 2022. I do not expect AMD to pull ahead at all for the next 2 years unless Zen 4 is an absolute monster of a chip or they have a secret product in between. That is a good thing, however. Having performance parity between the two companies means that prices will mostly remain in check.

For 56 Golden Cove cores with SPR at 350W vs 96 Zen 4 cores (also top end SKU is rumoured to be 320W for Genoa, not 350W+).

After what we've seen of Tiger Lake-H, I'm skeptical that SPR will be particularly more efficient than Milan as well now. Which is a shame because I was expecting Intel to the lead in power efficiency, even if that lead is primarily brought about by Milan's awful uncore power eating up nearly 1/2 the 280W power budget.

And you're saying you don't see AMD having much of a chance even with Zen 4, but I disagree. I think in the server space Intel has no real chance for starters. SPR is nice but pulling off a 20% lead or so vs Milan is a bad position to be in with Genoa mere months away.

On the desktop, Intel's top end Alder Lake SKU will compete against the 5900X in MT performance. Even without Zen 4 AMD have one way they're currently untouchable in (by Intel at least). As for gaming, well Tiger Lake only sees very moderate gains in gaming performance vs Comet Lake from initial testing. With no major changes to cache layout (apart from perhaps some timing improvements?) I don't expect Alder Lake to mix things up much. Rather, I have concerns about Windows scheduler and big.LITTLE, and the same IMC as Rocket Lake is also concerning.

Mobile is Intel's real chance to shine. Competing against Rembrandt only helps a lot here. Particularly in larger laptops as ADL-P should be very capable of extremely strong MT performance, so it'll be very, very solid for productivity. Gaming concerns carry over for me. Rembrandt will only be a modest CPU improvement, although PCIe Gen4x8 should help out a fair bit with higher power GPUs.

As for thin and lights, I dunno. ADL-P/M will certainly have a ST lead, and the little cores are nice and all, but now that Rembrandt should clear up AMD's big SoC issue in lack of Thunderbolt (via USB4) and the iGPU will a big step up vs Xe.

Oh and I know you joked about the whole secret product thing, but Zoo over on Chiphell - one of the very few reliable leakers in regards to AMD products has already hinted that we should stay tuned for a roadmap update at some point and that AMD would have something that would absolutely crush ADL and co in gaming etc even without Zen 4. Although I have almost 0 confidence in that rumour playing out. Zoo is the only reason I'm even considering it.

 

[Exist50]

Something I've been wondering about regarding both Tiger Lake and Sapphire Rapids is how does FIVR factor into TDP? You'd think it would be included, at least from the perspective of TDP = cooling requirement, but then that would skew die level power consumption vs system level power consumption. Don't think anyone's done a good breakdown of this yet, nor have I seen anything from Intel on it.

 

[Ajay]

Eh, if Zen 4 launches mid-ish '22, it could be a full year until we see a proper desktop competitor from Intel. By which point Zen 5 might not be far away.

I think Apple's A15 is using the same N5P as AMD. This will bottleneck Zen4 chiplet output; so AMD is wise to wait.

 

[Exist50]

I think Apple's A15 is using the same N5P as AMD. This will bottleneck Zen4 chiplet output; so AMD is wise to wait.

Most of the A15 volume should be in the year before Zen 4 launches, starting roughly this summer. N4 might also interfere a bit, but they might as well launch what they can when it's ready.

 

[Gideon]

When I say cancellations, I mean COMPLETE cancellations. as in Rocket Lake getting completely canceled. If only a partial launch happens, Intel can usually still make bucket loads.

1. Cannon lake was launched: Intel's 10nm Cannon Lake and Core i3-8121U Deep Dive Review (anandtech.com) They lost money on this, but it allowed them to get their 10nm process whipped into shape.
2. Rocket Lake was launched, it was just mobile that was withheld. Intel will make more than enough money from Rocket Lake Desktop (and there are Rocket Lake Xeons btw) to cover that cost. Rocket Lake will be incredibly profitable for Intel, even if it is a crap product.

I agree on most of what you said, but most definitely not on Cannon Lake help in whipping 10nm in shape. The only reason it was released was to not get sued by the stockholders as they had commited a while ago to "ship for revenue" 10nm in 2017. Ian himself has all but confirmed it.

Not only didn't they have a working GPU and were slower than Skylake, they built an absurdly limited amount of them. You couldn't get them in retail, not even in China. Only chinese schools working with a certain distributor had one model. In the end they erased the original 10nm entirely making Ice-Lake the first iteration (which was actually 10nm+).

There is nothing in such quantities to qualify as "whipping in shape". The only reason for it was to get them out of lawsuits.

From Ian's article:

While officially 'shipping for revenue' by 31 December 2017, the only way we knew to get hold of an Intel 10nm x86 CPU was if you happened to be a Chinese school and work with a specific distributor to buy a specific laptop.We pulled in a few favors from within the industry and managed to source the laptop for review.

 

[uzzi38]

I agree on most of what you said, but most definitely not on Cannon Lake help in whipping 10nm in shape. The only reason it was released was to not get sued by the stockholders as they had commited a while ago to "ship for revenue" 10nm in 2017. Ian himself has all but confirmed it.

Not only didn't they have a working GPU and were slower than Skylake, they built an absurdly limited amount of them. You couldn't get them in retail, not even in China. Only chinese schools working with a certain distributor had one model. In the end they erased the original 10nm entirely making Ice-Lake the first iteration (which was actually 10nm+).

There is nothing in such quantities to qualify as "whipping in shape". The only reason for it was to get them out of lawsuits.

From Ian's article:

Not only did they attempt to erase the original 10nm from people's minds, they also took down the product page for Cannon Lake as well. The 8121U is still up, but click on the "products formerly Cannon Lake" and you get a 404.

Product Specifications

quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

 

[coercitiv]

Something I've been wondering about regarding both Tiger Lake and Sapphire Rapids is how does FIVR factor into TDP? You'd think it would be included, at least from the perspective of TDP = cooling requirement, but then that would skew die level power consumption vs system level power consumption. Don't think anyone's done a good breakdown of this yet, nor have I seen anything from Intel on it.

It has to be included from a TDP perspective, there's no going around that. However, TDP increase from FIVR shouldn't be a problem from the user's perspective. As long as heat dissipation at socket level isn't an issue, the savings we get in terms of simplified power delivery at the motherboard level are definitely worth it.

 

[Exist50]

It has to be included from a TDP perspective, there's no going around that. However, TDP increase from FIVR shouldn't be a problem from the user's perspective. As long as heat dissipation at socket level isn't an issue, the savings we get in terms of simplified power delivery at the motherboard level are definitely worth it.

I'm thinking more in terms of what is being measured when comparing between Intel and AMD. Or even Comet Lake to Tiger Lake. Package power vs platform power, basically.

 

[coercitiv]

I'm thinking more in terms of what is being measured when comparing between Intel and AMD. Or even Comet Lake to Tiger Lake. Package power vs platform power, basically.

If you want to normalize power usage between 2 different CPUs then you have to account for everything different inside the package. This includes both voltage regulation and chipset features. I believe the H platform still uses an external PCH.

 

[IntelUser2000]

Hmm, how large is each Willow Cove core + L3 in Tiger Lake?

In Tigerlake-U, the 4 cores take up 45mm2 or so. Tigerlake-H is already 190mm2. Just the addition of full Xe LP will make it 210mm2. The Gracemont clusters themselves are going to add 20-25mm2 minimum.

Alderlake may end up being large as Rocketlake in die size. Remember, Golden Cove cores are likely going to be larger.

If Gracemont cores were only half the size of Golden Cove cores like some thought it was a while ago based on a block diagram, it'll easily exceed 300mm2. That's the most convincing argument why the cores themselves must be tiny.

I agree on most of what you said, but most definitely not on Cannon Lake help in whipping 10nm in shape. The only reason it was released was to not get sued by the stockholders as they had commited a while ago to "ship for revenue" 10nm in 2017. Ian himself has all but confirmed it.

I don't think so. Nothing substitutes real world experience. So by actually getting it to a point where they can ship few units, they'd get actual feedback on what to do.

Marketing wise it may have been questionable but screw them. They have too much power in modern mega corporations.

 

[jpiniero]

My assumption is that a Gracemont cluster is roughly the size of a Golden Cove core.

I wonder if they will keep the IPU in Alder Lake-P.

 

[IntelUser2000]

My assumption is that a Gracemont cluster is roughly the size of a Golden Cove core.

That's my guess too.

I wonder if they will keep the IPU in Alder Lake-P.

They've enhanced the IPU IP on Tigerlake-H. They haven't done it on previous -H products.

I actually question AMD's ability to stay ahead when ADL-S rolls out. Especially when rumors say that Zen 4 isn't coming until the end of 2022. I do not expect AMD to pull ahead at all for the next 2 years unless Zen 4 is an absolute monster of a chip or they have a secret product in between.

I actually believe that leak where Zen 4 is 30% faster per clock. Genoa will end up being twice as fast as Milan which will keep everyone in check(ARM, Power, Intel).

 

[IntelUser2000]

Some of the modules look a bit spaced out where the zen3 ccx was unceremoniously dumped into the renoir soc design.
Going forward I see it as a good thing where the companies can design products much more iteratively, see all the various mobile cpus amd has planned with no easily apparent consistency.

Empty silicon has nothing to do with Keller. It's simply a result of having flexibility in design.

We're dealing with billions of transistors and billions of connections here. Yes some are automated but there's significant input by the engineers. When you design in a way that you can easily add/subtract blocks, you'll waste some space but you decrease time-to-market.

Sure you can spend extra months adding blocks to that space, but why?

 

[jpiniero]

They've enhanced the IPU IP on Tigerlake-H. They haven't done it on previous -H products.

I'm pretty sure this is the first time H has gotten the IPU.

 

[Hulk]

I actually believe that leak where Zen 4 is 30% faster per clock.

I would be (happily) astounded if this turns out to be true.