Discussion Intel current and future Lakes & Rapids thread

[IntelUser2000]

Tiger Lake-U recently was said to have shipped 4 million units. Maybe as of a month ago? So if Intel already has 1 million TigerLake-H shipped to OEMs and if a significant percentage of that is 8c, then the yields may actually be improving to the point that they aren't completely embarassing.

40 million not 4.

 

[beginner99]

Sucks that you can’t use the unlocked i9-11980HK in a desktop board. Just sad, really. Who needs it in a laptop that sounds like a jet plane? Intel is mad, really. No common sense at all.

Why make a product people want if you can't ship it? This way they keep up appearances and people don't directly notice how little of these they can actually make.

 

[eek2121]

Chinese review vs Cezanne.

https://www.reddit.com/r/Amd/comments/n9zmsu

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...

 

[Accord99]

Wonder if that's down to a difference in silicon quality, intra-SKU binning (even within each SKU line, AMD/Intel/Nvidia offer differently bins of products to OEMs) or CB R20 and R23 both being more AVX and AVX2 heavy that allows TGL-H to catch up.

Given the low performance increase at 25-35W compared to 4-core Tiger Lake at similar TDP and the relatively linear increase in TDP and performance from 45W+, I'm more inclined to believe that it's not decreasing voltage as one would expect at lower multipliers.

 

[eek2121]

They might not perform as well there as in some other benchmarks but their power draw has been totally average in cinebench flr a AVX2 load. Not looking great whichwver way you slice it.

Now obviously more data is required before drawing any conclusions, but all-core power is not looking competitive:

https://twitter.com/x/status/1392084542312394752

16 - 19% multicore uplift (on a marketing slide!) from 2 generations of uarch and 10nm SF?

AMD pretty much doubled MT perf @ same power draw going from 14 to 7nm ...

You have that backwards. It took AMD 4 generations and 3 node jumps (14nm, “12nm”, 7nm) to beat Intel “14nm” in performance.

 

[tomatosummit]

They seem to have a considerable amount of dark silicon, where the GPU used to be. I wonder why couldn't they just add a couple of EUs instead:

https://twitter.com/x/status/1392088812600365057

I think that dark silicon is the real symptom of Keller. Under him both amd and now intel said that their design process was broken down so teams can work independantly on various modules within IC design and how it applies to dies with less time. Both tgl-h and rocketlake have been relatively quick to respond to the market needs.
For tigerlake-H it looks like the block in the top right (display and fixed function?) couldn't be shrunk down, possibly the pinout phy and I've not seen anything XE that's more granular that 32eus. So it's 32 and black space or 64 and pushing 200mm^2.
Rocketlake die looks like it has some black space next to the memory phy. Perhaps the main soc was designed in parallel to the backport cores and when mated there wasn't enough time to save the last ~5% of die space.
Cezanne as well. 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.

 

[Gideon]

You have that backwards. It took AMD 4 generations and 3 node jumps (14nm, “12nm”, 7nm) to beat Intel “14nm” in performance.

I was talking about AMD going from GlobalFoundries 14nm to TSMC 7nm, where they managed to deliver 2x the power efficency in the optimal power range (2.5-3.5 Ghz) being able to deliver chips with 2x the cores running at approximately the same clocks at the same total power draw.

Granted that's with putting more cores but the fact remains that I expected more Going from:

1. 2015 era Skylake to (originally ~2018 era) TigerLake
2. 14nm +++ to 10nm SF (10nm++ in original notation)

According to Andrei F. in a more power constrained environment the MT perf increase is: 11950H vs 10885H = +16% (on a marketing slide whic is usually optimistic)

AMD managed about the same going from 2700X -> 3700X in a single generation of uarch and process and also dropping peak power by 25% (117W -> 90W) while using a hugely power-inefficient MCM I/O die solution. The difference in power would be much bigger if both were monolithic chips.

Intel just jumped from 2015 to the best they could do in 2020. It's perfectly reasonable to expect more than 16%

 

[Asterox]

If 80-90% people buy cheep or cheeper Prebuilt PC and use it on defoult setings hm.Well, whose fault is it.

 

[jpiniero]

If 80-90% people buy cheep or cheeper Prebuilt PC and use it on defoult setings hm.Well, whose fault is it.

Totally shocking that a power hungry CPU is faster at 140 W compared to 65 W.

 

[uzzi38]

You have that backwards. It took AMD 4 generations and 3 node jumps (14nm, “12nm”, 7nm) to beat Intel “14nm” in performance.

I prefer phrasing things a different way.

It took AMD 4 generations and 3 node jumps of solid execution to overthrow a decade and a half of Intel's complete leadership in performance in all possible aspects. 3 generations and 3 node shrinks if it's just performance leadership.

 

[rainy]

You have that backwards. It took AMD 4 generations and 3 node jumps (14nm, “12nm”, 7nm) to beat Intel “14nm” in performance.

With all respect but Zen+ in 12nm (which is tweaked 14nm) should be not counted as generation/node.

Btw, yes, you are defending Intel even if you're saying opposite.

 

[JoeRambo]

It took AMD 4 generations and 3 node jumps of solid execution to overthrow a decade and a half of Intel's complete leadership in performance in all possible aspects. 3 generations and 3 node shrinks if it's just performance leadership.

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.

 

[eek2121]

With all respect but Zen+ and 12nm (which is tweaked 14nm) should be not counted as generation/node.

Btw, yes, you are defending Intel even if you're saying opposite.

No, I'm saying technology is hard, and AMD is hardly beating Intel by any significant margin. Keep in mind I own AMD stock and all my PCs are AMD based. This does not mean I am unable to have an objective conversation about Intel. I am going to purchase a laptop this year, and given what I've seen thus far, Tiger Lake H and Cezanne are competitive, which means it is going to come down to specs and build quality.

Currently the most promising laptop out there for my needs appears to be the new Razer Blade Advance 15 that is going on preorder. There is not a single AMD or Intel based laptop that is comparable to it.

1. Core i7-11800H (8 cores)
2. 240hz QHD panel (not 4k, but not 1080p)
3. RTX 3060
4. 4.4 lbs/1.99 kg

I need something that can function as both a workstation (hence the 8 cores) and a mobile gaming machine. I would also prefer something a bit lighter/slower over one of those giant 5-7 pounder as I will be carrying it around when travelling. AMD does not have a similar offering from what I've seen thus far. I prefer a higher resolution screen for software development and other workstation related tasks.

EDIT: Ideally I would prefer for it to be a bit lighter, but we can't have everything.

 

[dullard]

Currently the most promising laptop out there for my needs appears to be the new Razer Blade Advance 15 that is going on preorder. There is not a single AMD or Intel based laptop that is comparable to it.

1. Core i7-11800H (8 cores)
2. 240hz QHD panel (not 4k, but not 1080p)
3. RTX 3060
4. 4.4 lbs/1.99 kg

I need something that can function as both a workstation (hence the 8 cores) and a mobile gaming machine. I would also prefer something a bit lighter/slower over one of those giant 5-7 pounder as I will be carrying it around when travelling. AMD does not have a similar offering from what I've seen thus far. I prefer a higher resolution screen for software development and other workstation related tasks.

EDIT: Ideally I would prefer for it to be a bit lighter, but we can't have everything.

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

 

[Gideon]

I prefer phrasing things a different way.

It took AMD 4 generations and 3 node jumps of solid execution to overthrow a decade and a half of Intel's complete leadership in performance in all possible aspects. 3 generations and 3 node shrinks if it's just performance leadership.

14nm zen1 is a relatively decent apples to apples starting point to skylake (while being 1.5 years late).

12nm is a marketing name, its very similar to intel's 14nm +(+) nodes in practice so i wouldn't really count it. Zen+ is also just zen with full (instead of partial) turbo implementation and origikal design latencies.

And all of this completely misses the point that in doing the aforementioned AMD also beats everything intel could design before 2020.

And the whole "intel has designs ready but can't ship due to process" argument is also long since depleted. Intel has 10nm SF out for a while. If Alder lake was only held back by process they could have just skipped Tiger Lake, they didn't, so they are all caught up now.

Alder lake is the best they can do in (very late) 2021, no designs waiting in the drawer anymore ...

 

[jpiniero]

If Alder lake was only held back by process they could have just skipped Tiger Lake

You have to figure yields into it. Alder Lake-P is going to be over 200 mm2 easy.

 

[moinmoin]

AMD is hardly beating Intel by any significant margin.

Well, that's great news to Intel if you say so. But for Intel's sake I sure hope they don't rely on such impressions.

 

[Panino Manino]

I think that dark silicon is the real symptom of Keller. Under him both amd and now intel said that their design process was broken down so teams can work independantly on various modules within IC design and how it applies to dies with less time.

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

 

[uzzi38]

You have to figure yields into it. Alder Lake-P is going to be over 200 mm2 easy.

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

 

[Gideon]

You have to figure yields into it. Alder Lake-P is going to be over 200 mm2 easy.

That is true. But if Golden Cove core design was just sitting in a drawer since 2018 as some claim, they would have had plenty of time to redesign Tiger Lake with newer CPU cores

 

[eek2121]

That is true. But if Golden Cove core design was just sitting in a drawer since 2018 as some claim, they would have had plenty of time to redesign Tiger Lake with newer CPU cores

People always go off the deep end regarding AMD/Intel product launches. It takes a significant amount of time (years) and money (many millions) to develop and launch a product. You don't simply "skip" or "cancel" a launch. You also don't fast track launches. Decisions regarding Rocket Lake, Tiger Lake, and Alder Lake were made 2+ years ago. The only thing that affects timing is node availability and unforeseen issues. Skipping Rocket Lake, for example, would have costed Intel a LOT of wasted time and money (and money equals time, so there is that). Intel would never cancel Rocket Lake because it runs hot. Canceling the product would cost more than launching it, and if they launch it, chances are good they will profit from it.

 

[eek2121]

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

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.

 

[RTX]

Is SF a completely new design compared to 10nm+/10nm++ and 14nm+/14nm++/14nm+++ ( increased fin pitch with each + generation vs other optimizations? )
In ESF they say they increased mim capacitor performance in one of the slides for Alderlake.

 

[Gideon]

People always go off the deep end regarding AMD/Intel product launches. It takes a significant amount of time (years) and money (many millions) to develop and launch a product

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

 

[B-Riz]

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.

Ah, I don't think Intel is doing that great for being an x86 leader for years, they just have cheap chips now.

Cheap does not mean a good value.

Intel eff'd up so bad they are throwing crap together and acting like it is great (11900K?).