Discussion Intel current and future Lakes & Rapids thread

[uzzi38]

China has had them for a month now? Wrong thread for Renoir though. More interested in when TigerLake-U is out, at least in this thread.

Computex announcement, products 1-2 months later, depending on how human malware is tackled.

 

[Exist50]

Computex announcement, products 1-2 months later, depending on how human malware is tackled.

More likely to be former Computex timeframe, considering the postponement.

 

[uzzi38]

More likely to be former Computex timeframe, considering the postponement.

Ah yeah, that's what I meant.

 

[IntelUser2000]

There's no need to speculate. Back in January(no hype about the virus) when Ryzen Mobile 4000 launched, all vendors were pointing at March/April as the release date. So it makes sense why we're seeing it now.

If next gen launches exactly a year from Renoir, we'll see it at March/April 2021.

On another note, again no detail on Xe!

 

[Dayman1225]

There's no need to speculate. Back in January(no hype about the virus) when Ryzen Mobile 4000 launched, all vendors were pointing at March/April as the release date. So it makes sense why we're seeing it now.

If next gen launches exactly a year from Renoir, we'll see it at March/April 2021.

On another note, again no detail on Xe!

https://twitter.com/x/status/1241067782168440832

 

[mikk]

If next gen launches exactly a year from Renoir, we'll see it at March/April 2021.

This is always the ideal case. Picasso was available in February last year, Renoir will be available in April (probably), so it is more than a year for this generation. The next update could easily slip into a timeframe around June Computex with regards to availability which makes a paper launch in January during CES implausible because the time gap from paper launch to hard launch is too big.

 

[maddie]

This is always the ideal case. Picasso was available in February last year, Renoir will be available in April (probably), so it is more than a year for this generation. The next update could easily slip into a timeframe around June Computex with regards to availability which makes a paper launch in January during CES implausible because the time gap from paper launch to hard launch is too big.

AMD stated that they will release models in time periods suited for the mobile OEMs. This pretty well means annual launches at around the same time period. This is what our clients want, so we will design and release to suit them.

 

[Dayman1225]

A Look at Intel Lakefield: A 3D-Stacked Single-ISA Heterogeneous Penta-Core SoC - Wikichip

Looks like the leaked Lakefield die shot was real as confirmed by this article. Intel also released die sizes and transistor count for base and compute dies on Lakefield

Compute Die (10nm+) comes in at 82mm^2 with 4050000000 Xtors which is roughly 50MTr/mm^2

Base Die (22nm FFL) comes in at 92mm^2 with 650000000 Xtors which is roughly 7MTr/mm^2

Another interesting tidbit a singular Tremont core is around 0.88mm^2 and Intel claims it can reach around 70% of the performance of a SNC core

 

[DisEnchantment]

Compute Die (10nm+) comes in at 82mm^2 with 4050000000 Xtors which is roughly 50MTr/mm^2

In comparison to TSMC's N7 at 62.82 MTr/mm^2 in Renoir. Although just SRAM would indeed be a lot denser in both cases.

 

[mikk]

Compute Die (10nm+) comes in at 82mm^2 with 4050000000 Xtors which is roughly 50MTr/mm^2

Is it the marketing 10+ (Tigerlake) or the real 10+ (Icelake). 2nd-generation 10-nanometer implies it's 10+ from Icelake which makes sense (Sunny Cove+Gen11).

 

[ksec]

I wonder what is Lakefield's target customer? It doesn't look cheap with its packaging, and if it isn't cheap enough it is fighting with 4 Core Tiger Lake.

 

[bullzz]

Microsoft Surface PCs, Computers, Laptops, 2-in-1s, Dual-Screen & All-in-Ones | Microsoft Surface

Discover a range of Microsoft Surface devices built for all the ways you create, work, play and connect. Unleash your creativity with Microsoft Surface laptops, dual screens, 2-in-1 PCs and All-in-One devices.

www.microsoft.com

I wonder what is Lakefield's target customer? It doesn't look cheap with its packaging, and if it isn't cheap enough it is fighting with 4 Core Tiger Lake.

 

[RetroZombie]

I wonder what is Lakefield's target customer? It doesn't look cheap with its packaging, and if it isn't cheap enough it is fighting with 4 Core Tiger Lake.

And the arm guys with 8 cores.

 

[IntelUser2000]

I wonder what is Lakefield's target customer? It doesn't look cheap with its packaging, and if it isn't cheap enough it is fighting with 4 Core Tiger Lake.

One thing current Intel chips have lot to improve on it battery life. There's still a 30-50% gap.

Now I know they can close the gap. They have done it with Bay/Cherry Trail. Also the Gemini Lake chips do better.

If they do well on scheduling both of the cores, then we'll get 15W U level ST performance at 3-5W TDP its targeting and better integration along with being able to run flexibly depending on workload means better battery life. That's way better than what the Y chips are doing. It's about 30% behind U chips in ST. Don't forget the big space savings it allows due to integration.

Like I said before, this is a proper return to Tablet chips since they abandoned it with Broxton in 2015. You can do it with Core, but its very clunky. Tablets based on Core weigh much more while the subpar integration sacrifices battery size and design flexibility. Core addresses 13 inch-plus systems while Lakefield will go down to 8-10 inches like every other Tablet, or have a very portable clamshell.

Its a bit unfortunate its taking so long to get Lakefield out. Hopefully future generations improve on timing so to catch up with Core timelines. I believe despite the rocky start this will bring on a new future for x86.

(Also this should result in far better Core M than Core M ever was. Boy that was a disappointment for me. All that marketing hype for what seemed like different binning. It didn't deserve the Core "M" moniker)

 

[RetroZombie]

I believe despite the rocky start this will bring on a new future for x86.

Great analysis, and you painted a very pretty and rosy scenario for intel.
The problem is intel abandoned it because the arm guys did all that much better at that time, let's not even talk today.
I cant see where intel (or even amd) can have a upper hand in a market where the smartphones guys dominate.

 

[SAAA]

I wonder what is Lakefield's target customer? It doesn't look cheap with its packaging, and if it isn't cheap enough it is fighting with 4 Core Tiger Lake.

The ultramobile market would have been, if this thing were out a couple years ago and tablets were still a trend rather than at saturation level.
Now I do expect a couple decent but otherwise poor welcomed devices who will use this SOC then die off like most revolutionary projects. Folding screen? Pfft, surface neo with two distinct ones will be perfect but stay on the shelves for this and that reason, price won't help too.

And the arm guys with 8 cores.

It's not going to be much behind with 4+1 cores, Tremont itself should be comparable to current ARM cores aside from Apple's:

The right graph tells you the sunny cove core is worth 2 Tremonts so MT will be worth 6 small cores total and ST will punch 50% higher.
Not a bad combo honestly, when the die area used it that of 7.5 to 8 small cores. The bad part is the missing features like AVX etc, probably a thing worth doing only for ultraportable devices.

 

[ksec]

Well I would be very surprised if it was for Tablet. But Surface Neo seems interesting. Ignoring Apple, most tablet's SoC cost less than $30. With Mediatek's going below $20. Intel® Celeron® N4120 is going for around $100, for manufacturing and rebate we know it could go down to sub $40 as with previous Gemini Lake. But I thought Intel abandoned the idea of low profits low margin SKUs, but At 80mm2+ Die Size with State of the Art packaging I just dont see this being sold at this price point.

 

[bullzz]

I cant see where intel (or even amd) can have a upper hand in a market where the smartphones guys dominate.

Lakefield is not targeting smartphones though. It is going after premium tablet market. that market is owned by apple. even qualcomm is not doing great in it. check out reviews of MS Surface Pro X reviews.

Well I would be very surprised if it was for Tablet. But Surface Neo seems interesting. Ignoring Apple, most tablet's SoC cost less than $30. With Mediatek's going below $20. Intel® Celeron® N4120 is going for around $100, for manufacturing and rebate we know it could go down to sub $40 as with previous Gemini Lake. But I thought Intel abandoned the idea of low profits low margin SKUs, but At 80mm2+ Die Size with State of the Art packaging I just dont see this being sold at this price point.

It is a myth that high end smartphones chips cost $30. qualcomm and samsung use leading edge process and their die sizes are not small either. in some ways snapdragon chips are more complex that apples. chinese manufacturers are exceptions since they use off-the-shelf logic from ARM and do not use leading edge process.

 

[DrMrLordX]

@bullzz

Some of the smaller-market cell phone companies are skipping Snapdragon 865 for now due to cost (and maybe the heat output, too). Huawei opted to re-release an A76-based SoC (Kirin 990) instead of adopting A77 like Qualcomm, presumably to reduce costs and heat.

 

[RetroZombie]

Lakefield is not targeting smartphones though. It is going after premium tablet market. that market is owned by apple.

Yes. But those smartphones cpus, some of them at least can scale up to service the premium tablet market, they just cant go desktop* yet.
For the same reason desktop chips from intel can service the mobile market, almost the tablet* and never the smartphone*.

*Thinking overall here, anyone can use any chip in anywhere they want even if it ends a fluke.

 

[IntelUser2000]

The right graph tells you the sunny cove core is worth 2 Tremonts so MT will be worth 6 small cores total and ST will punch 50% higher.
Not a bad combo honestly, when the die area used it that of 7.5 to 8 small cores. The bad part is the missing features like AVX etc, probably a thing worth doing only for ultraportable devices.

This is why when Gracemont arrives it'll be much better. I'm counting on the "improved vector performance" to mean having AVX support.

Full support is dependent on software though. Windows 10 X* is supposed to have one for the hybrid chips. Lenovo said for the Fold, initially it'll be Windows 10, with X support coming later. This is why Surface Neo is coming late this year(nothing to do with using Lakefield-R).

I'm not a big fan of dual screens or folding screens. It increases cost a lot and it reduces battery life. I hope with Lakefield we'll see super portable 2-in-1s such as was possible with older Atoms.

*10 X also looks much better for UI. The tile interface is crap. It takes too much space on the screen, making you scroll too much. Ironically it was Android/iOS that optimized the Windows approach of using icons into mobile. Thankfully they are back to using icons for 10 X.

 

[coercitiv]

This is why Surface Neo is coming late this year(nothing to do with using Lakefield-R).

Microsoft reportedly delays Surface Neo beyond 2020

The Surface Neo, Microsoft’s dual-screen PC that runs a new operating system called Windows 10X, will miss its planned release date of holiday 2020, according to reports from ZDNet and CNBC. The company is said to have shifted its focus to get Windows 10X to run well on single-screen devices like laptops and 2-in-1s.

 

[mikk]

https://twitter.com/x/status/1248188783578591233

It could be shocking to some people, there is a slide which says Alder Lake Desktop 10nm. Apart from this nothing new though.

 

[jpiniero]

It could be shocking to some people, there is a slide which says Alder Lake Desktop 10nm. Apart from this nothing new though.

Given that it's release would be two years from now, you would think they would use 7 nm.

 

[coercitiv]

That twitter thread now contains translated content for both slides:

Slide 1

Moore's law evolution: process returns to two-year cadence.

10NM

• substantial yield improvement
• substantial increase in production capacity
• series of 10nm products launching in 2020

7NM

• lead product launching in 2021
• complete product portfolio in 2022
• performance improvement every year

Slide 2

A series of new products coming in 2020

ALDER LAKE: 10nm Desktop Processor
TIGER LAKE: 10nm Core Mobile Processor
DG1: First Discrete Graphics Card based on Xe Architecture
ICE LAKE: 10nm Xeon Scalable Processor
SNOW RIDGE: First 5G-Ready 10nm SoC for Base Station

Given that it's release would be two years from now, you would think they would use 7 nm.

If these slides are true, then Intel is trying to launch a 10nm desktop product in 2020.