Discussion Intel current and future Lakes & Rapids thread

[IntelUser2000]

It is truly astonishing what AVX512 can bring on a modern efficient design."

I doubt its AVX512 that's making it that much faster compared to Icelake. It's the ability to hold clocks better in multi-threaded workloads.

 

[itsmydamnation]

I doubt its AVX512 that's making it that much faster compared to Icelake. It's the ability to hold clocks better in multi-threaded workloads.

thats the same difference really, AVX512 clock was slow.

 

[IntelUser2000]

thats the same difference really, AVX512 clock was slow.

That's true for Skylake server with the extra 512-bit unit, but on client and in Icelake it only has 2x256 units.

Ice Lake AVX-512 Downclocking

Examining the extent of AVX related downclocking on Intel’s Ice Lake CPU

travisdowns.github.io

Of course I'm pretty sure 10nm helps a bit.

Along this axis, there is almost no downclocking at all! Only for a single active core count is there any decrease with wider instructions, and it is a paltry only 100 MHz: from 3,700 MHz to 3,600 MHz when any 512-bit instructions are used.

Tigerlake's gains are nearly all due to the 10nm SF changes and whatever physical design changes they did.

 

[IntelUser2000]

I did use the picture without the orange outline yesterday and got 45mm².

13.3*10.6mm=140.98mm2

Some new info.

 

[coercitiv]

As for outsourcing production to TSMC: Intel just doesn't have enough production lines to make everything themselves.
It's absolutely a bummer and one of the reasons for worse forecasts and stock drop.

AFAIK the TSMC outsourcing is done to compensate for the 7nm delays, not for 10nm capacity. Having capacity problems that are not yield related would not lead to stock price drops, they'd be drowning in high margin revenue.

 

[coercitiv]

I don't see it as coincidence reviewers receiving 4800U laptop samples just a week before Tigerlake launch.

We'll also be seeing the 4900H against TGL, it's already announced in the Lenovo Sim 7 Pro (14" screen, 1.45kg). Throughput will likely be similar to 4800U due to power/thermal limits but ST performance will jump up a bit.

 

[IntelUser2000]

We'll also be seeing the 4900H against TGL, it's already announced in the Lenovo Sim 7 Pro (14" screen, 1.45kg). Throughput will likely be similar to 4800U due to power/thermal limits but ST performance will jump up a bit.

It's the opposite. The 4900HS is upwards of 30% faster than the 4800U in MT. About the same in ST. The Zephyrus has been using 4900HS months ago.

 

[lobz]

TGL appears to be a great little product. Waiting to see power consumption, but the i7 performs better than a 4700U.

A lot of people are bashing Intel and making claims they can’t deliver. Clearly many of us have short memories. Time and time again Intel has been beaten, and they’ve always come back. Tiger Lake and Intel’s discrete graphics are the first indication that 10nm is beginning to ramp up. While 10nm is (extremely) late, we must keep in mind that until AMD makes the leap to 5nm, they won’t continue to have a process node advantage. With 5nm chips from AMD possibly not landing until 2H2022, Intel and AMD could very well find themselves on even footing from a process technology standpoint going forward, and believe it or not this is a good thing.

Dude, this is exclusively mobile with up to 4 cores. FOUR CORES. Nobody ever said that Intel was bad in the mobile sector. In fact, they developed super cool energy saving techniques over the years. It's just... just not as good ones as ARM vendors, but we'll have to wait and see with A14X and his buddies.

 

[coercitiv]

It's the opposite. The 4900HS is upwards of 30% faster than the 4800U in MT. About the same in ST. The Zephyrus has been using 4900HS months ago.

I was talking in the context of the slim body of the Lenovo unit. There's still a chance they managed to do an excellent job with the cooling and keep it at 35W in a 14" unit with 1.45kg weight, but I would err on the side of caution and expect under 35W TDP, 25W would be way more doable.

 

[clemsyn]

Intel Core Gen11 TigerLake ULV (i7-1165G7) Review & Benchmarks – CPU AVX512 Performance – SiSoftware

www.sisoftware.co.uk

Surprised to see this review early, but TGL looks like a beast!

Quoting from the article in case it is taken down:

"While due to high power consumption, AVX512 was never a good fit for mobile and their meager power envelopes (15W, 25W, etc.) – somehow “Tiger Lake” (TGL) manages to run them much faster, 40-50% faster than “Ice Lake” and thus beating the competition.

TGL’s performance with a meager 15W power budget in a think & light laptop is pretty compelling and it soundly beats not only older (bigger) mobile processors with more cores (at 35-45W) but also older desktop processors! It is truly astonishing what AVX512 can bring on a modern efficient design."

Looks impressive but I'll have to wait for Anandtech's review on this. I'm pretty sure they are itching to release it tomorrow. I hope they price this right so I can upgrade my dell e6430.

 

[IntelUser2000]

I was talking in the context of the slim body of the Lenovo unit. There's still a chance they managed to do an excellent job with the cooling and keep it at 35W in a 14" unit with 1.45kg weight, but I would err on the side of caution and expect under 35W TDP, 25W would be way more doable.

Cooling continues to improve. Just couple of years ago, 5W was the level for fanless.

Do you remember this?

Huawei MateBook X 2020 will feature a novel fanless design, come with an Intel CPU/Nvidia GPU combo, and could be priced under 10,000 yuan (US$1,438)

More details about the upcoming Huawei MateBook X 2020 laptop have started being shared online. It’s believed the device sports a novel cooling solution that doesn’t rely on fans. It is also rumored that the MateBook X 2020 could sport Intel CPUs and an Nvidia GPU, come in four different colors...

www.notebookcheck.net

It features Cometlake and MX250 GPU while being fanless due to the novel cooling design.

They absolutely can fit a 35W chip using a cooler in a 1.45kg design. Alternatively we should see ever thin and light designs using 15/25W chips.

Also, 4900H is a 45W part while 4900HS is 35W. If they are going to lower it to 25W why not a 4800U?

 

[piokos]

AFAIK the TSMC outsourcing is done to compensate for the 7nm delays, not for 10nm capacity. Having capacity problems that are not yield related would not lead to stock price drops, they'd be drowning in high margin revenue.

Intel's 7nm will not compete against TSMC 6N but a more dense node in the future (5N?). 6N is an improved 7N. Intel will get there with 10nm and a finite number of "+".

If Intel has something very specific in mind for their 7nm chip that isn't possible on 10nm, it probably won't be possible on 6N as well.
IMO the biggest gain for them is just adding a "safety margin" (which they currently don't have on their 10/14nm combination). This is exactly the issue that hit them during early stage of transition to 10nm: 14nm volume was limited as well and they couldn't keep up with OEM orders.

 

[uzzi38]

Intel's 7nm will not compete against TSMC 6N but a more dense node in the future (5N?). 6N is an improved 7N. Intel will get there with 10nm and a finite number of "+".

If Intel has something very specific in mind for their 7nm chip that isn't possible on 10nm, it probably won't be possible on 6N as well.
IMO the biggest gain for them is just adding a "safety margin" (which they currently don't have on their 10/14nm combination). This is exactly the issue that hit them during early stage of transition to 10nm: 14nm volume was limited as well and they couldn't keep up with OEM orders.

N6 is a significantly cheaper node than N5 with much more available volume. It makes more sense for Intel to use N6 over N5 as it does provide notable density and perf benefits over N7 variants even still. With Apple ramping up N5 volume, Qualcomm potentially still seemingly around (rumours are mixed here) and surprisingly enough AMD becoming a much more major player by the node, there's not a whole lot of volume to go around for Intel.

PVC compute dies are going onto N6 afaik, Intel are just ready to have to sacrifice some performance to ensure they get the volume they need.

 

[exquisitechar]

Intel Core Gen11 TigerLake ULV (i7-1165G7) Review & Benchmarks – CPU AVX512 Performance – SiSoftware

www.sisoftware.co.uk

Surprised to see this review early, but TGL looks like a beast!

Quoting from the article in case it is taken down:

"While due to high power consumption, AVX512 was never a good fit for mobile and their meager power envelopes (15W, 25W, etc.) – somehow “Tiger Lake” (TGL) manages to run them much faster, 40-50% faster than “Ice Lake” and thus beating the competition.

TGL’s performance with a meager 15W power budget in a think & light laptop is pretty compelling and it soundly beats not only older (bigger) mobile processors with more cores (at 35-45W) but also older desktop processors! It is truly astonishing what AVX512 can bring on a modern efficient design."

Not a very high quality review, I can see a few problems at a glance.

 

[piokos]

Cooling continues to improve. Just couple of years ago, 5W was the level for fanless.

Actually cooling didn't improve that much.
Check if the laptops you think are passively cooler really don't have fans. People tend to think that slim ultrabooks (say: Macbook Air or Asus Zenbook) are passively cooled. And most of them aren't.

Sure - we have better heatpipes and clever vapor chambers. But all they do is move heat from the chip as quickly as possible.
But from a device perspective you have the same 5W or 25W that have to be pushed through the same tiny vents.

The whole point of forced airflow in a laptop is to push hot air through the small side panels. But if you make a passively cooled laptop (or just badly designed actively cooler one), most of the heat will exit through the top panel (keyboard and palm rest), which isn't exactly improving user experience.

 

[A///]

I don't want to go back 20 pages, but wasn't TSMC developing a special 5nm node for AMD or some of their preferred clients? I can see Intel going for that N6 node for savings, and also because they just delayed a major delivery they need to get their stuff sorted and deliver on a new date.

 

[DrMrLordX]

"4c mobile chips" is Intel's most important and profitable consumer lineup.

Times change. They need 6c and 8c products on 10nm, stat. They can't do it until next year.

If they're going to make it exclusively on Tiger Lake, it clearly shows a lot of confidence in both platform and process.

No it doesn't. It shows that they can't get an 8c or even a 6c Willow Cove to yield correctly. The iGPU is enormous to compensate. Where's the much-touted TigerLake-H? Still not here.

As for outsourcing production to TSMC: Intel just doesn't have enough production lines to make everything themselves.

Yeah, that's why Murthy got canned. Sure.

AFAIK the TSMC outsourcing is done to compensate for the 7nm delays, not for 10nm capacity. Having capacity problems that are not yield related would not lead to stock price drops, they'd be drowning in high margin revenue.

It's actually because:

DG1 and DG2 were originally intended to chew up "excess" 10nm production, which never materialized because a). Intel had to burn too many 10nm wafers just to yield enough working silicon to get IceLake out the door and b). Intel slowed down the process of converting fabs from 14nm to 10nm since they couldn't effectively replace their 14nm lineup with anything on 10nm (yet). It looks like Intel may have their act together by next year - we think - but decisions like the TSMC 6nm buy were made with 2019's and 2020's problems in mind.

Remember, 10nm was supposed to be responsible for TigerLake, Alder Lake-P and -S, Sapphire Rapids, DG2, and parts of Ponte Vecchio (the parts that were not on Intel 7nm).

 

[uzzi38]

I don't want to go back 20 pages, but wasn't TSMC developing a special 5nm node for AMD or some of their preferred clients? I can see Intel going for that N6 node for savings, and also because they just delayed a major delivery they need to get their stuff sorted and deliver on a new date.

Yes and no.

Working with TSMC on nodes and new technologies is what all priority customers are doing now. AMD, Apple, you name it, they have their own customisations and tweaks, and in return they also help iron out the creases in nodes/packaging technologies too.

 

[A///]

Remember, 10nm was supposed to be responsible for TigerLake, Alder Lake-P and -S, Sapphire Rapids, DG2, and parts of Ponte Vecchio (the parts that were not on Intel 7nm).

I can't recall when ADL was announced for 2021, but I would be surprised if it comes out sooner than 4Q22.

 

[DrMrLordX]

I can't recall when ADL was announced for 2021, but I would be surprised if it comes out sooner than 4Q22.

Alder Lake is supposed to be a 2021 CPU according to some Intel roadmaps. Not saying I agree with it, it's just . . . that's what's supposed to happen. If Intel can't launch Alder Lake next year, then I really have no idea what they're going to do. Tiger Lake Refresh?

edit: Alder Lake in 2021:

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

www.anandtech.com

 

[A///]

Alder Lake is supposed to be a 2021 CPU according to some Intel roadmaps. Not saying I agree with it, it's just . . . that's what's supposed to happen. If Intel can't launch Alder Lake next year, then I really have no idea what they're going to do. Tiger Lake Refresh?

I think it's safe to say they won't be on time given recent historic setbacks spanning the last 12 quarters, and historically behind behind and getting caught with their pants around their ankles. If Tiger Lake is as good as the rumors suggest it'll be (it isn't close) then they can do a refresh. AMD is supposed to launch by year's end. And if that interim processor refresh is anything go by, it'll be a mid to late '21 product, and I don't see AM5 launching until late 2Q22. Given the tech stack for the new platform, they'll use up their suggested cadence cycle. Hopefully to make sure it's a smooth launch. Not to be that guy, but the AMD launch cycle for Ryzen has an interesting pattern at the moment. I posted it a while back but removed it because even I thought I was crazy for posting some weird theory.

I'm more interested in Rocketlake if it launches. Core reduction with new core architecture may prove to be interesting. It's become the forgotten child of the Intel lineup. People want to talk about TGL only it seems.

 

[DrMrLordX]

If Tiger Lake is as good as the rumors suggest it'll be (it isn't close) then they can do a refresh.

The problem there is that, at least at this point, Intel can't get (or won't try to get) Tiger Lake to yield at anything higher than 4c. Which is troubling. It's like the "SuperFETs" are helping them with the power/clockspeed problems but not with yields? Maybe by next year they can have 8c Tiger Lake-H finally ready, but if they can do that, why not Alder Lake?

It's all really confusing.

 

[A///]

The problem there is that, at least at this point, Intel can't get (or won't try to get) Tiger Lake to yield at anything higher than 4c. Which is troubling. It's like the "SuperFETs" are helping them with the power/clockspeed problems but not with yields? Maybe by next year they can have 8c Tiger Lake-H finally ready, but if they can do that, why not Alder Lake?

It's all really confusing.

I can't really comment on TGL because I can't seem to find much discussion online that isn't back and forth debate over silly stuff or the mental masturbation over dumb stuff. I'm way more interested in the big/little approach for ADL. Logic tells me it won't be good, but I'm very curious despite that. If that is what Intel will do from now on or for however many years, I'll very likely build a secondary system that uses ADL.

 

[piokos]

Times change. They need 6c and 8c products on 10nm, stat. They can't do it until next year.

No they don't. Just wait for Tiger Lake U reviews. 😉

Magic of "more cores" doesn't work on most consumers. Thankfully.

No it doesn't. It shows that they can't get an 8c or even a 6c Willow Cove to yield correctly. The iGPU is enormous to compensate. Where's the much-touted TigerLake-H? Still not here.

Much touted? You mean: much discussed by enthusiasts on forums like this one where most people stay on desktops anyway?

Intel's most important client for high-end -H chips is Apple.
If Apple really plans to replace Macbook Pro with an ARM system (which Intel surely knows by now), they'll have even less reason to invest in the -H lineup.
I mean, 20 years from now, they'll probably sell just server processors and -U SoCs. So budgets for developing high-power mobile and desktop chips will probably be decimated over the next decade anyway.

If Intel isn't able to make an 8-core mobile competitor for Zen2 out of Tiger Lake, they may just sacrifice this market altogether.
An even if they can, it might not be an interesting niche.
They've already given up on HEDT - a move that was expected already before 2017 - when the main treat from AMD was that it'll collapse and US gov will force Intel to split.

 

[jpiniero]

Intel's most important client for high-end -H chips is Apple.
If Apple really plans to replace Macbook Pro with an ARM system (which Intel surely knows by now), they'll have even less reason to invest in the -H lineup.

With Alder Lake, they are merging U and H into P anyway.