[Ashraf] 10nm "Lakefield" SoC with Intel big + little cores

[coercitiv]

Problem with NBC's browser benchmarks is that they don't say the browser used. For Windows, would it be IE, Edge, Chrome, MS Chrome?

They started mentioning it after the test version, example from the Qualcomm powered Galaxy S review: NBC WiFi Websurfing Battery Test 1.3 (Edge 83.0.478.45)

It may remain inconsistent between different reviewers though, with only some mentioning the browser.

 

[naukkis]

Looks like task manager load info ignores Sunny Cove in this case, we can see from the screenshot there a 4 cores fully loaded and the fith core is loaded with 10%, Notebookcheck says the fith cores runs with only 200-300 Mhz. Or possibly Sunny Cove is running but with 200-300 Mhz only?

Is there power budget to run Sunny Cove at higher frequencies at MT load? I don't think so, whole idea of that package is to get some MT performance from Tremont cores and single thread performance for same power budget from Sunny Cove. Maybe that Cinebench R20 load also just goes over power budget which makes it to switch to Tremont core.

 

[mikk]

Is there power budget to run Sunny Cove at higher frequencies at MT load? I don't think so, whole idea of that package is to get some MT performance from Tremont cores and single thread performance for same power budget from Sunny Cove. Maybe that Cinebench R20 load also just goes over power budget which makes it to switch to Tremont core.

This could be a reason, who knows. The Cinebench R10 run is also a 32 bit run only which probably is easier than the 64 bit version, they should check the 64 bit version. Browser benchmarks are easier as well usually. However if this is the reason it reveals the weakness of Sunny Cove/10nm, it's too power hungry.

 

[coercitiv]

Is there power budget to run Sunny Cove at higher frequencies at MT load? I don't think so, whole idea of that package is to get some MT performance from Tremont cores and single thread performance for same power budget from Sunny Cove.

Of course there's enough power budget, you drop clocks to fit the other core as well. Not having the SNC core working in MT loads is a big loss in efficiency, either in perf/power or alternatively perf/area.

Renoir supports cTDP of 10W for 8X Zen 2 cores and you honestly think Lakefield can't handle 7W with 4X Tremont + 1X SNC? The 8500Y can do cTDP of 3.5W, meaning you could have Skylake 4c/8t still working within 7W on 14nm.

 

[IntelUser2000]

8500Y is a 2C/4T part. You can see from the ST performance that it does not clock up all the way to its rated Turbo even in ST with workloads such as Cinebench, and that's at 7W.

Yes it could still be better. Even if it ran at lower clocks, having Sunny Cove enabled would make it perform much better in MT. And if it can somehow find a way to have it prioritize SNC in ST that would be awesome.

 

[coercitiv]

8500Y is a 2C/4T part. You can see from the ST performance that it does not clock up all the way to its rated Turbo even in ST with workloads such as Cinebench, and that's at 7W.

That doesn't change the fact that doubling the core count of 8500Y would still make it more efficient in terms of power usage. If there were 2 more cores in 8500Y, you would gain performance in Cinebench by using them. It wouldn't be efficient in terms of area, but thats another story.

 

[IntelUser2000]

Hopefully its a scheduler issue that can be fixed. If it really is going to come with the 10X version, then there's no reason to have the product this early.

 

[jpiniero]

Hopefully its a scheduler issue that can be fixed. If it really is going to come with the 10X version, then there's no reason to have it this early.

I am under the impression that you won't see 10X until at least the end of the year.

 

[IntelUser2000]

I am under the impression that you won't see 10X until at least the end of the year.

I know. But you already have the Surface Neo and Lenovo dual screen device coming with 10X in 2021 and X1 Fold is October. Better than a terrible first impression.

Cinebench ST is 10% better than Gemini Lake and no better in MT. It's a big red flag if its just running on Tremont. Yea, it might count PoP DDR as 2W and maybe its really 5W, but still. There's zero advance here.

I wonder how much has to do with it using Icelake's 10nm variant and being crap?

 

[naukkis]

Of course there's enough power budget, you drop clocks to fit the other core as well. Not having the SNC core working in MT loads is a big loss in efficiency, either in perf/power or alternatively perf/area.

Renoir supports cTDP of 10W for 8X Zen 2 cores and you honestly think Lakefield can't handle 7W with 4X Tremont + 1X SNC? The 8500Y can do cTDP of 3.5W, meaning you could have Skylake 4c/8t still working within 7W on 14nm.

It works as Intel has stated it should:

https://www.extremetech.com/wp-content/uploads/2020/06/Tremont-Performance-Curves-1.jpg

SunnyCove is meant for burst type ST-loads only. If they give power budget to SunnyCove in MT-loads it will result lower MT performance as seen in published graphs.

SunnyCove performance/watt is extremely poor, don't compare it to Zen2 as SunnyCove loses in efficiency to even Zen1 Picasso. Skylake has much better performance/watt at low clock speeds than SunnyCove.

 

[mikk]

In Cinebench R15 MT x64 Tremont runs only at 1.9 Ghz says Notebookcheck, so there is no chance Sunny Cove can actively render this scene with a proper cock frequency. The big issue for me is the 1T score and may be Sunny Cove there is too power hungry as well. Sunny Cove on 10nm is really a big power hog.

 

[randomhero]

I have to ask this.
When was the last time that Intel really delivered with any product???
While technology is potent, product is awful.
When will Intel deliver with this or any other tech they shout about???
They just talk, talk, talk.

 

[DrMrLordX]

It has been Intel's corporate directive not to "really deliver" with any product. Overall, that makes it easier for them to sell hardware, so long as their hardware remains desirable compared to the competition.

 

[randomhero]

It has been Intel's corporate directive not to "really deliver" with any product. Overall, that makes it easier for them to sell hardware, so long as their hardware remains desirable compared to the competition.

This will not sell. That's the thing.

Remember TSX in Haswell? Never worked then, still doesn't. But Haswell was really good cpu/soc.This is not.

What about Optane?
Doesn't work. Regular NAND is eating its breakfast.

Graphics?(Edit:I mean latest effort, we all know how larabee ended.)
Only silicon deities know.

And list goes on. What desirable products will be left to sell?

 

[RetroZombie]

So it's the flop i predicted in my previous posts.
The only interesting thing is the packaging, but if it's not going inside a smartwatch with all that tremendous power efficiency...

 

[DrMrLordX]

But Haswell was really good cpu/soc.This is not.

It looks that way NOW, but at the time Haswell was underwhelming for anything that wasn't using AVX2. Your typical 5 GHz 2600x was "fast enough" compared to a 4.5 GHz 4770k. It was a sidegrade in a lot of situations. Intel has been dribbling out minor increases in performance for awhile. The difference now is that they can't accelerate the pace of innovation when they clearly need to do so.

As to when Intel will finally start delivering . . . there's a lot of lag in their development process right now, and it's still not certain that some of these near-future 10nm and 7nm products will deliver. Alder Lake is probably their next, best hope within its respective market. As for Lakefield, I suspect it's tech demo status makes it a sacrificial lamb. If/when Intel releases a successor to Lakefield, it could be much-improved.

 

[mikk]

This will not sell. That's the thing.

Remember TSX in Haswell? Never worked then, still doesn't. But Haswell was really good cpu/soc.This is not.

What about Optane?
Doesn't work. Regular NAND is eating its breakfast.

Graphics?(Edit:I mean latest effort, we all know how larabee ended.)
Only silicon deities know.

And list goes on. What desirable products will be left to sell?

Tigerlake and Xe iGPU in combination with improved 10+ looks really good from what we have seen so far. The only downside might be the multithread speed against 8C Renoir U, even though in the ultrabook market this is not a big factor. Overall Tigerlake looks promising, this is a real improvement over Icelake in everything.

 

[Roland00Address]

I have to ask this.
When was the last time that Intel really delivered with any product???
While technology is potent, product is awful.
When will Intel deliver with this or any other tech they shout about???
They just talk, talk, talk.

22nm was glorious, so that is 2011. 2011 was the last time Intel truly delivered and met / exceeded expectations.

It looks that way NOW, but at the time Haswell was underwhelming for anything that wasn't using AVX2. Your typical 5 GHz 2600x was "fast enough" compared to a 4.5 GHz 4770k. It was a sidegrade in a lot of situations. Intel has been dribbling out minor increases in performance for awhile. The difference now is that they can't accelerate the pace of innovation when they clearly need to do so.

You are comparing an overclocked chip vs an overclocked chip. Not everyone cares for top performance, even if it makes the cpu use a whole lot of power, like 200w of power. That said you are correct sandybridge overclocked vs haswell overclocked was similar enough. You were just not the target customer.

And while desktop non oc sandybrdige (2nd gen) vs non oc haswell (4th gen) saw a decent improvement with their desktop chips, the real improvement of haswell was in laptop and convertible form factors giving both raw performance but also good battery life and thinner form factors.

-----

In theory the Skylake (6th) gen did not give much more increases in performance (both with desktop and laptops) but it was still enticing for it delivered a massive improvement in battery life, on top of what haswell (4th gen) gave. That said skylake had a major problem. The problem with Haswell though was the buggy chipsets and firmware and no on wanted that. They eventually fixed that stuff roughly a year later but by 7th gen these increases no longer looked so good. And with 7th, 8th, 9th, generation we are still talking slight improvements of Skylake on the same process, so we been stuck in the same REAL generation of hardware from 2015 to 2019.

 

[IntelUser2000]

22nm was glorious, so that is 2011. 2011 was the last time Intel truly delivered and met / exceeded expectations.

22nm was 2012, and that was with Ivy Bridge. Other than the graphics which were playing catchup with Llano, it was pretty underwhelming. I have an Ivy Bridge ultrabook too. I should have really waited 1 additional year for Haswell. Battery life didn't advance, and whatever performance there is easily dwarfed by misbehaving power management. 22nm was all about Atom, which didn't pan out.

And while desktop non oc sandybrdige (2nd gen) vs non oc haswell (4th gen) saw a decent improvement with their desktop chips, the real improvement of haswell was in laptop and convertible form factors giving both raw performance but also good battery life and thinner form factors.

Agreed.

In theory the Skylake (6th) gen did not give much more increases in performance (both with desktop and laptops) but it was still enticing for it delivered a massive improvement in battery life

Actually Skylake regressed in some scenarios compared to Broadwell in the battery life department. It wasn't until Kabylake they were able to exceed Broadwell.

Broadwell was what got the battery gain over Haswell.

Tigerlake and Xe iGPU in combination with improved 10+ looks really good from what we have seen so far.

I think it looks better because even Icelake and 10nm used for that chip was far from optimal. Based on Lakefield, Icelake's 10nm looks pretty bad. We got Tremont cores no faster than Goldmont Plus.

 

[IntelUser2000]

Let's say in a hypothetical scenario that updates to the Windows 10 scheduler results in 40% better performance in 1T Cinebench, and MT Cinebench gets SNC to work which gets it to 400 points from current 255. And battery life goes from 11-12 to 13 hours.

Would that change opinions? I bet it wouldn't. While the -Y parts go to 9W, inevitably most will compare against Tigerlake.

The biggest problem with Lakefield is still that its taking so long to come out. Let's say instead we have Lakefield-R in November with Xe graphics, and 30% higher clocks with 10nm+(TGL process). It'd be much better received.

 

[coercitiv]

Let's say in a hypothetical scenario that updates to the Windows 10 scheduler results in 40% better performance in 1T Cinebench, and MT Cinebench gets SNC to work which gets it to 400 points from current 255. And battery life goes from 11-12 to 13 hours.

Would that change opinions? I bet it wouldn't.

400 points would be spectacular but improbable to say the least. Ice Lake Y at 10-12W does ~300 points in a 2c/4t chip and ~400 points in 4c/8t chip.

If Lakefield were doing ~400 points with half the power and half the silicon area... that would be something I would openly applaud. And it should crush Ice Lake in battery life.

 

[insertcarehere]

It looks that way NOW, but at the time Haswell was underwhelming for anything that wasn't using AVX2. Your typical 5 GHz 2600x was "fast enough" compared to a 4.5 GHz 4770k. It was a sidegrade in a lot of situations. Intel has been dribbling out minor increases in performance for awhile. The difference now is that they can't accelerate the pace of innovation when they clearly need to do so.

As to when Intel will finally start delivering . . . there's a lot of lag in their development process right now, and it's still not certain that some of these near-future 10nm and 7nm products will deliver. Alder Lake is probably their next, best hope within its respective market. As for Lakefield, I suspect it's tech demo status makes it a sacrificial lamb. If/when Intel releases a successor to Lakefield, it could be much-improved.

For Intel's sake they definitely need to do much better than Lakefield to compete in this space, if 8cx with two generation old A76 competes so well already imagine how this would fair against an ARM X1/A78 solution that Qualcomm are inevitably cooking up right now...

 

[IntelUser2000]

400 points would be spectacular but improbable to say the least. Ice Lake Y at 10-12W does ~300 points in a 2c/4t chip and ~400 points in 4c/8t chip.

Spectacular is really pushing it. This is the competition:

Asus ZenBook 14 UM433IQ review (AMD Ryzen 7 4700U, Nvidia MX350)

This here is the ZenBook 14 UM433IQ from Asus.

www.ultrabookreview.com

At 10W, the 4700U gets 650 points. The point is that Icelake actually sucks, but it has been hidden due to lack of comparison. That's a big part of why Tigerlake is supposed to have such a huge gain.

Since the gap between LKF and TGL is in the range of 2-3 months(and that's with a single LKF product), it'll go against Tigerlake, which is a far superior product.

Not to mention the hybrid nature of threading should show big advantages in certain categories because it'll complicate coding for it, like how special function units are fast.

I had hoped better for battery life, but preliminary indications are disappointing. The Galaxy Book S should be the pinnacle of optimization, considering how well the Snapdragon version does, and how poorly Surface X does.

Icelake is unfortunately also a disappointment here.

How will it do against the 8500Y?

7 hours with a 23.4WHr

VAIO A12 (Core i7-8500Y, 16 GB RAM, 512 GB SSD, FHD) Convertible Review

Notebookcheck.com reviews the VAIO A12, a 12.5-inch convertible laptop that can also be used as a tablet with a stylus. We review the most expensive version of the device, which VAIO equips with an Intel Core i7-8500Y processor, 16 GB RAM and a 512 GB SSD. Read on to find out how the VAIO...

www.notebookcheck.net

8.6 hours with a 31.9WHr

Acer Swift 7 SF714-52T Review: Slim and light subnotebook for a fat wallet

We test the Acer Swift 3 SF714-52T equipped with a Core i7-8500Y CPU, Solid State Disk, and Full-HD touchscreen display. How does the 14-inch notebook fare against the competitors?

www.notebookcheck.net

7 hours with 45.6WHr

Porsche Design Ultra One laptop short review: Noisy SSD antagonizes the expensive laptop

We review the Porsche Design Ultra One with the Intel Core i7-8500Y, Intel UHD Graphics 615, 16 GB of RAM and a 1 TB SSD.

www.notebookcheck.net

14 hours with a 54.28WHr

HP Spectre Folio 13 (i5-8500Y) Convertible Review

The leather-clad HP Spectre Folio is an aesthetic marvel that combines a timeless material with the latest in ultra-thin performance and wireless LTE technology. The good news is it works, but it's not without a list of notable disadvantages.

www.notebookcheck.net

The Icelake-Y Macbook Air regresses compared to Amberlake(latter doing 17% better)

Apple MacBook Air 2020 Core i5 Review – The best MacBook you can get? – No

Notebookcheck reviews the new Apple MacBook Air with the quad-core i5 CPU. The keyboard has been improved, but the performance is still the Achilles heel of the laptop.

www.notebookcheck.net

 

[Asterox]

For Intel's sake they definitely need to do much better than Lakefield to compete in this space, if 8cx with two generation old A76 competes so well already imagine how this would fair against an ARM X1/A78 solution that Qualcomm are inevitably cooking up right now...

Samsung is also cooking, and adds spicy AMD ingredients.

AMD RDNA-Infused Samsung SoC Benchmarks Obliterate Galaxy S20 Graphics

Keeping an eye on Samsung's developments with AMD RDNA

www.tomshardware.com

 

[mikk]

Spectacular is really pushing it. This is the competition:

Asus ZenBook 14 UM433IQ review (AMD Ryzen 7 4700U, Nvidia MX350)

This here is the ZenBook 14 UM433IQ from Asus.

www.ultrabookreview.com

At 10W, the 4700U gets 650 points. The point is that Icelake actually sucks, but it has been hidden due to lack of comparison. That's a big part of why Tigerlake is supposed to have such a huge gain.

This is CPU TDP versus Soc package TDP isn't it? Aside from this, an 8 core won't lose in Cinebench MT against 4 cores unless it has a really big IPC and clock deficit which is not the case.