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

Page 13 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136
Ok now we know why this hasnt been done before.
Something just looks better in theory and on paper, and takes years to mature, if at all.
Still likes the idea though...lol
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Maybe Lakefield will be better at balancing multiple processes than trying to commit all cores to one multithreaded process.
 
  • Like
Reactions: mastercoin

piokos

Senior member
Nov 2, 2018
554
206
86

First benchmarks. It does manage to beat the 8cx in the Galaxy Book S in the browser tests but only barely in most of them.

Frequency is pretty low on the R15 run, only getting 2.4 Ghz on the single core test and 1.9 on the multi core test. That gives a score of 255 on the MT test which is pretty low and is about what an n5000 gives anyway.
These benchmarks apparently haven't used the big core even in single-thread tests. It's essentially a 4-core Tremont and the results are pretty good seen as such.

And we shouldn't be surprised because heterogenous core global scheduling will be added in Windows 10X.
This preproduction Samsung Book S probably uses normal Windows.

Author should have included W10 version (essential) and power consumption per core though the whole test (nice bonus), which would make this obvious. Pretty lame article.
Notebookcheck remains the best source for laptop reviews, because the test team is really solid.
And the simple fact is: the guy writing articles (Osthoff) just isn't technical enough to properly cover components.
 
  • Like
Reactions: mastercoin

piokos

Senior member
Nov 2, 2018
554
206
86
Geekbench 5

Lakefield (Sunny Cove x1 + Atom tremont x4)

SC 843 | MC 1631

This makes a lot more sense - as expected - matching a 1035G4 in slim devices (Surface Pro etc).
Also MC/SC~=2, which was a design choice Intel mentioned a lot.

So someone knows how to benchmark Lakefield properly (or got a better Windows version ;))
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
These benchmarks apparently haven't used the big core even in single-thread tests. It's essentially a 4-core Tremont and the results are pretty good seen as such.

Appears to be bouncing between the big and small clusters on the R15 single core test. Seems to be working more properly on the browser tests.
 

piokos

Senior member
Nov 2, 2018
554
206
86
Appears to be bouncing between the big and small clusters on the R15 single core test. Seems to be working more properly on the browser tests.
Well, putting aside the lacking article and the poorly functioning benchmark, Tremont looks really good.
These 4 cores are expected to pull 2-3W under load. They match 6-10W Pentium/Celerons (admittedly, pretty old ones).

I really wanted to buy a NUC with N5000 or J4105. I just lost all interest.
 
  • Like
Reactions: mastercoin

mikk

Diamond Member
May 15, 2012
4,112
2,105
136
@Antey

J5005 makes a more-interesting point of comparison.


It's a desktop chip and can use more power.

Cinebench clearly didn't use the big core in the singlethread test. The question is why is this, could it be a software issue which possibly could be fixed.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
It's a desktop chip and can use more power.

True! However, there are some GB5 results for it that run at nearly-fixed clocks close to its turbo limits. It's very easy to get data from those runs for some kind of half-arsed IPC comparison.
 

mikk

Diamond Member
May 15, 2012
4,112
2,105
136
Cinebench clearly didn't use the big core in the singlethread test. The question is why is this, could it be a software issue which possibly could be fixed.


From a Notebookcheck forum writer:
These results are meaningless without the Windows version.
Scheduling for Lakefield will be added in W10X.

If this uses existing W10, you're essentially testing a 4-core Atom.


It sounds plausible and it's a huge mistake from Notebookcheck to ignore the windows version, also the driver infos are missing.
 
  • Like
Reactions: mastercoin

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
It sounds plausible and it's a huge mistake from Notebookcheck to ignore the windows version, also the driver infos are missing.

Obviously it's not going to be 10X. It's not available yet.

The Sunny Cove core is definitely being used, it's just buggy and not the way you would expect. You can see why MS decided to wait.
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,684
136
From a Notebookcheck forum writer:
Check the name of that writer.

The Samsung Galaxy Book S is an ultra-thin laptop that will be sold with Win 10 Home/Pro.
galaxy-s.png

We still don't have confirmation that the Galaxy S that Notebookcheck got their hands on is a final retail unit with mature scheduler, but a forum poster repeating his idea of truth as fact is not the best way to get to the bottom of this.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
We still don't have confirmation that the Galaxy S that Notebookcheck got their hands on is a final retail unit with mature scheduler, but a forum poster repeating his idea of truth as fact is not the best way to get to the bottom of this.

One of the articles about the Lenovo X1 Fold said 10X will have optimizations for it. I assumed battery life but surely scheduler is another?

I was worried scheduling will be difficult. If you look at NBC's Task Scheduler screenshot, there's one core that's not working, and I bet that's Sunny Cove.

These 4 cores are expected to pull 2-3W under load. They match 6-10W Pentium/Celerons (admittedly, pretty old ones).

I really wanted to buy a NUC with N5000 or J4105. I just lost all interest.

I don't know how you came to that conclusion. The MT performance, therefore Tremont performance is no better than Goldmont Plus. They are pretty much at the same TDP level.
 
Last edited:
  • Like
Reactions: spursindonesia

bullzz

Senior member
Jul 12, 2013
404
23
81
I didnt expect intel low power graphics to be this close to qcomm considering they have been doing this longer than intel. it will be interesting to see what tigerlake does
 
  • Like
Reactions: mastercoin

Thala

Golden Member
Nov 12, 2014
1,355
653
136
And we shouldn't be surprised because heterogenous core global scheduling will be added in Windows 10X.
This preproduction Samsung Book S probably uses normal Windows.

Does not need to be added, because it works nicely on the Surface Pro X. At least it would be suprising if x86 Windows does not have all the scheduler updates from Windows ARM.
 
Last edited:

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,684
136
If you look at NBC's Task Scheduler screenshot, there's one core that's not working, and I bet that's Sunny Cove.
Yeah, I posted the CB15 MT screenshot on the previous page, aligns perfectly with your previous theory based on benchmark scores.

The ST benchmark screenshot looks equally weird though, the Tremont cluster is under some type of continuous load according to the Task Manager.
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,684
136
Notebookcheck also published a first impression article on the new Galaxy S, and the supposed strong point of Lakefield is not really there:
The Intel model uses a bit more power in almost every scenario, which is also noticeable in the battery runtime. Samsung advertises a lower runtime, and our results so far confirm that: The load test runs for almost 4 hours (without outdoor mode), 2 hours less than the ARM model. The Wi-Fi test at 150 nits is still running, but it should result in about 11-12 hours compared to more than 16 hours for the ARM model.

They are actually reviewing it, which means they believe it's a final retail unit or equivalent.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
The ST benchmark screenshot looks equally weird though, the Tremont cluster is under some type of continuous load according to the Task Manager.

This is how it looks on my desktop i3 7100 under Cinebench R15 ST.
1593253017686.png

I tried assigning only 1 thread to it. The power usage doesn't change much. Sure it looks different in Task Manager, but my scores were nearly 10% better just letting Windows do its own thing. Even HWInfo is showing it as the load being spread over 4 cores. Does perform like a 1 core so I have to assume its just the graphics that's off. I know Windows lets threads hop between cores.

Of course that won't apply to Lakefield with a whole new paradigm.

In order to know the whole story, someone has to sit down and just analyze Cinebench. I would have HWInfo up and track power and clock numbers as I'm changing thread assignment through Task Manager. No site is going to do that though.

and the supposed strong point of Lakefield is not really there:

Yup. It's still better than the U series but seems about the same as Y. Intel talks in their presentation that enabling Sunny Cove allows 33% better performance and 17% better perf/watt under WebXPRT 3.
 
Last edited:

mikk

Diamond Member
May 15, 2012
4,112
2,105
136
Notebookcheck:
Interessanterweise sehen wir bei Single-Core-Belastung maximal 2,4 GHz und bei der Verwendung von allen Kernen laufen 4 Kerne mit ~1,9 GHz und 100 % Last, während ein Kern mit knapp 10 % Last und 200-300 MHz arbeitet.

Notebookcheck says the chip only clocks up to 2.4 Ghz in Singlecore workloads and ~1.9 Ghz for 100% multithreads workloads which uses 4 cores, apparently the Tremont cores. The fifth core is idling. The sreenshot is from Cinebench, so it might look different in other applications. But it's obvious that Sunny Cove isn't working in Cinebench R15/R20 Singlethread.

Did someone notice the Cinebench R10 32 bit scores? Singlethread looks much better there, this must be Sunny Cove.

Cinebench R20 1T
i5-1030NG7 330
i5-L16G7 182
Celeron N4020 175

Cinebench R10 32 Bit 1T
i5-1030NG7 4250
i5-L16G7 4079
Celeron N4020 2184
 
  • Like
Reactions: lightmanek

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
I compared using all available 8500Y and N5000 data. You just type the name of the CPU click it.

Observations-

Under browsing benchmarks its competitive and and Cinebench R10 ST(32-bit), its drastically faster than Geminilake Refresh N5030. 63% advantage tells me its running on Sunny Cove.

@mikk
8500Y outperforms the 1030NG7 only in Cinebench R10 32-bit ST. But it performs worse in newer benchmarks.

Why is Task Manager saying its at 100% load under multi-threaded loads? Is it actually running Sunny Cove or is it a misreport?
 
Last edited:

mikk

Diamond Member
May 15, 2012
4,112
2,105
136
Why is Task Manager saying its at 100% load under multi-threaded loads? Is it actually running Sunny Cove or is it a misreport?

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?
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Or possibly Sunny Cove is running but with 200-300 Mhz only?

Reviewers should take advantage of HWInfo more.

So I downclocked my CPU by approximately half, which is 1.9GHz and when I run MT Cinebench Task Manager says its at 48% utilization. HWInfo reports it correctly and says its 100%.
 
  • Like
Reactions: Tlh97 and coercitiv

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
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?