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

Page 16 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Looks like the Tremont cores run at a constant frequency.

1593777752920.png

Another result got 540 points for Cinbench R20. They should have just waited for the final result. I assume the reason they got a "preview" article ahead of the real one is so they get more hits. Not like we aren't filled with clickbait articles all over the web!

The 1T result for both Cinebench versions and MT result for the R20 is what's really low. R15 MT isn't going to improve since its already beating the 8500Y. The issue is that N5000 beats it too.
 
Last edited:

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
Looks like the old Cinebench R20 version was the culprit. The are saying this device runs with decreased PL1 of 5W instead 7W, PL2 is unchanged 9.5W.
So two myths killed with one stone:
  • It doesn't take Win 10X to get hybrid scheduling up and running.
  • Sunny Cove can run together with the Tremont cluster even with very low power budget. (5W TDP)
 

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126

bullzz

Senior member
Jul 12, 2013
404
23
81
This makes me happy for we now see there is a use case for this hardware, it achieved its objectives. The only question is price and does it make sense for you.

The price question has been answered. Samsung replaced an ARM SoC in their previous Galaxy S with Lakefield. It is safe to assume they would not have unless Lakefield was price competitive. From a die size point of view it makes sense - qualcomm 8cx is bigger (112mm2) vs lakefield (82mm2) and Intel does not have pay the TSCM tax.
 
  • Like
Reactions: Zucker2k

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
The price question has been answered. Samsung replaced an ARM SoC in their previous Galaxy S with Lakefield. It is safe to assume they would not have unless Lakefield was price competitive. From a die size point of view it makes sense - qualcomm 8cx is bigger (112mm2) vs lakefield (82mm2) and Intel does not have pay the TSCM tax.

Lakefield has multiple dies, more expensive packaging, and a lower yield process. And that's ignoring wafer cost.
 

Doug S

Platinum Member
Feb 8, 2020
2,202
3,405
136
The price question has been answered. Samsung replaced an ARM SoC in their previous Galaxy S with Lakefield. It is safe to assume they would not have unless Lakefield was price competitive. From a die size point of view it makes sense - qualcomm 8cx is bigger (112mm2) vs lakefield (82mm2) and Intel does not have pay the TSCM tax.

What "TSMC tax"? Intel has much higher margins than TSMC, companies buying from them pay a much higher "Intel tax".

In the past Intel paid companies to use their mobile chips by giving them more in "market development funds" (free advertising and other support) than they paid for the chips, and despite this costing them billions in chips with an effectively negative price they STILL failed in the market.

I wouldn't be shocked if Intel isn't trying that strategy again hoping to break into the mobile market, and if so of course it would make sense for Samsung to use them so long as Intel wants to give them free chips.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136

Performance improvements but the battery life is merely ok.

I wonder if it underperforms in Cinebench R15 because its not the latest?

Sunny Cove can run together with the Tremont cluster even with very low power budget. (5W TDP)

I don't think so. In R20, which supports the configuration properly, the scores are just above 2x, which is well within the range of the graphs they presented. For comparison the top N5000 gets 500. If Sunny Cove does kick in, its affecting it very little.

What "TSMC tax"? Intel has much higher margins than TSMC, companies buying from them pay a much higher "Intel tax".

Intel's product margins are high, but they spend a lot.

Net margins comparison(Q1 2020)-
Intel - 5.7 billion net revenue on 19.8 billion revenue = ~29%
TSMC - 3.9 billion net revenue on 10.31 billion revenue = ~38%
 
Last edited:
  • Like
Reactions: Tlh97

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
It seems to do ok on Cinebench R11.5.

1.4/3.2, ST/MT

Comparison-
N5000: 0.9/2.8
8500Y: 1.5/2.5
1000NG4: 1.5/2.7
1030NG7: 1.2/3.4

The ARM version has significantly better battery life but the Intel version runs noticeably cooler. I wonder if the battery life will vary significantly based on browser versions?

They are saying by manually setting the affinity to the Sunny Cove core, the R15 ST performance can be improved from 88 points to 111 points(26%).

No performance reduction under battery mode. Doesn't look like it throttles over time either.
 
Last edited:
  • Like
Reactions: Tlh97

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
From a die size point of view it makes sense - qualcomm 8cx is bigger (112mm2) vs lakefield (82mm2) and Intel does not have pay the TSCM tax.

In some countries it'll cost $30 more for the Intel version.

Lakefield has multiple dies, more expensive packaging, and a lower yield process. And that's ignoring wafer cost.

Yup. An 80mm2 die is $5-7, but the package adds $5. You have two dies, one that's on a cheaper process but larger so it'll add another $5. Then the package is a new one so it'll be low volume which will raise the cost of production.

Update: Translated from the NBC review:
...We introduced the Samsung Galaxy Book S with the Intel® Core ™ i5-L16G7 processor with a processor power of five watts. We are happy to announce today that we can probably increase this processor performance again in the next two weeks by an update. This enables a processor power of seven watts.
 
Last edited:

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
It seems to do ok on Cinebench R11.5.

1.4/3.2, ST/MT

Comparison-
N5000: 0.9/2.8
8500Y: 1.5/2.5
1000NG4: 1.5/2.7
1030NG7: 1.2/3.4

Just a reminder for everyone at home. The 2013 Baytrail chip, z3770 usually got 0.4 Single Thread, and 1.48 Multithread. Also released that year was the slower z3740 which you saw in more devices for the z3770 was the premium chip.

So in 7 years we are talking roughly a 2.14x in Multi Thread and 3.55x in Single Thread. Of course the price for these devices also has gone up by that amount as well 😁 😒
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
I don't think so. In R20, which supports the configuration properly, the scores are just above 2x, which is well within the range of the graphs they presented. For comparison the top N5000 gets 500. If Sunny Cove does kick in, its affecting it very little.
They show a task manager graph with full CPU utilization, and as you repeatedly said in the thread, keep in mind about the low(er) 5W TDP and RAM power being factored in. That relative power/perfomance graph looks like the holy grail of napkin math until you realize nobody but Intel knows the target TDP for that data.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
They show a task manager graph with full CPU utilization, and as you repeatedly said in the thread, keep in mind about the low(er) 5W TDP and RAM power being factored in.

You know what? On second look, it does seem its used, but I wouldn't agree with it at 100%. Task Manager does some funny things. Show us HWInfo data or something. Sunny Cove is just too power hungry to be at 100% under MT workloads while at the same time quad Tremont cores are active.

If they up the TDP to 7W, its actually a significant power headroom increase. Think about it-

0.5W TDP for memory = 4.5W to 6.5W, or a 44% increase in headroom
1W TDP for memory = 4W to 6W or 50% increase

I don't think we'll see huge gains. But I'd put peak ST gains at 15%. For MT it depends on whether Tremont can mostly stretch its legs already. Could be as low as 10% to high as 25%.
 
  • Like
Reactions: Tlh97

amosliu137

Junior Member
Jul 12, 2020
22
38
61
If we use the usual calculation of mobile's soc, the notebookcheck's review shows the power comsumption of Lakefield is 14.9-4.8=10.1w when stressing CPU and GPU in 5w tdp. 8cx is 13.2-4.3=8.9w when stressing cpu only .
 

Attachments

  • Screen Shot 2020-07-12 at 9.31.02 PM.jpg
    Screen Shot 2020-07-12 at 9.31.02 PM.jpg
    126.2 KB · Views: 26

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
If we use the usual calculation of mobile's soc, the notebookcheck's review shows the power comsumption of Lakefield is 14.9-4.8=10.1w when stressing CPU and GPU in 5w tdp. 8cx is 13.2-4.3=8.9w when stressing cpu only .
While that is disappointing looking at the idle power consumption. 0.9w difference at idle can mean several more hours of battery life. The load and full load numbers are okay, but I wished the idle was better for we were promised that part of this intel product (compared to other intel products) was going to be so much better.

A reminder that we are talking a 40 watt hour battery (actually 39.8 watt hour) a difference of idle power consumption of 0.9 watts can be the difference of several hours of battery.
 
Last edited:

Thala

Golden Member
Nov 12, 2014
1,355
653
136
Looks like Lakefield is somewhat between a rock and a hard place. Higher power and lower performance then the ARM competition but more compatible - however compared to other x64 machines not really much better efficiency.

Notebookcheck summarizes: "Das Galaxy Book S mit dem ARM Prozessor hatte einige Vorteile gegenüber "normalen" Windows Notebooks. Mit dem Intel Prozessor verbessert sich zwar die Kompatibilität, doch im Gegenzug muss man auf das LTE-Modul verzichten, und auch die Akkulaufzeiten sind deutlich kürzer. Insgesamt bieten die Konkurrenten für diesen Preis einfach deutlich bessere Gesamtpakete. "

Translation:"The Galaxy Book S with ARM processor has some advantages compared to "normal" Windows Notebooks. With Intel processor you get better compatibility, but you need to pass on the LTE module and the battery runtime decreases dramatically. Overall the competition offers for the asking price a clearly better overall package."
 
  • Like
Reactions: Tlh97

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
While that is disappointing looking at the idle power consumption. 0.9w difference at idle can mean several more hours of battery life. The load and full load numbers are okay, but I wished the idle was better for we were promised that part of this intel product (compared to other intel products) was going to be so much better.

10-20% difference in AC power consumption doesn't explain the 60% difference in battery life. Many times I've seen results where the AC results are very different from battery runtimes.

Also, the XPS 2-in-1 can get 35 hours in idle battery life(several times better than predecessor while regressing in actual tests.

I think a further improvement in scheduler/OS along with testing by others will show better results.

If we use the usual calculation of mobile's soc, the notebookcheck's review shows the power comsumption of Lakefield is 14.9-4.8=10.1w when stressing CPU and GPU in 5w tdp. 8cx is 13.2-4.3=8.9w when stressing cpu only.

Power tests aren't all CPU/GPU. Storage comes into play, RAM uses more power, even the motherboard isn't free. There are 20 different things that contribute to power consumption to a device under load.
 
Last edited:

amosliu137

Junior Member
Jul 12, 2020
22
38
61
10-20% difference in AC power consumption doesn't explain the 60% difference in battery life. Many times I've seen results where the AC results are very different from battery runtimes.

Also, the XPS 2-in-1 can get 35 hours in idle battery life(several times better than predecessor while regressing in actual tests.

I think a further improvement in scheduler/OS along with testing by others will show better results.



Power tests aren't all CPU/GPU. Storage comes into play, RAM uses more power, even the motherboard isn't free. There are 20 different things that contribute to power consumption to a device under load.
I just want to compare with mobile soc. Usually, we cannot read the package power in mobile soc just like intel.
 
  • Like
Reactions: Tlh97

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
View attachment 25763
When I disable turbo boost of 1030NG7, it take about 7.5w to get 652 points. L16g7 got 609 when limit 5w.

That supports the argument that its basically running on Tremont core.

If you assume 25% gain for HT, then it takes 4 Sunny Cove cores ~1.4GHz to get 650. Then Tremont is in average only running about 2.1GHz.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Golem.de has a through review, and hardwareluxx has an abbreviated version of Golem's review.

Tremont ST/Sunny Cove ST/Total MT
Cinebench R11.5-2.91,08 ptk1,35 ptk (+25 %)3,09 ptk (+286 % bzw +128 %)
Cinebench R15-0.3.868 ptk107 ptk (+57 %)251 ptk (+369 % bzw +135 %)
Cinebench R20-0.6.0211 ptk272 ptk (+29 %)577 ptk (+273 % bzw +112 %)
Geekbench 5.1.1529 ptk884 ptk (+67 %)1758 ptk (+332 % bzw +99 %)

It's extremely poor in R15, but does much better in R11.5 and R20.

Also, Tremont is an excellent core if it does that with a 200MHz clock disadvantage.

Their conclusion seems positive on the battery aspect. They say their Schenker ultrabook with 73WHr cannot match the Lakefield Galaxy Book S with a 42WHr battery.

Hybrid support requires enablement from OS/Firmware/Hardware. They say Intel/MS has got it going in the first try.
 
  • Like
Reactions: lightmanek

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
Golem.de has a through review, and hardwareluxx has an abbreviated version of Golem's review.

...

Their conclusion seems positive on the battery aspect. They say their Schenker ultrabook with 73WHr cannot match the Lakefield Galaxy Book S with a 42WHr battery.

Hybrid support requires enablement from OS/Firmware/Hardware. They say Intel/MS has got it going in the first try.

Repeating what I said earlier in this thread but with slightly variant words. It appears Lakefield is "good enough" that it hit some of its goals, and other goals it surpassed. This was in doubt previously. Yet Lakefield showed up to play the game and it did a good enough job.

Now it is a question not of the intel SoC but instead the OEMs can they make devices that are interesting enough in some way, and/or release it at prices that make it interesting on the value front.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Now it is a question not of the intel SoC but instead the OEMs can they make devices that are interesting enough in some way, and/or release it at prices that make it interesting on the value front.

I hope Lakefield-R is a significant improvement. Just moving to Xe is not enough. The main issue is the CPU. I expect something like 25% faster in ST and 35% faster in MT along with battery life improvements at minimum. Really though I'm not expecting anything amazing until the third generation(successor of Lakefield-R, Ryefield or whatever its called).

I feel for devices that doesn't have to be razor-thin and feather-light Tigerlake-Y is going to completely demolish it. Significantly better battery life is one thing I counted for on Lakefield, but it doesn't seem to be enough, and Tigerlake with 10nm+ may do a fair bit better than Icelake.

Manufacturers seem to be aware of the deficiency of the product as its coming in troublingly few number of products.

It needs to aim for something even smaller like a 8-10 inch device.