Discussion Intel current and future Lakes & Rapids thread

Page 428 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Asterox

Golden Member
May 15, 2012
1,026
1,775
136
His other twitter post is within the same hour as that Intel AMA
Base PL2 is 107w
There's also a PL2 135w performance



Compared to PL2 for 4/8 Tiger Lake, well red dragon is kind of expected.

It's all just not a mobile CPU, that CPU power consumption on Mobile is grotesque and absurd.

 

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
So far it looks like the 11800H may have been a significant part of that. So there are definitely some 8-core parts.



Boost clocks are significantly higher, base clocks are lower. AVX will drag clocks down like always. wait for the reviews…

Sorry, I should have been more specific in my post. What I meant to say is that at a given frequency 8 core TGL seems to have comparable power draw to Zen 3.

If it's true Intel has already shipped a million of these parts then it seems like they should have skipped Rocket Lake and simply gone Tiger Lake to the desktop at 10SF along with the high performance mobile. Comet Lake could have filled the gap in demand for the desktop and they would have had some 8 core parts that would have actually been good competition for the 5800X. And of course avoided the cost and review disaster that appears to be Rocket Lake. I realize that Rocket was probably initiated years ago and whomever led that course of action made a wrong turn. Easy to see in hindsight of course.

Let me elaborate. In hindsight, it seems like Intel should have continued on with Comet Lake and split Tiger Lake H production between high power mobile and high end desktop, using Comet Lake to fill in any gaps in demand from big partners like Dell. Tiger Lake H would have been better competition for Zen 3, they would have avoided the backport cost, both in time and labor for Rocket Lake, and they would have still kept their 14+++++++ foundries operating with Comet Lake. While Tiger Lake H assumed the new desktop position, the old Comet Lake fabs could be transitioned to 10ESF for Alder Lake.

I have limited knowledge compared to many here I admit and I'm sure there are plenty of technical reasons this couldn't have happened. But if I don't put it out there I won't learn anything about why my thinking is misguided.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
IMO a million is not that much, and that they were able to even do Tiger Lake H45 was probably unexpected.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,573
146
Yikes if true. . .
It's true.

The slides here are part of the press briefing for Chinese media outlets. Tbh I don't think they were supposed to be released out on the internet, but apparently someone didn't care.

As for the discussion about efficiency from earlier:

1620758640303.png

'nuff said.

EDIT: Ooooh, the OG source is a PM at Lenovo. That makes a tonne of sense now.
 
Last edited:

Magic Carpet

Diamond Member
Oct 2, 2011
3,477
231
106
Sucks that you can’t use the unlocked i9-11980HK in a desktop board. Just sad, really. Who needs it in a laptop that sounds like a jet plane? Intel is mad, really. No common sense at all.
 
Last edited:

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
It's true.

The slides here are part of the press briefing for Chinese media outlets. Tbh I don't think they were supposed to be released out on the internet, but apparently someone didn't care.

As for the discussion about efficiency from earlier:

View attachment 44228

'nuff said.

EDIT: Ooooh, the OG source is a PM at Lenovo. That makes a tonne of sense now.

That's just one bench. Not really enough to make a case.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,573
146
That's just one bench. Not really enough to make a case.
It very much gives a good general idea of what to expect for non-AVX-512 productivity tasks. Cezanne to have a very sizable lead in efficiency.

The question left remaining is gaming. At the very least in these benchmarks they're not seeing a huge gain in gaming performance over Comet Lake.
 

dullard

Elite Member
May 21, 2001
24,998
3,325
126
That's just one bench. Not really enough to make a case.
It does show a weakness of Intel's chips. But, you are correct. That is a 7.5 year old benchmark that generally does poorly on all Intel chips. More data is needed for a full conclusion.
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,572
136
It does show a weakness of Intel's chips. But, you are correct. That is a 7.5 year old benchmark that generally does poorly on all Intel chips. More data is needed for a full conclusion.
They might not perform as well there as in some other benchmarks but their power draw has been totally average in cinebench flr a AVX2 load. Not looking great whichwver way you slice it.

Now obviously more data is required before drawing any conclusions, but all-core power is not looking competitive:


16 - 19% multicore uplift (on a marketing slide!) from 2 generations of uarch and 10nm SF?

AMD pretty much doubled MT perf @ same power draw going from 14 to 7nm ...
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
They seem to have a considerable amount of dark silicon, where the GPU used to be. I wonder why couldn't they just add a couple of EUs instead:

Using the comparison to the U die it looks like either they can go with full blown 96 EUs or go with 32EUs. They'd have figured no one needs the 96EU version so going with 32EUs reduces die a bit plus you get empty area which will help with yields.

Someone can check me on this but here's what I came up with.

CPU​
TDPCores/ThreadsBaseMax/All Boost
Intel 11980HK45 W8/162.6 GHz5.0/4.5 GHz
Intel 11980HK65 W8/163.3 GHz5.0/? GHz
AMD 5980HS35 W8/163 GHz4.8/? GHz
AMD 5980HX45+ W8/163.3 GHz4.8/? GHz
Intel 11900H35 W8/162.1 GHz4.9/? GHz
Intel 11900H45 W8/162.5 GHz4.9/4.4 GHz
AMD 5900HS35 W8/163 GHz4.6/? GHz
AMD 5900HX45 W8/163.3 GHz4.6/? GHz
Intel 11400H35 W6/122.2 GHz4.5/? GHz
Intel 11400H45 W6/122.7 GHz4.5/4.1 GHz
AMD 5600HS35 W6/123 GHz4.2/? GHz
AMD 5600H45 W6/123.3 GHz4.2/? GHz

Yea the Zen 3 cores themselves are more efficient. We knew that from Icelake/Tigerlake's massive core size.

Not really comparable since you are talking about AVX-512 versus AVX2 presumably. Also pretty sure AMD doesn't technically have any kind of multiplier limits.

It's a client mobile chip so it doesn't have the second AVX-512 unit, thus power impact should be much smaller.

If it's true Intel has already shipped a million of these parts then it seems like they should have skipped Rocket Lake and simply gone Tiger Lake to the desktop at 10SF along with the high performance mobile.

They probably weren't sure about 10nm production other than maybe they could fill the mobile parts.

I also have a feeling by skipping Tigerlake they can advertise "huge gains" with Alderlake-S.
 
Last edited:

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Based on the steepness of the curve between performance and frequency, there's a possibility 10nm SF is somewhat behind TSMC 7nm used in Zen 3.

They said Zen 3 uses an enhanced version of 7nm. So 10nm SF = 7nm original. It might take 10nm ESF to be equal/exceed 7nm in Zen 3. I think the ESF will end up slightly better than the latter 7nm, but for AMD 5nm will be right around the corner.

The saving grace for Intel is that 5nm is more like a 0.6 node jump so they won't be greatly behind. But they will be.
 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
The saving grace for Intel is that 5nm is more like a 0.6 node jump so they won't be greatly behind. But they will be.

Still a huge jump. In addition the worse architecture (xxxCove) does not pair well with the worse process - so i am not convinced we can conclude about the saving grace here :)
Times were surely more comfortable for Intel when they had the better architecture and the better process.
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
Based on the steepness of the curve between performance and frequency, there's a possibility 10nm SF is somewhat behind TSMC 7nm used in Zen 3.

They said Zen 3 uses an enhanced version of 7nm. So 10nm SF = 7nm original. It might take 10nm ESF to be equal/exceed 7nm in Zen 3. I think the ESF will end up slightly better than the latter 7nm, but for AMD 5nm will be right around the corner.

The saving grace for Intel is that 5nm is more like a 0.6 node jump so they won't be greatly behind. But they will be.

I think in addition to general Intel 10nm vs TSMC 7nm differences, leakage probably comes into play. Historically, Intel's H and S series chips (same die) were based on a higher leakage version of the process. Combine that with an overly large core, and it can partially explain the curve of the graph, if not its offset.
 

cortexa99

Senior member
Jul 2, 2018
318
505
136
I wonder either 10nm process or architecture is broken? Or am I missing something? Alderlake use same 10nm process as Tigerlake-H right?? This perf/watts is ugly and should not be a mobile chip......am I the only one start worrying Alderlake which is on same 10nm process?
OTOH the source article has been deleted, I wonder if something has been released why still has to delete the review?
 
Last edited:

Saylick

Diamond Member
Sep 10, 2012
3,084
6,184
136
Based on the steepness of the curve between performance and frequency, there's a possibility 10nm SF is somewhat behind TSMC 7nm used in Zen 3.

They said Zen 3 uses an enhanced version of 7nm. So 10nm SF = 7nm original. It might take 10nm ESF to be equal/exceed 7nm in Zen 3. I think the ESF will end up slightly better than the latter 7nm, but for AMD 5nm will be right around the corner.

The saving grace for Intel is that 5nm is more like a 0.6 node jump so they won't be greatly behind. But they will be.
For what it's worth, AMD will not be using the base N5 node from TSMC. They will be using some enhanced version of it. If it's N5P, then it will bring an additional 5% perf at iso-power or -10% power at iso-perf.
 
  • Like
Reactions: Tlh97

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
For what it's worth, AMD will not be using the base N5 node from TSMC. They will be using some enhanced version of it. If it's N5P, then it will bring an additional 5% perf at iso-power or -10% power at iso-perf.

Yes, N5P. I think it's reasonably safe to say Intel will be a ~full node behind when Zen 4 comes out. They may have Raptor Lake by then, but an Alder Lake derivative, even with more cores, isn't going to save them. AMD will win until at least Intel's next desktop gen.
 
  • Like
Reactions: Tlh97

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
For what it's worth, AMD will not be using the base N5 node from TSMC. They will be using some enhanced version of it. If it's N5P, then it will bring an additional 5% perf at iso-power or -10% power at iso-perf.

Because it won't be this year. It's Zen 3+ which will face Alderlake. Zen 4 is year after that.

@cortexa99 Tigerlake and Tigerlake-H uses 10nm SF. Alderlake uses Enhanced SF.
 
  • Like
Reactions: Tlh97

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
Because it won't be this year. It's Zen 3+ which will face Alderlake. Zen 4 is year after that. If they get their 7nm process ready it'll be close for once at least on the process side.

Eh, if Zen 4 launches mid-ish '22, it could be a full year until we see a proper desktop competitor from Intel. By which point Zen 5 might not be far away.
 

cortexa99

Senior member
Jul 2, 2018
318
505
136
Find another one:

2.png

1.png

3.png


It seems TGL-H use a high-perf process instead of low-power process in mobile segment which is weird&insane, when 11800H's TDP is around 45w the perf/watt is underwhelming and vastly outperformed by 5800H, only when >80w it might be ahead but the tester's rig is too hot to hold the heat due to heatsink limitation......
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Yikes if true. . .

It does look like 8c Tiger Lake-H can perform if you feed it enough power. Might have been good for a desktop chip, but not really appropriate for mobile/NUCs.

IMO a million is not that much, and that they were able to even do Tiger Lake H45 was probably unexpected.

Tiger Lake-U recently was said to have shipped 4 million units. Maybe as of a month ago? So if Intel already has 1 million TigerLake-H shipped to OEMs and if a significant percentage of that is 8c, then the yields may actually be improving to the point that they aren't completely embarassing.

Sucks that you can’t use the unlocked i9-11980HK in a desktop board. Just sad, really. Who needs it in a laptop that sounds like a jet plane? Intel is mad, really. No common sense at all.

See above, I could see it as a bridge between Rocket Lake-S and Alder Lake-S to tide over customers. Unfortunately Rocket Lake-S was delayed by so much that Intel's product stack would get really confused that way.
 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,573
146
Find another one:

View attachment 44251

View attachment 44250

View attachment 44252


It seems TGL-H use a high-perf process instead of low-power process in mobile segment which is weird&insane, when 11800H's TDP is around 45w the perf/watt is underwhelming and vastly outperformed by 5800H, only when >80w it might be ahead but the tester's rig is too hot to hold the heat due to heatsink limitation......
Well at least this is still better than the earlier results. If nothing else we're not talking about needing >100W but now >80W.

Wonder if that's down to a difference in silicon quality, intra-SKU binning (even within each SKU line, AMD/Intel/Nvidia offer differently bins of products to OEMs) or CB R20 and R23 both being more AVX and AVX2 heavy that allows TGL-H to catch up.