Discussion Intel current and future Lakes & Rapids thread

Page 187 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
  • Like
Reactions: uzzi38

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
And what OS is that system running?

Linux, but the prior score of 1372 was running ICL on Windows at basically the same frequency.

I'm not a huge fan of GB but you can't say the ICL score is too high, you need to look at this list to compare when the same score is in your list and we have Linux results for TGL as well all showing basically the same perf/clock from the latest entries.
 

csbin

Senior member
Feb 4, 2013
838
351
136
No, but we have Ice Lake U RVP to compare to Tiger Lake U RVP



RVP systems are what Intel sends out to OEMs so they have something to test before they design products. I'd say that's a far better comparison than the best possible Geekbench run for that CPU.



So..Single-Core Score 1240

ICL-U RVP frequency

3687, 3589, 3706, 3775, 3678, 3758, 3692, 3159, 3652, 3759, 3732, 3641, 3413, 3509, 3608, 3752, 3675, 3540, 3765, 3814, 3355, 3432, 3460, 3612, 3668, 3627, 3650, 3621, 3659, 3681, 3742, 3681, 3268, 3272, 3480, 3600


Single-Core Score 1400

TGL-U RVP frequency

4165, 4150, 4164, 4181, 4180, 4165, 4143, 4166, 4175, 4179, 4185, 4169, 4183, 4180, 4187, 4179, 4186, 4177, 4180, 4183, 4179, 4147, 4178, 4185, 4182, 4183, 4166, 4178, 4184, 4180, 4182, 4180, 4184, 4184, 4177, 4182, 4182, 4184, 4181, 4184, 4161, 4180
 

FriedMoose

Member
Dec 14, 2019
48
28
51
Linux, but the prior score of 1372 was running ICL on Windows at basically the same frequency.
This is why you don't compare scores between operating systems:



Note the highest 1065G7 score on windows is only 1372. Yet crossing 1400 on Linux is fairly easy. Here's a result from page 3:

 

mikk

Diamond Member
May 15, 2012
4,111
2,105
136
I doubt it's a good idea to use the best ICL-U score because we haven't seen enough TGL-U results to use a comparable best score. Averaged ICL-U is probably a better comparison and because this TGL sample is a DDR4 variant means Icelake-U with LPDDR4 is not a good idea either. 2.3 Ghz base might point to a 28W SKU, however we already have seen TGL-U with 2.7 Ghz base (not on Geekbench yet). Interesting is this:

Intel(R) Gen12 Mobile Graphice Controller
Maximum Frequency 1.50 GHz


1.50 Ghz GPU boost on a Stepping 0. The fastest Icelake SKUs can go up to 1.1 Ghz for its GPU, looks like Intel is looking for a beefy GPU turbo increase with Tigerlake.
 
  • Like
Reactions: Dayman1225

FriedMoose

Member
Dec 14, 2019
48
28
51

So..Single-Core Score 1240

ICL-U RVP frequency

3687, 3589, 3706, 3775, 3678, 3758, 3692, 3159, 3652, 3759, 3732, 3641, 3413, 3509, 3608, 3752, 3675, 3540, 3765, 3814, 3355, 3432, 3460, 3612, 3668, 3627, 3650, 3621, 3659, 3681, 3742, 3681, 3268, 3272, 3480, 3600


Single-Core Score 1400

TGL-U RVP frequency

4165, 4150, 4164, 4181, 4180, 4165, 4143, 4166, 4175, 4179, 4185, 4169, 4183, 4180, 4187, 4179, 4186, 4177, 4180, 4183, 4179, 4147, 4178, 4185, 4182, 4183, 4166, 4178, 4184, 4180, 4182, 4180, 4184, 4184, 4177, 4182, 4182, 4184, 4181, 4184, 4161, 4180

This one has higher clocks:

 

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,570
146
So let's take that score then why don't we.

ICL-U: Approx 3.8GHz reported clocks. Score of 1273. Performance per clock: 335pts/GHz
TGL-U: Aprrox 4.175GHz reported clocks. Score of 1400. Performance per clock... 335pts/GHz. And I even went and underestimated the clock a little too.

They're about the same even still.
 
  • Like
Reactions: spursindonesia

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
This is why you don't compare scores between operating systems:



Note the highest 1065G7 score on windows is only 1372. Yet crossing 1400 on Linux is fairly easy. Here's a result from page 3:

And yet there are plenty of scores where the perf/Hz is basically equal between Windows and Linux so just because it's on Linux doesn't guarantee it higher perf/Hz. So we have multiple ICL entries on Windows performing the same perf/Hz as the best TGL score available, ICL on a reference platform running Linux performing the same in perf/Hz as TGL score available, but we're supposed to just ignore those ICL scores because they're too good?
So let's take that score then why don't we.

ICL-U: Approx 3.8GHz reported clocks. Score of 1273. Performance per clock: 335pts/GHz
TGL-U: Aprrox 4.175GHz reported clocks. Score of 1400. Performance per clock... 335pts/GHz. And I even went and underestimated the clock a little too.

They're about the same even still.

Exactly, it's not just about the score, TGL seems like it will run higher clocks (which is good) but won't improve on IPC much if at all (at least in a way that will be shown by GB). If you use the exact max frequency reported by both runs it comes out to 334.6 pts/Ghz for TGL and 331.3 pts/GHz for ICL or a 1% increase in perf/Hz. Obviously we'll have to wait for actual reviews to confirm and also see if other tests show more improvement in other workloads, but we work with the best we have so far.
 

Richie Rich

Senior member
Jul 28, 2019
470
229
76
Oh, I should clarify that even the Ocean Cove name itself is dead, hence my hope for another Architecture Day as validation. I was told that much, at least.
Ocean Cove as a new uarch is dead? But Why?
Why would Intel stop development of high performance uarch when actually having hard time with AMD's IPC? I can imagine it was stopped due to bad results (performance or perf/watt)
Is this move somehow related to Keller?
 

FriedMoose

Member
Dec 14, 2019
48
28
51
And yet there are plenty of scores where the perf/Hz is basically equal between Windows and Linux so just because it's on Linux doesn't guarantee it higher perf/Hz. So we have multiple ICL entries on Windows performing the same perf/Hz as the best TGL score available, ICL on a reference platform running Linux performing the same in perf/Hz as TGL score available, but we're supposed to just ignore those ICL scores because they're too good?
You don't compare them because several people per day score higher than the Windows world record, and Linux scores are consistently higher across the board. If you used the world record Linux score you might conclude TGL is a 5-10% step back in IPC. But we all know that's a completely ridiculous thing to do.
 

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
Don't worry peeps, Intel is gonna be having another Conroe moment, REAL SOON NOW.

How do I know?

Well, Francois told me so.


That's not even his worst take on the subject, he tweeted right after that:


To show how bad Intel is going to "pulverize" AMD in mobile offerings, he linked to a comparison of an unreleased, most likely top of the line TGL product against a 2500u. Yes, he compared intel's best unreleased product to AMD's mid tier 14 nm (not even 12nm) mobile processor launched over 2 years ago.
 

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
You don't compare them because several people per day score higher than the Windows world record, and Linux scores are consistently higher across the board. If you used the world record Linux score you might conclude TGL is a 5-10% step back in IPC. But we all know that's a completely ridiculous thing to do.

In isolation you would have a point, but the scores aren't in isolation and we have lots of ICL data on both operating systems for comparison. Believe whatever you want, I'm out. I'll see you all when the reviews arrive.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Allegedly, it's 7nm Meteor Lake in Q3/Q4 2022. Or whenever.
Intel better hope that isn't true. Aurora is supposed to be ready in Q4 2021 with other 7nm coming in 2022 . . . that's their lifeline.
...I'm sorry, I'm an idiot. I've mixed up my years, you're correct.

In which case, I'll update to Q1 2022 for ADL-S and Q4 MTL-S I guess?

We just don’t have any visibility into Intel 7nm EUV yet (as Charlie D has noted). Aurora only has a 7nm GPU, IIRC. It's just a big fat pipe cleaner. I would expect Intel to Prioritize Server and laptop CPUs on 7nm, because of the higher margin, but that is just an educated guess - we really don’t know what is going on.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
That there isn't much of an IPC increase from Sunny Cove does fit with what Intel said about the performance gains ("double digit", which is mostly the increase in boost clock from ~3.8 to ~4.2). Except for AES, it seems that it's pretty much even against the 10510U.
 

mikk

Diamond Member
May 15, 2012
4,111
2,105
136
Ocean Cove as a new uarch is dead? But Why?
Why would Intel stop development of high performance uarch when actually having hard time with AMD's IPC? I can imagine it was stopped due to bad results (performance or perf/watt)
Is this move somehow related to Keller?

Intel Oregon is dead according to him, he didn't say high performance uarch in general is dead because Intel IDC is alive. Sunny Cove, Willow Cove, Golden Cove and the next generation comes from IDC.

That there isn't much of an IPC increase from Sunny Cove does fit with what Intel said about the performance gains ("double digit", which is mostly the increase in boost clock from ~3.8 to ~4.2). Except for AES, it seems that it's pretty much even against the 10510U.


Double digit: 10-99%....there is some room for IPC improvements.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Intel Oregon is dead according to him, he didn't say high performance uarch in general is dead because Intel IDC is alive. Sunny Cove, Willow Cove, Golden Cove and the next generation comes from IDC.

A little confused here. CPU design at Hillsboro is gone? Only Israel is doing 'Core' development?
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
Don't worry peeps, Intel is gonna be having another Conroe moment, REAL SOON NOW.

How do I know?

Well, Francois told me so.

It's an interesting idea, and I'm sure it's what Francois wants Intel to do. But i don't see it happening. Intel has already dished on their future 2020/2021 sockets - LGA1200 and LGA1700. A high core-count Willow Cove-based CPU using EMIBed 4c dice (or whatever) would likely require a larger package size than can be accommodated by LGA1200, but LGA1700 isn't until 2021 (Alder Lake-S).

And comparing to a 2500u = lulz

We just don’t have any visibility into Intel 7nm EUV yet (as Charlie D has noted).

True. There's all kinds of potential for problems/delays. My take on it is that Intel needs to hit a home run here to stay in the game. I'm cautiously optimistic that they've got their 7nm process moving along nicely since they learned so much from 10nm (and moving to EUV probably made some things easier for them). Without working 7nm for their entire CPU lineup in 2022, Intel is going to be in bad shape. I really don't want to consider that seriously until evidence begins to indicate that might be the case.
 
Last edited:
  • Like
Reactions: CHADBOGA

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,570
146
A high core-count Willow Cove-based CPU using EMIBed 4c dice (or whatever) would likely require a larger package size than can be accommodated by LGA1200, but LGA1700 isn't until 2021 (Alder Lake-S).

And comparing to a 2500u = lulz
Don't fully rule out Tiger Lake/Willow Cove based options. There is a very, very, very low chance Intel might pull something off. Maybe.
 

scannall

Golden Member
Jan 1, 2012
1,944
1,638
136
Don't fully rule out Tiger Lake/Willow Cove based options. There is a very, very, very low chance Intel might pull something off. Maybe.
They may have something. They may not. One thing they are lacking at the moment is credibility though. When there are actually parts on shelves you can buy. And reviews from trusted sources I will read them with interest. But I can't take marketing slides and leaks the least bit serious.
 
  • Like
Reactions: Markfw

Richie Rich

Senior member
Jul 28, 2019
470
229
76
Intel Oregon is dead according to him, he didn't say high performance uarch in general is dead because Intel IDC is alive. Sunny Cove, Willow Cove, Golden Cove and the next generation comes from IDC.
Thanks, very interesting info. So Oregon stuff was reduced to keep Atom uarch development only? Because if not one wilde idea crossed my mind, that Keller started Intel's high-performance ARM core development instead of Ocean Cove, something like K12 he was not able to finish in AMD. It might be more efficient not to duplicate x86 Core development and instead be ready for hypothetical ARM/Nuvia server offensive. Can we exclude there is no ARM core development in Intel right now?
 

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
Thanks, very interesting info. So Oregon stuff was reduced to keep Atom uarch development only? Because if not one wilde idea crossed my mind, that Keller started Intel's high-performance ARM core development instead of Ocean Cove, something like K12 he was not able to finish in AMD. It might be more efficient not to duplicate x86 Core development and instead be ready for hypothetical ARM/Nuvia server offensive. Can we exclude there is no ARM core development in Intel right now?
Last time I checked Intel was developing their Atom cores in Austin