Intel Cannonlake, Ice Lake, Tiger Lake & Sapphire Rapid Thread

Page 97 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

coercitiv

Diamond Member
Jan 24, 2014
3,198
490
136
I'll repeat, the last time they got 2x was going from i5 4200U to a i5 8265U. The multi-threaded performance of the latter chip is 3x as fast, and the single thread performance is 50-60% better.
And that was an apples to apples comparison: 15W vs. 15W to showcase the combined gains in real world workloads (arch, process node). Would you have considered the the 2X in productivity a real gain if it had been done on WHL 4c/8t @ 28W TDP vs. Haswell 2c/4t @ 15W?! Because that's what 9W TGL vs. 5W AML means in terms of relative power ratio.

Heck, do you even think the 9W TGL can work in the same form factors as 5W AML?
 

TheGiant

Senior member
Jun 12, 2017
365
46
76
IMO the lower the node, the more difficult to cool because of density and hotspots it will be
we need some innovation in cooling too
 

IntelUser2000

Elite Member
Oct 14, 2003
5,994
169
126
This is important, because its in single thread and light thread where the gap is small and hard to get. If it gains 50-60% there, it'll match the top end 45W quad core CFL-H.

Top end gains likely won't change much so we might be entering an era where the lowest power chip to the highest power one has minimal difference in light workloads.

The market moved from 35W to 15W chips because suddenly there wasn't enough of a gap between the two.

Heck, do you even think the 9W TGL can work in the same form factors as 5W AML?
Some already do. The XPS 13 2-in-1 is a fanless device that can put itself up to 9W. 9W is not an official Intel spec but Dell is using it.

I would challenge you to find Y systems with screen size under 11-inch. If you look at it that way, it won't change things. Also, from Notebookcheck reviews, significant amount of Y chips are set at 7W. Of course, significant amount of U chips are set at 25W.

What 3-5W with Lakefield allows it to do(along with better integration) is go into form factors with 10-inch or smaller screens.
 
Last edited:

ApTeM

Junior Member
Jan 12, 2019
12
10
41
Some relatively recent but missed Ice Lake U results running @1.3GHz (?) or maybe that's just a base frequency detected by GeekBench because there's no way Ice Lake U can score 7233 in single threaded mode @ 4GHz. That'd be just an insane result.

Edit: it was running at ~1700MHz. At 5GHz it would score ~6915 vs 6100 for Coffee Lake which makes Ice Lake's IPC 13.4% higher. Not too much I have to admit. At least it's the first IPC increase in the past 4 years.
 
Last edited:

IntelUser2000

Elite Member
Oct 14, 2003
5,994
169
126
Edit: it was running at ~1700MHz. At 5GHz it would score ~6915 vs 6100 for Coffee Lake which makes Ice Lake's IPC 13.4% higher. Not too much I have to admit. At least it's the first IPC increase in the past 4 years.
Where did you get the 1700MHz figure from?

If you search for Ice Lake, it also shows a Kaby Lake client named result with the internal naming being Icelake. The 2.6GHz version gets 4150 in ST with Linux 64-bit.

The 8265U gets 5200 for the top result. So if you assume that's the real result, you end up with 19%. Then again, a result just below that gets 4950. So maybe that's the better comparison because Icelake 2.6 might do better than 4150.

User submitted results are too unreliable to rely on.
 

jpiniero

Diamond Member
Oct 1, 2010
6,327
226
126
I mean, right, Rocket Lake; Comet Lake; and Cooper Lake make things look bad for them in the desktop/workstation/server market since they wouldn't need any of those things if a full raft of IceLake products were ready for every market segment.
Comet and Rocket are also being released for mobile. It's not just for desktop.

Because of the bad yield due to high defects, Intel is going to have to bin very loosely with the clock speeds/voltages to ensure that any chip that isn't defective is used. So then it becomes a problem as to whether the Icelake chips would be actually better than anything 14 nm Intel puts out.

Tigerlake using chiplets would help greatly.
 

jpiniero

Diamond Member
Oct 1, 2010
6,327
226
126
In looking at the Icelake Client die, the IPU is as big as a core and the USB Type-C is bigger than a core. Maybe they aren't getting that good of scaling on those parts.

To me it seems pretty "simple"; move what you can to the 14 nm PCH and connect the two using EMIB or Foveros. And if you can also turn the CPU and GPU into separate chiplets, all the better.
 
Apr 27, 2000
11,512
844
126
Comet and Rocket are also being released for mobile. It's not just for desktop.
Pretty sure Rocket Lake isn't even going to the desktop. Or the roadmap leaks made it look like desktop is only getting Comet Lake through 2020.
 

jpiniero

Diamond Member
Oct 1, 2010
6,327
226
126
Pretty sure Rocket Lake isn't even going to the desktop. Or the roadmap leaks made it look like desktop is only getting Comet Lake through 2020.
Rocket on desktop might not be until early 2021.

It makes sense- Comet in early 2020, Rocket in early 2021, and then a 7 nm product in 2022.
 

amdfan111

Junior Member
Feb 9, 2018
16
11
16
To everyone complaining about the "missing" 10 nm desktop parts, how can you still be unaware that Intel couldn't give a rat's ass about desktop? The last time Intel made a desktop CPU was the Pentium 4 (which was terrible). Everything since then has been an overclocked laptop processor.

The only company producing true desktop CPUs is AMD, which was first to 8-core and will soon be first to 16-core. In 2022, Intel will for sure have 10 nm and maybe even 7 nm, and it will not be on the desktop, because they sure as hell don't care about it! Meanwhile, AMD will probably have 5 nm Zen4 with 32 cores.

Sigh. No thread-crapping.

Take your viewpoints on AMD's product lineup
to the appropriate thread. (Hint: It's not this one).


AT Mod Usandthem
 
Last edited by a moderator:

Zucker2k

Senior member
Feb 15, 2006
757
65
136
Intel also gave an update on their 10-nanometer node. 10 nm has gone into high-volume production at the beginning of the year. “We are delivering, without compromise, on the original performance and scaling targets,” said Renduchintala. Their first-generation 10-nanometer node will follow by two additional inter-node optimizations – 10nm+ and 10nm++ planned for this year through 2021. It’s worth noting that even after 10 ramps, Intel will continue to leverage their 14-nanometer node for mix applications such as cost-sensitive products and to speed various developments wherever it makes sense.
Wow! Interesting times ahead, if he's not equivocating.
 

jpiniero

Diamond Member
Oct 1, 2010
6,327
226
126
Did a quick and dirty estimation of the Icelake 4+2 die size and got in the low 80s mm2, which is lower than I thought. Still think the fully enabled die yield is somewhere is in the teens %.
 

TheGiant

Senior member
Jun 12, 2017
365
46
76

jpiniero

Diamond Member
Oct 1, 2010
6,327
226
126
Looking at the benchmark config disclosure, seems that the Tigerlake 15W 96EU model would get around 5k in Firestrike. Of course being Firestrike, it's not actual games we are talking about. This is much faster than any mobile APU AMD offers today but assuming the Zen 2 memory controller can do 4200+ the 7nm APUs should be able to hit at least that 5k level.

Does sort of confirm that Tigerlake will be using chiplets, if they are continuing to increase the EU count.
 

Dayman1225

Senior member
Aug 14, 2017
869
68
96
Ice Lake Brings A New CPU, GPU, IPU, and I/Os, To Follow By Tiger Lake Next Year - WikiChip

David thinks each core has FIVR next to it, he also thinks System Agent has 1 and the GPU has 2.

Icelake is on 10+, not regular 10 like discussed at Investor Day.

Icelake also integrates Intel's 4th Generation IPU and the MIPI interface.

TB3 over Type-C is now natively on die apart from the high voltage power delivery subsystem which is still off die.

Tigerlake 4c 15w will have 96EUs in GT2 config

Since Tigerlake is back from the fab and booting with an Xe Graphics Engine it is likely that Xe dGPUs are too
 

coercitiv

Diamond Member
Jan 24, 2014
3,198
490
136
Maybe coffee intake is too low for monday morning, but if ICL is on 10+, what's being done on vanilla 10?
 

Brunnis

Senior member
Nov 15, 2004
490
12
91
Cannonlake is dead and buried. Intel won't be producing any more of those CPUs. Also, if IceLake-U/Y is 10nm+ then obviously Intel still has yield problems.
Well, whether CNL is in production or not and whether ICL-U/Y yields or not, the simple answer to coercitiv's question is what I wrote.
 

coercitiv

Diamond Member
Jan 24, 2014
3,198
490
136
Why would there be a vanilla 10 to speak of?
Because the "+" defines an improvement over a node that works. If the vanilla 10nm node is not be spoken of, then by definition 10+ is actually 10, more exactly their first 10nm working node.

Well, whether CNL is in production or not and whether ICL-U/Y yields or not, the simple answer to coercitiv's question is what I wrote.
Cannon Lake is no longer on Intel's roadmaps and keynotes, in fact they behave as if it never really existed in the first place.

The Cannon Lake based CPU needs to be searched by SKU id on Intel's CPU database, it's not listed in the 8th gen CPU list, nor is there a Cannon Lake page on Intel's site as all other codenames have. It's only purpose was to enable Intel claim their 10nm node was working, and it failed at that as well.
 


ASK THE COMMUNITY