• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."
  • Community Question: What makes a good motherboard?

Discussion Intel current and future Lakes & Rapids thread

Page 316 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

RTX

Junior Member
Nov 5, 2020
12
2
11
U series chips are also limited by the cooling of the laptop dimensions, heatsink, and the VRM components.

For a Desktop part, they'd shouldn't have problems extracting another 300-400 mhz out of it without those limitations out of the way? The 10810U can boost to 4.9 while the desktop part with 4 more cores boost to 5.3.
 

SAAA

Senior member
May 14, 2014
471
80
91
You are saying it will be based on 10nm, to be exact it will be (most likely) based on 10ESF, I don't think the max clock speed is a problem. Intels big cores are too power hungry unless they are clocked really low, on 10nm at least. They cannot add the same amount of big cores as AMD. Also Willow Cove is quite a lot bigger than Zen3 and Golden Cove likely will be even bigger than Willow Cove, this is another issue with a limited 10nm capacity and subpar yields. There are a few reasons for Intel to go for the big+little route.
On the issue of size I have to say Intel 10SF process has definitely one good point: same density as 7nm from TSMC.
Looking at the picture I posted above the L2 is 0.5 MB vs 1.25 MB in 0.8 and 2 mm^2 square respectively, the density comes out equal for SRAM in actual products, at about the same running frequencies. With the caveat that Tigerlake 8 core might push above 5GHz turbo, without overclocking.

As for big little… if small cores are this small I seriously wonder why they don't plan to release parts entirely made of them, or like 4+16 small cores, 8+32 small and so on.
Even servers where frequencies are usually lower so not as big an issue… with rumors for Icelake server parts in the 370 to 640mm^2 range why not release a 128 or 256 core Gracemont? Those 256 cores could fit in the small dice range!
 

mikk

Platinum Member
May 15, 2012
2,955
767
136
U series chips are also limited by the cooling of the laptop dimensions, heatsink, and the VRM components.

For a Desktop part, they'd shouldn't have problems extracting another 300-400 mhz out of it without those limitations out of the way? The 10810U can boost to 4.9 while the desktop part with 4 more cores boost to 5.3.

Singlethread boost clock speed isn't the problem for a big core count sample. I believe 6 cores would have worked out nicely for TGL-U 28W with a MT clock speed reduction but 8C I don't think so considering that they are a lot bigger and more power hungry than Renoir/Zen 2. Another 300-400 Mhz isn't enough for a desktop part, at 28W 4C Willow Cove can run with 3.2 Ghz in Cinebench R20, they need around 4 Ghz for a desktop part. Ryzen 9 5950X (16C!) can run with 4 Ghz in MT workloads, I really doubt 16 big cores from Intel can do this on 10nm.
 

IntelUser2000

Elite Member
Oct 14, 2003
7,249
1,839
136
As for big little… if small cores are this small I seriously wonder why they don't plan to release parts entirely made of them, or like 4+16 small cores, 8+32 small and so on.
I hope to see that with their 7nm parts. Alderlake is just a second iteration of hybrid and first try on desktops.

Even servers where frequencies are usually lower so not as big an issue… with rumors for Icelake server parts in the 370 to 640mm^2 range why not release a 128 or 256 core Gracemont? Those 256 cores could fit in the small dice range!
If you take the 2nd generation Xeon Phi, cut the vector width in half and remove the HMC memory you'll get what you are suggesting. I am also very curious how a latest version based on Gracemont would have turned out but I doubt they'll go that route again.

What you are saying would work in an ideal world(and on paper) but in reality it takes a lot of effort and resources just to get it working. There's no guarantee it won't miss the original goals by a lot and end up being behind Core-based server chips.

It's much easier to slap cores together. It's much harder to actually get it working better than what's out there, especially against the team that's been doing it for years and have more resources behind it.

Many server applications also need high per thread performance and many cores.
 
  • Like
Reactions: uzzi38

RTX

Junior Member
Nov 5, 2020
12
2
11
Singlethread boost clock speed isn't the problem for a big core count sample. I believe 6 cores would have worked out nicely for TGL-U 28W with a MT clock speed reduction but 8C I don't think so considering that they are a lot bigger and more power hungry than Renoir/Zen 2. Another 300-400 Mhz isn't enough for a desktop part, at 28W 4C Willow Cove can run with 3.2 Ghz in Cinebench R20, they need around 4 Ghz for a desktop part. Ryzen 9 5950X (16C!) can run with 4 Ghz in MT workloads, I really doubt 16 big cores from Intel can do this on 10nm.
For a -S series chip, 4.5+ should be achievable with 95-125 TDP. I mean it's only 3.1 at 45W for the 8 core. Going from 10980HK ( 2.4 ghz ) to 10700K ( 3.8ghz ) is about 1.4ghz diff.
 

mikk

Platinum Member
May 15, 2012
2,955
767
136
For a -S series chip, 4.5+ should be achievable with 95-125 TDP. I mean it's only 3.1 at 45W for the 8 core. Going from 10980HK ( 2.4 ghz ) to 10700K ( 3.8ghz ) is about 1.4ghz diff.

I don't think so. Even AMD cannot do this with their 16C if it runs on default. The power consumption isn't going linearly up, 4+ Ghz is out of the efficiency sweetspot.
 

RTX

Junior Member
Nov 5, 2020
12
2
11
Approximately? If we assume 3.1 is at 0.8v and it required 1.15v for 4.5, that should bring it to ~125W? ( 120w + 5w for the igpu? )
 

Exist50

Member
Aug 18, 2016
139
186
116
I hope to see that with their 7nm parts. Alderlake is just a second iteration of hybrid and first try on desktops.



If you take the 2nd generation Xeon Phi, cut the vector width in half and remove the HMC memory you'll get what you are suggesting. I am also very curious how a latest version based on Gracemont would have turned out but I doubt they'll go that route again.

What you are saying would work in an ideal world(and on paper) but in reality it takes a lot of effort and resources just to get it working. There's no guarantee it won't miss the original goals by a lot and end up being behind Core-based server chips.

It's much easier to slap cores together. It's much harder to actually get it working better than what's out there, especially against the team that's been doing it for years and have more resources behind it.

Many server applications also need high per thread performance and many cores.
You forget to factor in politics and corporate culture. Core was Intel's hammer for every computing nail. I bet they didn't even consider using Atom in a normal server product. But Core has stagnated in a way that Atom has not, so now it actually seems viable.

As for an actual product, there's no reason that a manycore Gracemont chip wouldn't work well. Sure, there are some markets that need the best single thread performance, but for the majority of cloud workloads, Skylake-esque performance should be plenty sufficient.
 
  • Like
Reactions: Tlh97 and uzzi38

Dave2150

Senior member
Jan 20, 2015
601
131
116
Any news on Rocket Lake's release date? I know 2021, though as Alder Lake is also rumoured to launch in 2021, I'm guessing Q1 for Rocket Lake and Q4 for Alder Lake?
 

Nereus77

Member
Dec 30, 2016
139
248
86
Any news on Rocket Lake's release date? I know 2021, though as Alder Lake is also rumoured to launch in 2021, I'm guessing Q1 for Rocket Lake and Q4 for Alder Lake?
Yes, Q1 for RL and Q4 / Q1 2022 for AL. Probably.
I think your guess is as good as mine at this point.
 

Dave2150

Senior member
Jan 20, 2015
601
131
116
Yes, Q1 for RL and Q4 / Q1 2022 for AL. Probably.
I think your guess is as good as mine at this point.
Fair enough. My 6700k@4.7Ghz, with 32GB of 3600Mhz C15 memory, is seemingly adequate to power my 3080 at 4k, though too many games are forcing it to 70-80% utilization for my liking, so keen to upgrade soon.

Considered Ryzen 5000, but not keen to jump into a dead platform in it's last hour.
 

Dave2150

Senior member
Jan 20, 2015
601
131
116
That would eliminate LGA1200 from consideration, including Rocket Lake.
LGA1200 isn't a dead platform, as it has a future CPU that will be supported on current boards - Rocket Lake.

The true next generations platforms though as as follows:

LGA1700 - Intel
AM5 - AMD

Not sure I want to wait that long for the DDR5 platforms, will see.
 
  • Like
Reactions: shady28

Ajay

Diamond Member
Jan 8, 2001
7,584
2,738
136
LGA1200 isn't a dead platform, as it has a future CPU that will be supported on current boards - Rocket Lake.

The true next generations platforms though as as follows:

LGA1700 - Intel
AM5 - AMD

Not sure I want to wait that long for the DDR5 platforms, will see.
I don’t think it's a big deal to be on a current gen platform with no future upgrade path, most of us will keep our systems for the 2-3 years that retains socket compatibility (if you buy as soon as a new socket comes out). You seem to have plenty of good RAM any current 6-8 core cpu will give you great performance for several years to come. Wait for the Rocket Lake reviews and buy whatever suits your needs. You buy good graphics cards, so a few frames +/- wont really matter.
 
  • Like
Reactions: Tlh97

jpiniero

Diamond Member
Oct 1, 2010
8,489
1,484
126
What the actual flabbergast?!
There were rumors of a 4 core Tiger-H die, but how it was going to be branded remained to be seen. Maybe they decided to just use the Tiger-U die instead. As to why i7, I thought that perhaps it was a mistake, but the 4.8 turbo suggests otherwise.
 

jpiniero

Diamond Member
Oct 1, 2010
8,489
1,484
126
Not sure why you think that excludes their being an 8 core H model.
It doesn't, but all of the Comet Lake H i7 models were 6 or 8 core. Obviously the U die only has 4 cores max so that's the limit if they go that way.

For anything higher than 4 they would presumably need to use the 8 core die. But if they can't get it to yield, using the U die and just jacking up the TDP makes for a plausible backup plan.
 

mikk

Platinum Member
May 15, 2012
2,955
767
136

Maybe Tiger Lake-H is just going to be a 35W model of Tiger Lake-U. Only 4 cores max.

Intel wouldn't launch TGL-H if it's TGL-U with just 7W higher TDP, from the SKU number 113 it's clear the 4/8 TGL-H is a low one and not the maximum. With a bit more logical thinking you could avoid such brain fart. And if this isn't enough there was an 8C TGL-H userbenchmark entry a few days ago.
 
  • Like
Reactions: Zucker2k and Tlh97

DrMrLordX

Lifer
Apr 27, 2000
16,627
5,634
136
LGA1200 isn't a dead platform, as it has a future CPU that will be supported on current boards - Rocket Lake.
Um, you don't realize what I was saying, do you?

Wait for RKL launch, then you can pick the best of two dead platforms.
That's what I meant. LGA1200 + Rocket Lake and AM4 + Vermeer - both dead-ends. It's not AMD's fault it's taking Intel so long to get Rocket Lake to market. Unless you really think Comet Lake is competition?

Maybe Tiger Lake-H is just going to be a 35W model of Tiger Lake-U. Only 4 cores max.
If yields on the 8c parts are bad enough then maybe that's the case. But still. Ughhhhh
 
  • Like
Reactions: Tlh97 and coercitiv

scineram

Junior Member
Nov 1, 2020
10
5
36
Intel wouldn't launch TGL-H if it's TGL-U with just 7W higher TDP, from the SKU number 113 it's clear the 4/8 TGL-H is a low one and not the maximum. With a bit more logical thinking you could avoid such brain fart. And if this isn't enough there was an 8C TGL-H userbenchmark entry a few days ago.
Fair point, but how do you explain the i7 branding for the 4 core? Maybe bigger die is really not viable. Did the 8 core entry come with a branding string?
 

ASK THE COMMUNITY