Discussion Intel current and future Lakes & Rapids thread

Page 107 - 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
I would assume, that under more realistic gaming conditions, that the user would reduce GFX setting to achieve a playable framerate, which would make the CPU usage going up - leaving less power for the GPU.
Wouldnt you think so?

No, and this is why Furmark is unrealistic. Games don't behave like Furmark. It's like the fundamental code is different. It's Linpack for 3D. Linpack doesn't care about memory bandwidth, or that even in HPC, codes have an ILP limit, or that it doesn't fit conveniently in the CPU caches, or that interconnects matter. You'll notice it right away if you have an x86 desktop and a laptop to run them on.

Real world games also have game developers working closely with video card vendors so people can play them. They'll have people sitting in front of the screen and reduce demanding features that stress the GPU too much relative to the visual impact. You wouldn't enable AF 16x on the lowest setting. Furmark doesn't care because its a stress test that pretends its a 3D benchmark.

Do you know what a good 3D benchmark is? 3DMark before they started copying(Firestrike, Cloud Gate, Time Spy) mobile benchmarks. The latest is called 3DMark11. The older versions were good back in their days. I can approximate how the iGPUs perform simply by looking at 3DMark11 scores.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
That has little to do with security. Whiskeylake, Amberlake and Coffelake CPUs are not affected by the Hyperthreading security bug. The older chips are.

That's not true. Only the updated steppings of Whiskey Lake and Coffee Lake Refresh (R0) are not affected by MDS. Presumably Comet Lake is also not affected.

Now Intel not doing 10C20T because of heat does seem plausible. But that would really wreck the lineup. You'd be better off just leaving it at 8 cores.
 
  • Like
Reactions: Gideon

Thala

Golden Member
Nov 12, 2014
1,355
653
136
No, and this is why Furmark is unrealistic. Games don't behave like Furmark. It's like the fundamental code is different. It's Linpack for 3D. Linpack doesn't care about memory bandwidth, or that even in HPC, codes have an ILP limit, or that it doesn't fit conveniently in the CPU caches, or that interconnects matter. You'll notice it right away if you have an x86 desktop and a laptop to run them on.

This might very well be the case. Still are you implying that a 15TDP laptop never throttles its GPU under more realistic gaming conditions? I am asking because my conclusion that CPU power would go up under more realistic gaming conditions is still valid (compared to Furmark). And that the cores can easily saturate 15W on their own is pretty much given.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
This might very well be the case. Still are you implying that a 15TDP laptop never throttles its GPU under more realistic gaming conditions? I am asking because my conclusion that CPU power would go up under more realistic gaming conditions is still valid (compared to Furmark). And that the cores can easily saturate 15W on their own is pretty much given.

It technically "throttles" if you see that the top Turbo is at 1.15GHz and 15W parts only runs it at ~1GHz. Also because the iGPUs in 35W and 45W CPUs are faster. At one point, there was a 50% difference between the one in the 15W one and a 45W one. I attributed 20% to the iGPU and 25% to the faster CPU back in Ivy days with 1.25GHz Turbo. The differences may have shrunk with the latest generations though.

The CPU power may increase with games, but it does not take high frequencies to offer enough power for the GPU, thus the GPU ends up being dominant power user in most cases.

https://www.anandtech.com/show/13471/the-microsoft-surface-pro-6-review-more-than-a-color/4

Iris with 48EUs start running into power limit issues. After accounting for that discrepancy, the GT3e Iris parts are 30-50% faster than GT2 HD ones. Without throttling the differences may be 50-80%. The GT2 may simply be undersized or their drivers target GT2 and behave better under games.

They have anemic performance, but its also maintains performance under battery power. Most dGPU systems throttle quite heavily. My sig Ivy Bridge ultrabook runs some games better under Balanced than with high Performance settings, which is an interesting demonstration of power management being effective.
 
Last edited:

cortexa99

Senior member
Jul 2, 2018
318
505
136
hmmmm anybody have source about this image? i dont know whether or not this has been post before.
pls click to open for original size cuz it's too large i have to just post thumbnail for not interrupt reading lol
it shows icelake has 40% increase against coffeelake in cpuz benchmark(avx512?)

9945564e9258d109330a483bdf58ccbf6d814d5b.jpg
 

Bouowmx

Golden Member
Nov 13, 2016
1,138
550
146
Alleged results exceed aggregate results by a bit. Also, comparing Intel Core i9-9900KS 5000 MHz scoring 600 to i9-9900K 5400 MHz scoring 660: 1.1x score for 1.08x frequency, which is kind of odd to see such "superlinear" relation.

Finally, Sunny Cove 6-core: I have no memory of this config of ever being a thing (could be wrong).
 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
hmmmm anybody have source about this image? i dont know whether or not this has been post before.
pls click to open for original size cuz it's too large i have to just post thumbnail for not interrupt reading lol
it shows icelake has 40% increase against coffeelake in cpuz benchmark(avx512?)

View attachment 7411


Because of the Sunny Cove 6C12T I would say this is a fake. There is nothing about a Sunny Cove 6C in the wild. From where did you get this, why don't you post your source?
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Yeah, unless Rocket Lake is using Sunny Cove it's definitely fake. I still think it's more likely to be using Willow Cove if it's not Skylake. There's been no indication that Intel is planning a 6 core model for Icelake (or Tigerlake for that matter)
 

birdie

Member
Jan 12, 2019
98
71
51
You must be mad to compare CPU architectures performance based on the CPU-Z benchmark alone. Basically nothing is known about it but what I know is that it severely penalizes previous-gen CPUs, e.g. a 4GHz Sandy Bridge CPU is significantly slower than a 4GHz Sky Lake CPU in this benchmark, however in a lot of other benchmarks/workloads they are very close.
 
Last edited:

ondma

Platinum Member
Mar 18, 2018
2,718
1,278
136
You must be mad to compare CPU architectures performance based on the CPU-Z benchmark alone. Basically nothing is known about it but what I know is that it severely penalizes previous-gen CPUs, e.g. a 4GHz Sandy Bridge CPU is significantly slower than a 4GHz Sky Lake CPU in this benchmark, however in a lot of other benchmarks/workloads they are very close.
I suppose you could find some cherry picked benchmarks where SB is close to skylake, but overall the average IPC gain is about 25%, even according to Anand's early skylake tests, which IIRC used slow ram for skylake.
 
  • Like
Reactions: mikk

Asterox

Golden Member
May 15, 2012
1,026
1,775
136
I suppose you could find some cherry picked benchmarks where SB is close to skylake, but overall the average IPC gain is about 25%, even according to Anand's early skylake tests, which IIRC used slow ram for skylake.

But overall, for example in gaming you can see 25% only on paper.

 

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
But overall, for example in gaming you can see 25% only on paper.



Because they are testing a GPU bottlenecked environment which is dumb if they want to find out the real speed difference for CPUs, the gaming difference between Sandy Bridge and Skylake is pretty large.
 
  • Like
Reactions: Lodix

birdie

Member
Jan 12, 2019
98
71
51
Because they are testing a GPU bottlenecked environment which is dumb if they want to find out the real speed difference for CPUs, the gaming difference between Sandy Bridge and Skylake is pretty large.

I know you want to believe that you payed for a worthy upgrade but it's not 25% in all tests. It's far from that.

https://www.anandtech.com/show/9483/intel-skylake-review-6700k-6600k-ddr4-ddr3-ipc-6th-generation/9
https://www.hardocp.com/article/2017/01/13/kaby_lake_7700k_vs_sandy_bridge_2600k_ipc_review/

Some people fail to realize that the progress in IPC has recently more or less stalled completely and we get anemic yearly gains. OK, in theory Ice Lake has a 18% higher IPC performance than Sky Lake, but the latter is already a four year architecture which means we get just 4% every year and again these 4% are not across all the workloads.
 
  • Like
Reactions: lightmanek

mikk

Diamond Member
May 15, 2012
4,112
2,108
136
I know you want to believe that you payed for a worthy upgrade but it's not 25% in all tests. It's far from that.

Pretty close to 25% in your anandtech link. The second link doesn't work. But what is your problem? I answered to someone with a GPU limited test which isn't any meaningful for a CPU test, I haven't even claimed the IPC gap is 25%, but for sure the gaming difference is pretty large if it isn't GPU limited, even though you came here with a flame.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
I've been out of the loop for a while. So it seems from the roadmap leaked by Tweakers that, in fact, for the desktop 10nm is dead like SemiAccurate reported late last year. So for desktop it will likely remain 14nm until a 7nm transition somewhere in 2022/2023. Ouch.

For the architecture, it appears Rocketlake is 14nm but Sunny/Willowcore based, and Comet lake is yet undetermined whether it is Skylake or Sunny Cove based. Anyone have any rumors / info on that?

For now, Comet Lake being pushed back from late 2019 to mid 2020, the lack of info, and the broad launch accross mobile and desktop IMO points to Sunny Cove for consumers moving forward, whether on 10 or 14nm.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Comet is definitely a Skylake core with Gen 9.5 graphics. While the desktop version isn't coming out until next year the U/Y parts should be announced soon.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
On a related note, not sure whether discussed here, but the mobile roadmap from 2018-2020 seems to be slightly older, as it shows a Comet Lake U launching in 19Q3. In the second roadmap it's pushed back to 20Q2, and launching across a wide range of lines (Y/U/H/G/S). This to me points to a reset moment of Comet Lake, probably around the same time they decided to cancel the 10nm desktop (and mainstream mobile) and as such a decision to port Sunny Cove to 14nm.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
Comet is definitely a Skylake core with Gen 9.5 graphics. While the desktop version isn't coming out until next year the U/Y parts should be announced soon.

Yes I know you've been harping on this for ages without any proof to show for it, and in spite of all the announcements by Intel and changes in the roadmap.

See the wide deployment of Comet Lake in mobile, no way in hell a niche low-power product will remain their most advanced architecture two years down the road.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Yes I know you've been harping on this for ages without any proof to show for it, and in spite of all the announcements by Intel and changes in the roadmap.

Pretty sure it's been confirmed from the Linux support that Comet Lake is using a Skylake core. The IGP is definitely Gen 9.5 from the driver.

By the way the IGP team has been working on supporting Elkhart Lake but no sign of Gen 12 yet.
 

Spartak

Senior member
Jul 4, 2015
353
266
136
Like I said, the initial Comet Lake seems to be cancelled. It's been pushed back and will launch across almost their complete client line-up next year. That's one and a half year after the Sunny Cove announcement where they decoupled architecture from process.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
On a related note, not sure whether discussed here, but the mobile roadmap from 2018-2020 seems to be slightly older, as it shows a Comet Lake U launching in 19Q3. In the second roadmap it's pushed back to 20Q2, and launching across a wide range of lines (Y/U/H/G/S).

No its not. The Client Commercial roadmap isn't regular client, but a stable platform for business. They have a different roadmap and come later. The roadmap shows Whiskeylake in Q2 of this year, when it was Q3 of last year. Coffeelake H is shown as Q3, when it was early Q2. It also does not show any Icelake.

The other roadmap is relevant for regular client. Whiskey/Amber/CFL H is correct, and Icelake is there too. Some may be little off but at least its the correct roadmap.
 
  • Like
Reactions: Spartak

Spartak

Senior member
Jul 4, 2015
353
266
136
No its not. The Client Commercial roadmap isn't regular client, but a stable platform for business. They have a different roadmap and come later. The roadmap shows Whiskeylake in Q2 of this year, when it was Q3 of last year. Coffeelake H is shown as Q3, when it was early Q2. It also does not show any Icelake.

The other roadmap is relevant for regular client. Whiskey/Amber/CFL H is correct, and Icelake is there too. Some may be little off but at least its the correct roadmap.

Thanks for clearing up. So it will still launch in 19Q3. We'll found out soon enough then.

If the delay is 3 quarters on average, this'd mean we'd see Rocket Lake with Sunny Cove for consumer desktop around 20Q3.
 
Last edited:
  • Like
Reactions: SK10H

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Thanks for clearing up. So it will still launch in 19Q3. We'll found out soon enough then.

If the delay is 3 quarters on average, this'd mean we'd see Rocket Lake with Sunny Cove for consumer desktop around 20Q3.

I imagine Rocket Lake Desktop will ship a year from when Comet Lake Desktop does.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
If the delay is 3 quarters on average, this'd mean we'd see Rocket Lake with Sunny Cove for consumer desktop around 20Q3.

Comet Lake S with 10 cores should be coming later this year, so if we add a year from that, we should see Rocket Lake S in 2020 Q3.

Because of the Sunny Cove 6C12T I would say this is a fake. There is nothing about a Sunny Cove 6C in the wild. From where did you get this, why don't you post your source?

Reading the google translation of the comments, he says it also contains scores of the Sunny Cove based desktop chip.

40% over SKL is possible, but of course its not the average. You can even see in Intel charts SNC being 40% faster in one application. Also, CPU-Z is a useless benchmark. They should just stick to identifying CPUs.