Discussion Intel current and future Lakes & Rapids thread

Page 276 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

ondma

Platinum Member
Mar 18, 2018
2,718
1,278
136
Based upon the leaked Geekbench score, RKL can hit 5 Ghz but ST could end up being still slower than the 1165G7 because the IPC is lower.
Max turbo for the 1165G7 is 4.7, but can it maintain that, even in sustained single threaded benchmarks?
 

eek2121

Platinum Member
Aug 2, 2005
2,904
3,905
136
If I were Intel, I would be riding 10SF hard and focus all effort on getting desktop and server chips out the door.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
If I were Intel, I would be riding 10SF hard and focus all effort on getting desktop and server chips out the door.

Desktop demand has been cripped by The Rona. Yields are still too bad to seriously do server chips. Can't cancel Ice Lake Server because Wall Street would freak out, but they will in practice.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
30-40% architectural performance improvement from Xe LP over Gen11 says Intel. LPDDR5 support is coming in 2021.

The cdyn improvement refers to power use. And in turn they can use that for frequency, which they did partially for Tigerlake. I say partially since frequency improvement is less.
 
  • Like
Reactions: mikk

mikk

Diamond Member
May 15, 2012
4,111
2,105
136
Seems unlikely. I thought RKL was expected to boost lower than Comet Lake due to being a backport of a core that was not designed for 14nm.

Stepping 0 already hits 5.0 Ghz and it's not unusual for Intel they increase it further in stepping 1 or 2. Tigerlake clocked at 4.3 Ghz with a Stepping 0 and most people thought this is final.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
I'm seeing parallels between Tigerlake and Ampere. Both seem to increase power consumption for the performance increases, and both are taking their sweet time getting released. Both also seem to be using benchmarks that use deep learning, and ray tracing to look better than it is.

The increased power consumption for better performance is likely a trend going forward.

Seems unlikely. I thought RKL was expected to boost lower than Comet Lake due to being a backport of a core that was not designed for 14nm.

Being a backport of a core meant for the next gen process is why it only has 8 cores.
 

ondma

Platinum Member
Mar 18, 2018
2,718
1,278
136
I'm seeing parallels between Tigerlake and Ampere. Both seem to increase power consumption for the performance increases, and both are taking their sweet time getting released. Both also seem to be using benchmarks that use deep learning, and ray tracing to look better than it is.

The increased power consumption for better performance is likely a trend going forward.



Being a backport of a core meant for the next gen process is why it only has 8 cores.
I am not really interested in multi core productivity apps, so except for the new consoles, I would have been more than satisfied with 8 cores, ~15% IPC gain, and still able to reach 5ghz all core. Now though, I have a feeling 8 cores may be a bit low for gaming going forward.
 

A///

Diamond Member
Feb 24, 2017
4,352
3,154
136
I believe I remember that "rumor" you're referring to. It was just someone on a Chinese forum speculating about what Intel could do, or what they wanted Intel to do. Not an actual leak.
Quite probably. It wasn't Chip Hell. I want to say I saw it on a thread on another site that had a twitter embed, but I'm not sure. Doesn't help that I clear my history often.
 

A///

Diamond Member
Feb 24, 2017
4,352
3,154
136
Yes, absent limited cooling, Tiger Lake can boost to 4.8Ghz, and AnandTech stated that apparently 5Ghz was not out of reach. In Geekbench TGL had no issue hitting 4.7-4.8ghz.

edit: https://browser.geekbench.com/v5/cpu/3695518.gb5
It's not that I'm looking to beat Intel's legs from under them here, but how long can TGL sustain that all core boost? Especially given how laptops always have gimped cooling solutions, some more than others, and that laptop idles are much higher than desktop equivalents and have less thermal room to play in.
 

mikk

Diamond Member
May 15, 2012
4,111
2,105
136
Quite probably. It wasn't Chip Hell. I want to say I saw it on a thread on another site that had a twitter embed, but I'm not sure. Doesn't help that I clear my history often.

It's from MebiuW, his 4.7 Ghz Tigerlake turbo boost prediction was spot on when nobody could believe in this. He has a source in the background which he occasionally uses afaik, it's not just a random speculation, although with all of such unconfirmed stuff nothing is safe.
 
  • Like
Reactions: Tlh97 and A///

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
It's from MebiuW, his 4.7 Ghz Tigerlake turbo boost prediction was spot on when nobody could believe in this. He has a source in the background which he occasionally uses afaik, it's not just a random speculation, although with all of such unconfirmed stuff nothing is safe.

Do you have a link? And I'll point out that I predicted 4.8GHz for Tiger Lake, yet I still think 5.5 is very unrealistic for Rocket Lake.
 

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
Do you have a link? And I'll point out that I predicted 4.8GHz for Tiger Lake, yet I still think 5.5 is very unrealistic for Rocket Lake.
IF 5.5GHz is even close to the real SC clocks then Zen 3 must be packing a hefty punch for Intel to want to push clocks that far, or, to maintain single-thread superiority on desktop instead of mobile, which some speculated could be a possibility seeing how the 1185G7 is turboing to 4.8GHz. All these is assuming these clocks are even realistic in the first place.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
IF 5.5GHz is even close to the real SC clocks then Zen 3 must be packing a hefty punch for Intel to want to push clocks that far,

Is it really 5.5GHz? Because the 10900K can realistically only do 5.1GHz, 5.2GHz in good days, and 5.3GHz for certain cores when its cool enough and for fractions of a second. Despite that, it uses 300W, and reaches 70C using liquid cooling. And it has nearly zero overclocking headroom.

That means that if games are really becoming multi-threaded, then those numbers are meaningless anyway. How is pretending it can reach 5.3GHz help the CPU in any way?

Oh I know. It pretends it can help Intel. Same as with the atrocious Tigerlake marketing, where they had to paper launch it 3 times. "We can reach 5.5GHz!" But for 1) Only for milliseconds 2) Only for certain cores 3) If its cool enough.

Nothing fundamental has changed in the last 20 years. The clocks that required water cooling, and sometimes even exotic cooling became more mainstream because we have heatsinks that are multiple the size and weight it was in those days.

This is the very problem of pushing clocks. At some point they need to actually and consciously back down, like they did with Netburst to Core transition. If they choose to do so architecturally, they'll net additional benefits such as a less complex core, and higher efficiency, which means it'll benefit the larger and higher ASP mobile and server market.

By choosing clocks it also means Rocketlake is going to be actually in a less favorable position against AMD. Intel has to cut cores, and whatever they gain due to using Sunny Cove uarch is mostly cancelled out by Zen 3. The focus on ever unattainable clocks mean power efficiency will further go down.

Now though, I have a feeling 8 cores may be a bit low for gaming going forward.

I'm pretty sure no articles exist that show whether games actually gain from 8C/16T to 10C/20T. If it does though, more boohoo for Rocketlake. 1/3rd of the hard earned uarch gains canceled out by insane quest for imaginary clocks!

Prescott-2M Round 2.
 
Last edited:

uzzi38

Platinum Member
Oct 16, 2019
2,565
5,568
146
It's from MebiuW, his 4.7 Ghz Tigerlake turbo boost prediction was spot on when nobody could believe in this. He has a source in the background which he occasionally uses afaik, it's not just a random speculation, although with all of such unconfirmed stuff nothing is safe.

His 4.7GHz "prediction" was him copying DavidBepo. DavidBepo treated it as a prediction. MebiuW paraded that information around Weibo as a leak with his source being Bepo. Eventually he posted Bepo's napkin maths on TGL-U's V/f curve there and he got shut down by the people there who actually knew about Tiger Lake, so he came onto Twitter instead.
 

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
By choosing clocks it also means Rocketlake is going to be actually in a less favorable position against AMD. Intel has to cut cores, and whatever they gain due to using Sunny Cove uarch is mostly cancelled out by Zen 3. The focus on ever unattainable clocks mean power efficiency will further go down.
Well, Intel is not going to fight an efficiency war with AMD right now. They'll need more cores to do that. Right now, their best option is to push fmax through the roof, power consumption be damned. It's a necessary brute force approach, albeit a temporary one until their process woes are solved. There's no other way around it, save abandoning their fabs or giving up competing altogether.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Well, Intel is not going to fight an efficiency war with AMD right now.

Process will only help so far. Dothan faired much better than Prescott, despite using the same 90nm process.

We know from Tremont their 10nm process can be competitive on the density front. It's just their big Cores that suck, and its the insane focus on clocks. Why not 4.5-4.7GHz like AMD? Leave the 5GHz+ for overclockers.

Instead, they overclock it themselves and sell it as 5GHz. Remember when we used to have incredible headroom?

There's high clocks and there are insane clocks. The latter what Intel is purusing right now. Not only it requires more transistors, but with less density.

For example, Ocean Cove should be a 4.5GHz chip with 25% higher perf/clock, rather than a 5.2GHz one with 20% higher perf/clock. Overall it might seem a loss, but not pushing it all the way means users can choose by overclocking. Servers and Laptops will benefit from a better architecture.
 

ondma

Platinum Member
Mar 18, 2018
2,718
1,278
136
Process will only help so far. Dothan faired much better than Prescott, despite using the same 90nm process.

We know from Tremont their 10nm process can be competitive on the density front. It's just their big Cores that suck, and its the insane focus on clocks. Why not 4.5-4.7GHz like AMD? Leave the 5GHz+ for overclockers.

Instead, they overclock it themselves and sell it as 5GHz. Remember when we used to have incredible headroom?

There's high clocks and there are insane clocks. The latter what Intel is purusing right now. Not only it requires more transistors, but with less density.

For example, Ocean Cove should be a 4.5GHz chip with 25% higher perf/clock, rather than a 5.2GHz one with 20% higher perf/clock. Overall it might seem a loss, but not pushing it all the way means users can choose by overclocking. Servers and Laptops will benefit from a better architecture.
Not saying you are wrong. However, designing a, lets call it "more efficient" core and process does not necessarily mean you can extract still more performance by overclocking. Just look at AMD as an example. Overclocking is not worth it in most cases; you are better served with tweaking ram.
 

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
Instead, they overclock it themselves and sell it as 5GHz. Remember when we used to have incredible headroom?
I agree completely. I've made this point many times before and been called out for being partial but I can tell you exactly when this new trend of factory overclocked chips began. Do you remember in 2017 when AMD dropped Zen on the desktop? That's when this trend began. The only reason why we're even discussing this right now is because Intel is running on a less efficient process so power consumption is necessarily higher. Look at the current releases from both companies - there's barely any headroom at all. In some ways, this is a consequence of competition; both companies are leaving no headroom for enthusiasts. Performance now relies heavily on cooling capacity, more than anything else.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Not saying you are wrong. However, designing a, lets call it "more efficient" core and process does not necessarily mean you can extract still more performance by overclocking

Intel almost always had much higher headroom. It's only recently starting with Coffeelake they started using all the headroom and selling it as stock.

A 125W CPU actually using 300W is proof they are pushing it way past reasonable limits.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Deceptive Intel marketing-

When they realized they made a fatal mistake not making a new core purpose-built for the 14nm process that would have been a successor to Skylake, they started the +++ branding. A purpose built 14nm core is better than a backport.

Kabylake to Coffeelake: Zero changes
Goldmont to Goldmont Plus: A massive 30% perf/clock gain and big uarch change. The Geminilake platform had little fanfare on release day. No mention of it to press.

They know that enthusiasts follow code names. The plusses may not be more than a glorified stepping or a respin.

AMD's Zen came exactly when Intel management was at its worst. It's not that the original Zen was fantastic, but its competitor was falling pretty badly.
 

mikk

Diamond Member
May 15, 2012
4,111
2,105
136
His 4.7GHz "prediction" was him copying DavidBepo. DavidBepo treated it as a prediction. MebiuW paraded that information around Weibo as a leak with his source being Bepo. Eventually he posted Bepo's napkin maths on TGL-U's V/f curve there and he got shut down by the people there who actually knew about Tiger Lake, so he came onto Twitter instead.

MebiuW is unrelated to David and it's the other way around he did use the predictions from MebiuW for his further "analysis"

This was the first high Turbo mention from Tigerlake on Twitter:

 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,784
136
I can tell you exactly when this new trend of factory overclocked chips began. Do you remember in 2017 when AMD dropped Zen on the desktop? That's when this trend began.

Good job calling out AMD in an Intel thread. Never mind that even the 1800x wasn't "factory overclocked" any more than Kaveri, Godavari, or desktop Carrizo (x4 945). Those chips had very little upper headroom due to process and/or cache latency (that was an issue for Carrizo, apparently). If you want to point to a "factory overclocked" CPU from AMD, you really have to go back to the FX9590 and 9370.

In any case, I don't think we'll see clocks from Rocket Lake-S that are as high as Comet Lake-S.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
@DrMrLordX Umm, I don't think Zucker2k is saying Zen itself started the factory overclocked trend. He's saying the release of Zen made Intel release factory overclocked chips in response to it.
 
  • Like
Reactions: dr1337

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,784
136
@DrMrLordX Umm, I don't think Zucker2k is saying Zen itself started the factory overclocked trend. He's saying the release of Zen made Intel release factory overclocked chips in response to it.

Maybe. At that point it devolves into an argument of what chips are "factory overclocked". Regarding Rocket Lake-S, I expect power usage in line with the 10900K but with lower clocks (well, certainly not anything above 5 GHz). Does that make it a factory-overclocked chip?