Discussion Intel current and future Lakes & Rapids thread

[eek2121]

No, it doesn't. @dullard claimed Intel were running the Gracemonts close to the voltage floor which Golden Cove can't drop down to. That would be FAR less than 3.9GHz. Actually probably close to or below the 1.8GHz base clock you say here.

3.9GHz is close to the absolute maximum they can clock to full stop. It is very much high frequency for the Gracemont cores, even if not high frequency in relation to Golden Cove cores.

I don’t make the rumors I just read them:

Specifications of Intel Core i9-12900K, i7-12700K and i5-12600K Alder Lake CPUs leak out - VideoCardz.com

Alleged specifications of Intel Alder Lake processors According to a post by 我用第三人称 on Zhihu forums, the specifications of three K-series processors from the upcoming Inte Alder Lake-S series have been leaked. The specifications are supposedly based on qualifications samples, which from what we...

videocardz.com

It is entirely possible that Gracemont runs efficiently at 3.9 ghz (3.7ghz all core). 10ESF is said to be 15% more efficient than 10SF.

However, I have always maintained that Alder Lake is about die area not power.

EDIT: Think of it also like this. Gracemont is the better core. It may be more efficient at these clockspeeds than Golden Cove.

 

[uzzi38]

I don’t make the rumors I just read them:

Specifications of Intel Core i9-12900K, i7-12700K and i5-12600K Alder Lake CPUs leak out - VideoCardz.com

Alleged specifications of Intel Alder Lake processors According to a post by 我用第三人称 on Zhihu forums, the specifications of three K-series processors from the upcoming Inte Alder Lake-S series have been leaked. The specifications are supposedly based on qualifications samples, which from what we...

videocardz.com

It is entirely possible that Gracemont runs efficiently at 3.9 ghz. 10ESF is said to be 15% more efficient than 10SF.

However, I have always maintained that Alder Lake is about die area not power.

Sigh, fine then, lets just come back to this when the chips launch eh? We'll see how highly Gracemonts clock then, and we'll be able to judge from that then - whether or not we're well within the efficient point for the cores or if they're pushed near the max.

 

[IntelUser2000]

Remember the TPU clock vs. power comparisons of Zen 3 and Tigerlake-H? How Tigerlake-H favors higher end of the frequency curve? It's way different from every other CPU.

For example:

https://images.anandtech.com/doci/15967/N2.png

Gracemont seems to follow a typical pattern but Cove cores follow their own pattern. You cannot extrapolate frequency from the graphs they've shown because Gracemont and Golden Cove curves are very different. Gracemont is designed to be efficient at the low end of the frequency range and Golden Cove is efficient at the high end of the frequency range.

The fact that Zen 3 starts to lose at the higher end of clock and power means they are also using similar methodologies as everyone else.

With the Intel 22nm process, we got absolutely nothing on the desktop chips and even lost frequency. But it resulted in Silvermont being a great gain(indeed this is where they claimed 37% better performance).

Maybe they've shifted things back to the pre-22nm era for Cove cores.

@uzzi38 I expect overall higher performance per clock. Skylake has higher number of pipeline stages and the uop cache hit is not as high as they claim.

Skylake could be better in games? But in overall Gracemont should outperform it.\

Sigh, fine then, lets just come back to this when the chips launch eh? We'll see how highly Gracemonts clock then, and we'll be able to judge from that then - whether or not we're well within the efficient point for the cores or if they're pushed near the max.

My rough guess:
-3.3GHz Tremont
-3.6GHz process
-3.9GHz unrestrained power/voltage

 

[eek2121]

Remember the TPU clock vs. power comparisons of Zen 3 and Tigerlake-H? How Tigerlake-H favors higher end of the frequency curve? It's way different from every other CPU.

For example:

https://images.anandtech.com/doci/15967/N2.png

Gracemont seems to follow a typical pattern but Cove cores follow their own pattern. You cannot extrapolate frequency from the graphs they've shown because Gracemont and Golden Cove curves are very different. Gracemont is designed to be efficient at the low end of the frequency range and Golden Cove is efficient at the high end of the frequency range.

The fact that Zen 3 starts to lose at the higher end of clock and power means they are also using similar methodologies as everyone else.

With the Intel 22nm process, we got absolutely nothing on the desktop chips and even lost frequency. But it resulted in Silvermont being a great gain(indeed this is where they claimed 37% better performance).

Maybe they've shifted things back to the pre-22nm era for Cove cores.

@uzzi38 I expect overall higher performance per clock. Skylake has higher number of pipeline stages and the uop cache hit is not as high as they claim.

Skylake could be better in games? But in overall Gracemont should outperform it.\



My rough guess:
-3.3GHz Tremont
-3.6GHz process
-3.9GHz unrestrained power/voltage

I was about to post something similar. At 45W, TGL-H drops to 2.7 ghz under full load per Anandtech. Golden Cove is bigger and much more power hungry…

 

[Hulk]

As Dullard suggest I adjusted the x and y axis so the Intel 40% is achieved. If a max frequency was obtained could these power curves be converted to frequency vs performance? I mean if enough data points were recorded?

 

[eek2121]

As Dullard suggest I adjusted the x and y axis so the Intel 40% is achieved. If a max frequency was obtained could these power curves be converted to frequency vs performance? I mean if enough data points were recorded?

The graphs are marketing graphs and are meaningless. Surely you understand this.

 

[Hulk]

The graphs are marketing graphs and are meaningless. Surely you understand this.

Hey take it easy. Surely you understand not everyone is as smart as you.

 

[eek2121]

Hey take it easy. Surely you understand not everyone is as smart as you.

I mean, come on, it is in the fine print…

(EDIT: yes I sense your sarcasm, I am just calling you out on it)

 

[Hulk]

I mean, come on, it is in the fine print…

(EDIT: yes I sense your sarcasm, I am just calling you out on it)

I'm using your language but the intent was serious. I admit you know more about this stuff than I do. I'm going to write stupid stuff now and then.

 

[Hulk]

I mean, come on, it is in the fine print…

(EDIT: yes I sense your sarcasm, I am just calling you out on it)

I'm using your language but the intent was serious. I admit you know more about this stuff than I do. I'm going to write stupid stuff now and then as I learn.

 

[RTX2080]

Geekbench in what reliability nowaday?

Intel Core i7-12700 non-K Alder Lake CPU is almost as fast as Ryzen 7 5800X in leaked Geekbench score - VideoCardz.com

Intel Core i7-12700 non-K with 8 cores and 16 threads? A new Alder Lake-S processor had just made its first appearance on the Geekbench website. A non-K series CPU called i7-12700 has just been tested in the Geekbench CPU test. The software had trouble reading out the information on the...

videocardz.com

The CPU scored 1595 points in the single-core benchmark and 10170 points in the multi-core, which puts it at the very end of the single-core benchmark list compared to the latest Intel 11th Gen Core series (Rocket Lake) and AMD Ryzen 5000 Series (Zen3).

 

[uzzi38]

Geekbench in what reliability nowaday?

Intel Core i7-12700 non-K Alder Lake CPU is almost as fast as Ryzen 7 5800X in leaked Geekbench score - VideoCardz.com

Intel Core i7-12700 non-K with 8 cores and 16 threads? A new Alder Lake-S processor had just made its first appearance on the Geekbench website. A non-K series CPU called i7-12700 has just been tested in the Geekbench CPU test. The software had trouble reading out the information on the...

videocardz.com

About as reliable as a benchmark can be when we have absolutely no clue about maximum boost clocks nor power consumption.

AKA practically useless to us at this early stage.

 

[coercitiv]

I'm using your language but the intent was serious.

Let me rephrase what @eek2121 tried to convey: these are not graphs, they are drawings. Rebuilding the axes is just as useful as applying lens correction on a painting.

Your intent was commendable, but there's nothing to salvage there.

 

[IntelUser2000]

Geekbench in what reliability nowaday?

User-submitted benchmarks like Geekbench are notoriously unreliable. You can have seemingly exact setups but have performance that vary by greater than 50%.

You can make it reliable but that's only doable after dozens of submissions and after release when you can compare the scores with already known information about the part.

Behavior of Gracemont is going to be steeper. So the lower you go in the frequency range, the lead over Skylake and Golden Cove will increase.

At the high end of the frequency, the Cove cores will start to become better.

It's a trade-off really. Even within the Cove lineup, the U chips use more power at the high frequency than H chips but with fraction of the leakage. Yes, crank it up high and at certain point the "U" chips use more power than H, even at the same voltage!

With the hybrid setup they're trying to have the cake and eat it too. Golden Cove is relatively efficient at the high frequencies. But when the workload is lighter,* Gracemont is better, and in burst scenarios the leakage power will be lower which will improve battery life.

As for why they couldn't have gone for 8+16, probably because at the time they decided on Alderlake, they were unsure of having 10nm everywhere, so they focused on mobile chips. Also the die size of Alderlake at 8+8 is quite large. 8+16 will be even larger.

3.9GHz isn't low nor an efficient range at all. Honestly I think even this is too high. I hope the Jasper Lake successor is at 3.3GHz because it'll be plenty fast, and be a power miser.

*Frequency running an application isn't always binary. For example even back with Ivy Bridge's HD 4000 graphics, in Warcraft III, the iGPU runs far below full frequency and thus save power. You'd think it runs at the full 1200MHz, but no it doesn't. On my XPS 12 it ran at about 500MHz. And the game runs no problem.

 

[JoeRambo]

Some real bad sub scores right there for 12700. Looks like it is held back by memory subsystem, without knowing exact setup it's hard to say, but probably DDR5 4400CL40ish.
Core seems to be very potent tho, some subtests advance almost 50% vs Skylake.

 

[moinmoin]

This is the raw data from that Geekbench results: https://browser.geekbench.com/v5/cpu/9487530.gb5
Uses Windows 10 Pro (so no optimizations).
Stock Processor Frequency: 2.10 GHz
Processor Minimum Frequency: 798 MHz
Processor Maximum Frequency: 4789 MHz
The frequencies shown for the benchmark look odd to me, but I'm too lazy to compare those with other raw results so can't tell either way.

 

[DrMrLordX]

It seems logical, so that's why the AVX512 is disabled on Alder Lake.

Not trying to nitpick, but I believe that has been pointed out before.

 

[dullard]

Old, somewhat complex power calculation from former member idontcare. The dominate term is Vcc (4th power). Of course, increases in voltage increase frequency. So knowing frequency = f(Vcc) would be a must.

View attachment 49269

The dominant term is the Pdynamic: D * frequency * Vcc^2. You are correct that we need to know how frequency is a function of Vcc. In most cases, though, it is acceptable to assume that frequncy is proportional to Vcc. Thus the dominant term becomes E * Vcc^3.

I'm not sure where you got the 4th power from.

 

[dullard]

Let me rephrase what @eek2121 tried to convey: these are not graphs, they are drawings. Rebuilding the axes is just as useful as applying lens correction on a painting.

They are actual graphs (note: of estimated performance which is different from final production performance) with the axis removed.

 

[Abwx]

The dominant term is the Pdynamic: D * frequency * Vcc^2. You are correct that we need to know how frequency is a function of Vcc. In most cases, though, it is acceptable to assume that frequncy is proportional to Vcc. Thus the dominant term becomes E * Vcc^3.

I'm not sure where you got the 4th power from.

Current, and thus switching frequency, of a mosfet increase as a square of voltage, wich mean that voltage has to be increased by 2^0.5 to have doubled frequency.
Of course that s within the favourable segment of the curve, as current increase in the mosfets there s second order effects that come to play, hence at the higher frequencies voltage has to be increased more than said ratio, leading to a slowly increasing voltage/frequency slope.

Some real bad sub scores right there for 12700. Looks like it is held back by memory subsystem, without knowing exact setup it's hard to say, but probably DDR5 4400CL40ish.
Core seems to be very potent tho, some subtests advance almost 50% vs Skylake.

Basically only the single core AES perf is doubtfull, although the MT AES score seems more or less in line.

Micro-Star International Co., Ltd. MS-7D09 vs Default string Default string - Geekbench

browser.geekbench.com

 

[JoeRambo]

Basically only the single core AES perf is doubtfull, although the MT AES score seems more or less in line.

Text Compression and Navigation subscores indicate real bad memory latency. Takes some horrible memory detuning to achieve scores as bad on Skylake.

 

[Abwx]

.

Text Compression and Navigation subscores indicate real bad memory latency. Takes some horrible memory detuning to achieve scores as bad on Skylake.

On ST Cypress Cove has 10% frequency advantage here, and likely 15-20% on MT since the 11700K run at 4.8-5GHz all cores.
Also no AVX512 for ADL.

 

[JoeRambo]

On ST Cypress Cove has 10% frequency advantage here, and likely 15-20% on MT since the 11700K run at 4.8-5GHz all cores.

I am comparing my own Skylake with fixed 5.1 clock.

Default string Default string vs Micro-Star International Co., Ltd. MS-7C71 - Geekbench

browser.geekbench.com

Scoring so bad in Navigation and Text Compression means any workload that hits memory is badly impacted.

Btw i am not claiming this is due to pre-release or whatever impact. Early DDR5 could be this bad easily. Even 4800CL40 or so is disaster levels of memory latency.

 

[eek2121]

.


On ST Cypress Cove has 10% frequency advantage here, and likely 15-20% on MT since the 11700K run at 4.8-5GHz all cores.
Also no AVX512 for ADL.

Aren’t the non-K variants locked to 65W?

I also find the L2 cache setup to be curious…

EDIT: could this chip possibly have 4 big cores and 8 small???

 

[Abwx]

I am comparing my own Skylake with fixed 5.1 clock.

Default string Default string vs Micro-Star International Co., Ltd. MS-7C71 - Geekbench

browser.geekbench.com

Scoring so bad in Navigation and Text Compression means any workload that hits memory is badly impacted.

Btw i am not claiming this is due to pre-release or whatever impact. Early DDR5 could be this bad easily. Even 4800CL40 or so is disaster levels of memory latency.

Including those two (and a two others subscores that perform not well) ADL is still like 17% faster than your CML clock/clock, and close to 20% faster than the original SKL since IPC has been slightly improved by CML.

Aren’t the non-K variants locked to 65W?

I also find the L2 cache setup to be curious…

EDIT: could this chip possibly have 4 big cores and 8 small???

And still perform better at 4.2GHz than a 5.1GHz 10C CML..?.