Question Alder Lake - Official Thread

[Hulk]

With the release of Alder Lake less than a week away and the "Lakes" thread having turned into a nightmare to navigate I thought it might be a good time to start a discussion thread solely for Alder Lake.

 

[Exist50]

Same die on some of the chips, yes. Other chips, no.

All P and H SKUs come from the same 6+8 die.

 

[dullard]

All P and H SKUs come from the same 6+8 die.

Yes, you are the 3rd person to say so (posts #1977, 1983, and 2001), not including myself saying it (post #1976). But the comment was about battery life of thin and light laptops (post #1974). Thin and Light includes both the P and U chips. The U SKUs do not come from the same die. Why am I having trouble here with that statement? Are we now supposed to forget about the U chips? Heck the post I commented on was specifically talking about both "TGL-U and TGL-H35" (post #1974). The best comparison to those chips would be both Alder Lake U in comparison to TGL-U and Alder lake P in comparison to TGL-H35.

Plus, you seem to be completely missing my point. Just because some mobile Alder Lake chips come from the same die as the 12900HK does not make them the same chip. Cores in use and frequencies matter massively for battery life.

 

[uzzi38]

Same die on some of the chips, yes. Other chips, no. But by talking about dies, you are really just trying to distract from your flawed comments above. Do you honestly think core counts and frequencies don't matter to power consumption (and thus battery life)?

What are you talking about? The entire ADL-P lineup uses the ADL-682 die.

 

[uzzi38]

Yes, you are the 3rd person to say so (posts #1977, 1983, and 2001), not including myself saying it (post #1976). But the comment was about battery life of thin and light laptops (post #1974). Thin and Light includes both the P and U chips. The U SKUs do not come from the same die. Why am I having trouble here with that statement? Are we now supposed to forget about the U chips? Heck the post I commented on was specifically talking about both "TGL-U and TGL-H35" (post #1974). The best comparison to those chips would be both Alder Lake U in comparison to TGL-U and Alder lake P in comparison to TGL-H35.

Plus, you seem to be completely missing my point. Just because some mobile Alder Lake chips come from the same die as the 12900HK does not make them the same chip. Cores in use and frequencies matter massively for battery life.

I said it will be used in thin and light laptops. Not ALL thin and light laptops. You're the one that assumed that's what I meant and started sperging about it.

 

[jpiniero]

What are you talking about? The entire ADL-P lineup uses the ADL-682 die.

In theory the i3 P could use the 2+8 die as well.

 

[dullard]

What are you talking about? The entire ADL-P lineup uses the ADL-682 die.

Yes. The entire ADL-P lineup uses the ADL-682 die.
The entire ADL-P lineup uses the ADL-682 die.
The entire ADL-P lineup uses the ADL-682 die.
The entire ADL-P lineup uses the ADL-682 die.
The entire ADL-P lineup uses the ADL-682 die.
The entire ADL-P lineup uses the ADL-682 die.
The entire ADL-P lineup uses the ADL-682 die.
The entire ADL-P lineup uses the ADL-682 die.
Is that enough for you? I agree that it uses the same die.

But it is NOT the same chip. How dense are you?

 

[jpiniero]

But it is NOT the same chip.

Err... physically it is.

 

[dullard]

Err... physically it is.

Oh great news! I'll buy the cheap i3-1220P and since it is the same chip as the $635 12900HK, what a steal I'll get!

This is quite possibly the best overclock in the history of the world! Same chip for a small fraction of the price!

Sadly, since this discussion is about battery life, the i3-1220P will clearly have the exact same battery life as the 12900HK, since they are the same chip, but I suppose I'll deal with that.

I'm very thankful for you educating me. For years, I thought the number of cores being used, the frequency of those cores, the amount of data going from core to core, the base power level, the turbo power level, the quality of the chip, the number of graphics execution units, the frequency of the iGPU, etc. would impact power usage and battery life. I'm so glad that I've now been corrected.

 

[IntelUser2000]

Oh great news! I'll buy the cheap i3-1220P and since it is the same chip as the $635 12900HK, what a steal I'll get!

You are not getting @uzzi38 point here. Yes clock speeds matter but different clocks are the easiest thing to do. Unless you are a computer luddite you can do it yourself as long as you have a tool. And it doesn't take a lot to do voltage adjustments.

If the difference is at binning or using different type of transistors then sure they are no longer the same chips. Like the question I asked whether one is using low leakage transistors that use more voltage or higher leakage transistors that need less voltage.

Also there's another point while Alderlake operate more efficiency at the lower power point, laptops require less power anyways so we'll see how much more efficient it is.

But half the reason why Intel is getting the performance win back is because they are throwing as much electricity as possible, like if this was a desktop?

Well not necessarily. We can see from AT review that both the Ryzens and the Alderlake is set to 85W CPU. Intel already was pretty close to Ryzen H on the laptop space so anything above that would have been a win, unlike in desktops where AMD outright had way more cores. It's as if 5950X was going against 16 core Tigerlake and Alderlake brings that to 12 P cores and 16 E cores.

 

[dullard]

You are not getting @uzzi38 point here. Yes clock speeds matter but different clocks are the easiest thing to do. Unless you are a computer luddite you can do it yourself as long as you have a tool. And it doesn't take a lot to do voltage adjustments.

Clearly not. He says the 1220P and the 12900HK are the exact same chips because they come from the same die. I absolutely do not get that point.

Also there's another point while Alderlake operate more efficiency at the lower power point, laptops require less power anyways so we'll see how much more efficient it is.

The underlined is the point that I am trying to make. They are not at the same power point, and thus the efficiencies are not the same. You cannot honestly claim that (A) the 12900HK at a less power efficient clock speed will perform the same as (B) the other Alder Lake CPUs at more power efficient clock speeds. Frequency matters massively here because that impacts power efficiency. They are not the same chips, they operate at different efficiency optimizations, and thus will have different impacts on battery life. Yes, the same die, but no not the same power efficiency.

 

[IntelUser2000]

The underlined is the point that I am trying to make. They are not at the same power point, and thus the efficiencies are not the same.

It doesn't matter really. Because we have to take into account the competition. If noone else existed sure it might make some sense. But Zen will scale down with frequencies and voltages too. There's also Intel competition with Tigerlake-H.

So will Alderlake end up being relatively better on laptop than they do with the -S chips. Yes, I do believe so, but not as profound especially on the -H parts.

When I asked whether they use the same chip meant whether the frequency/power curve is the same as Alderlake-S. If you have a steeper curve, then you will end up being better as you go lower. If you have a less steeper curve than you will end up being relatively better at the high end, which was the case of Tigerlake-H vs Cezanne. TGL-H did better vs. Cezanne as the power envelopes went up higher. Meaning you couldn't make the argument that AMD could pump up the power to get same performance gain as TGL-H since the Intel chip gained more.

12900HK review at Computerbase :

Intel Core i9-12900HK im Test

Mit Alder Lake-H bietet Intel wieder die schnellste CPU für Notebooks: Der Core i9-12900HK schlägt die gesamte Konkurrenz im Test.

www.computerbase.de

So you can see from the Blender package power graph it's outperforming Ryzen and TGL competitors by 20% despite actually using quite a bit less power. Seems like a success to me. That'll translate to 30%+ like in AT's review when it's set at same power levels.

 

[Abwx]

So you can see from the Blender package power graph it's outperforming Ryzen and TGL competitors by 20% despite actually using quite a bit less power. Seems like a success to me. That'll translate to 30%+ like in AT's review when it's set at same power levels.

If we normalize Cezanne s TDP at the same level then ADL (80W) does about 15% better perf than Cezanne, so a good result albeit with more cores to get this efficency level, i guess that things will be nullified by AMD s RMB.

Intel Core i9-12900HK im Test: Benchmarks und Erfahrungen

Intel Core i9-12900HK im Test: Benchmarks und Erfahrungen / Die Powerlimits und ihre Auswirkungen

www.computerbase.de

The references i m talking about is the Cezanne based XMG Core 17 that is set at 80W although the power curve do show 68W on average while the 12900HK manage to stay at 80W.

The Asus Rog Flow is also set at 80W but it doesnt hold the TDP, it start at 70W and end at 53W when finishing the run.

 

[IntelUser2000]

You can see the power normalizes at 69W for ADL, and still outperforms the best Ryzen by 18% in Cinebench R23.

Also Anand's tests have set both chips at 85W, or at least that's what he measured it.

 

[jpiniero]

Intel Celeron G6900 Benchmarks - Performance Of Intel's $40~60 Alder Lake Processor Review - Phoronix

www.phoronix.com

Review of the Celeron G6900. Note that Phoronix says they paid $69 for it (fitting, huh)

 

[Abwx]

You can see the power normalizes at 69W for ADL, and still outperforms the best Ryzen by 18% in Cinebench R23.

Also Anand's tests have set both chips at 85W, or at least that's what he measured it.

There s two curves for the 12900HK, one wich is for PL1/PL2 at 135/85W and the scond one for 135/65W, the scores are the one of 135/85W setting and the CPU use 80W on average, for some reason they dont display the scores for the 135/65W case where the CPU use about 70W on average, wich is surely the one you are talking about.

Btw, you can click on the squares in the graph legend to select the relevant CPUs.

AT review at 85W is somewhat of a stretch because Cezanne gain little perf at those levels, according to Computerbase there s a mere 6% difference between 67W and 95W.

It would be more relevant to do tests within a 25-45W range wich is the usual range of laptops while here we are more at DT level TDPs.

 

[dullard]

Intel Celeron G6900 Benchmarks - Performance Of Intel's $40~60 Alder Lake Processor Review - Phoronix

www.phoronix.com

Review of the Celeron G6900. Note that Phoronix says they paid $69 for it (fitting, huh)

Thanks for posting that. I'm certainly not one to buy a Celeron, and that still doesn't encourage me to buy a Celeron. It is fine for the lightest user group. But the G7400 should be much better for not much more than the cost of a good lunch.

But, I'm not sure why you bothered to post it, since it is the same chip as the 12600, 12500, 12400, etc and there are plenty of reviews of them. Apparently it doesn't take much to do voltage adjustments, so no need for reviews of other chips.

 

[jpiniero]

But, I'm not sure why you bothered to post it, since it is the same chip as the 12900K and there are plenty of reviews of the 12900K. Apparently it doesn't take much to do voltage adjustments, so no need for reviews of other chips.

Actually that is a different (6+0) die.

 

[dullard]

Actually that is a different (6+0) die.

Sorry, typed before thinking. I'm still adjusting to the new logic that any CPU from the same die is the same chip. I corrected it above. I'm not sure why you posted the review because it is the same chip as the 12600, 12400, etc.

 

[coercitiv]

From what I understood, the E-Cores are good because they, along with Thread Director, allows the P-Cores to concentrate on their tasks for better performance and system responsiveness.

There's two little lies Intel Marketing told us, or at least selectively gave us information so we could tell the lies to ourselves:

1. E-cores are energy efficient.
2. E-cores improve responsiveness with the help of Thread Director.

The truth is more nuanced:

1. E-cores are area efficient.
2. Thread Director mitigates potential responsiveness problems caused by the presence of E-cores.

To address your specific perceived benefit of E-cores and TD, if a system with 6 cores were to use Thread Director to optimize performance, does it matter what type of cores there are in the mix? As long as the hardware guided scheduling is there, shouldn't Intel enable Thread Director optimization on both a 2+4 configuration and a 6+0 configuration? Free performance is free, right?

Here's the catch, if the answer to the question above is that enabling TD optimization only makes sense in silicon with E-cores enabled, then it becomes obvious that TD's main goal is to mitigate potential scheduling issues, not enhancing overall performance. There may still be situations where there's bonus performance to be had, but they're part of a set of trade-offs that hybrid configurations come with.

 

[TheELF]

There's two little lies Intel Marketing told us, or at least selectively gave us information so we could tell the lies to ourselves:

1. E-cores are energy efficient.
2. E-cores improve responsiveness with the help of Thread Director.

The truth is more nuanced:

1. E-cores are area efficient.
2. Thread Director mitigates potential responsiveness problems caused by the presence of E-cores.

To address your specific perceived benefit of E-cores and TD, if a system with 6 cores were to use Thread Director to optimize performance, does it matter what type of cores there are in the mix? As long as the hardware guided scheduling is there, shouldn't Intel enable Thread Director optimization on both a 2+4 configuration and a 6+0 configuration? Free performance is free, right?

Here's the catch, if the answer to the question above is that enabling TD optimization only makes sense in silicon with E-cores enabled, then it becomes obvious that TD's main goal is to mitigate potential scheduling issues, not enhancing overall performance. There may still be situations where there's bonus performance to be had, but they're part of a set of trade-offs that hybrid configurations come with.

Intel sells the ecores in the way shown in these pictures.
Compared to a CPU without ecores it allows your foreground app to keep running at full speed while giving the background app a specified amount of compute that's always available to it, depending on the amount of e-cores.

The way you would keep your foreground app running at full speed on a CPU without e-cores would be to run the foreground app at real-time priority and the background app at idle priority meaning that the background app might never get any CPU cycles.

This is NOT mitigating potential responsiveness problems caused by the presence of E-cores, this is your foreground app getting the use of the full CPU (compared to a CPU without e-cores) meaning there is no slowdowns.

 

[coercitiv]

This is NOT mitigating potential responsiveness problems caused by the presence of E-cores, this is your foreground app getting the use of the full CPU (compared to a CPU without e-cores) meaning there is no slowdowns.

The Gaming + OBS marketing slide is showing just that, TD is mitigating potential problems introduced by foreground app threads ending on E-core clusters. It's amazing how clear the information is and you're still pretending otherwise.

The second slide is downright comical. They are actually claiming the first task is faster while a video export is in the background as opposed to the "serial run" scenario when no significant background task is active. Only marketing departments can come up with this stuff.

As I already mentioned, there are some benefits of TD separating background tasks and fencing them on a number of cores (E-cores in this case), but these benefits come with a cost, they're a trade-off. It has already been explained by both me and @Hulk that we've seen scenarios where performance drops with this approach, since locking background tasks on E-cores results in a net performance loss when P-cores are idle, for example when the user is looking at a webpage while a video export is happening in the background.

 

[TheELF]

The Gaming + OBS marketing slide is showing just that, TD is mitigating potential problems introduced by foreground app threads ending on E-core clusters. It's amazing how clear the information is and you're still pretending otherwise.

The perforated background box represents the e-cores, they don't show any gaming threads at all in there, obs exclusively.

The second slide is downright comical. They are actually claiming the first task is faster while a video export is in the background as opposed to the "serial run" scenario when no significant background task is active. Only marketing departments can come up with this stuff.

It's faster because they are comparing to a previous gen CPU and the e-cores handle the export so there is no slowdown for the foreground task.

since locking background tasks on E-cores results in a net performance loss when P-cores are idle, for example when the user is looking at a webpage while a video export is happening in the background.

Depends on if your webbrowser is running at a low enough priority, if it's tagged as a background task by the developers then yeah it's going to go to the e-cores and is going to fight for resources.

 

[IntelUser2000]

There's two little lies Intel Marketing told us, or at least selectively gave us information so we could tell the lies to ourselves:

1. E-cores are energy efficient.
2. E-cores improve responsiveness with the help of Thread Director.

The architecture of the E core themselves ARE more energy efficient. It's perhaps not showing in Alderlake-S as well, but detailed tests show otherwise:

https://i0.wp.com/chipsandcheese.com/wp-content/uploads/2021/12/image-51.png?ssl=1

You can see from the graph the "core" power starts off higher with Gracemont than with 6600K. So the "core" power includes very high overhead non-core components, as we see the 6600K overlapping and exceeding it at higher core counts, otherwise such graphs would be impossible!

Adding 3 additional cores results in almost exactly 15W for Gracemont, so about 5W per core. With Skylake it increases by 21.41W, so we could initially assume each Skylake cores are 7W.

However, we can see going from 1 to 2, then 2 to 3, then 3 to 4, with Gracemont it adds as expected at 5W. But for Skylake the first additional core adds 8W, the second adds a bit less, and the third adds less again, meaning the actual core power is 8W.

Now this is with Gracemont running at the high end of the frequency range. The V/F curve video showed that Gracemont requires higher voltages than Golden Cove.

Which is actually what happens with their Core chips when comparing -H versus -U. The latter requires more voltage than the former. So you'd think the -U is less efficient. On the contrary, the -U is just operating at a different point. The higher voltage required means the clock ceiling on the -U comes faster than the -H, and power wise it's less worth it to go there. The tradeoff is likely the -U uses fraction of the leakage power. Despite needing higher voltage, the static power goes down significantly reducing overall power.

Another graph of theirs shows interesting results as well: https://i0.wp.com/chipsandcheese.com/wp-content/uploads/2021/12/gmt_power.png?ssl=1

We can see up until 4 cores, the scaling of power for Gracemont is a constant 5W, but from the 5th core, the addition is less, meaning that's when per core clocks start to drop.

With Golden Cove, it scales pretty linearly all the way to the maximum of 8 cores. It's a constant 17-18W per core.

From Anandtech's updated 12900K review: https://www.anandtech.com/show/1704...hybrid-performance-brings-hybrid-complexity/9

the 8 E-cores are able to provide around 52-55% of the performance of 8 P-cores without SMT, and 47-51% of the P-cores with SMT. At first glance this could be argued that the 8P+8E setup can be somewhat similar to a 12P setup in MT performance, however the combined performance of both clusters only raises the MT scores by respectively 25% in the integer suite, and 5% in the FP suite, as we are hitting near package power limits with just 8P2T, and there’s diminishing returns on performance given the shared L3.

1/3rd the core power for 1/2 the performance means 50% better perf/watt.

So there's more that Intel can do to improve performance.
-Improved thread director
-Better balance between P and E cores in MT applications to favor E cores when it's more beneficial to get the full 0.5x benefit, rather than the 0.3x we're getting now per E core cluster.
-Separate voltage domains for P and E cores?
-Better ring bus frequency control so people won't be so inclined to disable the E cores for the higher ring bus

 

[uzzi38]

Take from it what you will. But I think their answer was pretty clear.

 

[Zucker2k]

Take from it what you will. But I think their answer was pretty clear.


View attachment 56601

Great. Just what I thought.