Question Anandtech.com article on 13900k and 7950x power scaling

[adamge]

A Lighter Touch: Exploring CPU Power Scaling On Core i9-13900K and Ryzen 9 7950X

www.anandtech.com

It's an interesting article, but IMO severely crippled by the reliance on "configured" power limit rather than measuring power usage. It's even more frustrating when one of the datapoints is measured power usage, so the author has access to this data.

In this scenario where configured power limits are basically hand waving guidelines, measured power usage (specifically, average power usage by the CPU during the benchmark run) is the only useful, meaningful, and concrete way to analyze the performance results relative to the power usage.

 

[Khanan]

I find this rather surprising as I see package power usage increasing slightly the higher the temperature gets with the 13900K. And with my MSI X570 (5900X) the hotter the VRMs got the higher system power usage increased under the same load (again within reason).

Yes higher temperatures decrease efficiency. Slightly, not by much.

 

[DrMrLordX]

Those temperatures were achieved by a high-end water cooler.
Laptops aren't cooled by water, except a few special cases.

We aren't even talking about mobile CPUs. We're talking about CPUs that are known to take up to 230W (7950X) and 250-350W (13900k) out of the box.

 

[TESKATLIPOKA]

We aren't even talking about mobile CPUs. We're talking about CPUs that are known to take up to 230W (7950X) and 250-350W (13900k) out of the box.

I would love to talk about or compare mobile CPUs, but for that we need reviews first and not just a few.
What do you expect will be different in Dragon Range or mobile Raptor Lake when they are the same CPUs as those used in desktop?
Better binned chips, lower power limits and probably target temperature.
How much difference can this actually make compared to what we saw with desktop parts limited to low power draw?

 

[Timur Born]

Can you provide an overview of your BIOS settings? Obviously you have it tweaked nicely.
Also does your mobo provide the SP rating of your CPU?
Finally what is your PL1/PL2 setting and what is your CB R23 score at that setting?

I have a "hot" core 5 so I might be able to do better if I cap that one at a lower frequency.

My core 5 also is the hottest core without question. Core 0 is the coolest, but not the most stable one. Currently I am using PL1=PL2=253W using 240-245W to reach 40-41k CB23 scores. At stock the same score need 290W. Since this still leaves headroom for per core OCs and CB23 seemingly being stable down to at least 232W I may lower the power limits and voltage further. But that also needs additional per core/clusters stability testing then, not just single and all-core.

Instead of a SP rating Gigabyte offers a "Biscuits" rating with a maximum of 100 (as far as I could find). At stock my CPU get 90 points, undervolted it just reads "out of range".

My undervoltage currently uses lowest AC/DC loadline (default is 2nd lowest), 3rd lowest CPU LLC (default is lowest) and -0.084v adaptive offset. As a result my CPU Vcore is 1.3v in BIOS, 1.2v after droop with P95 SFFT AVX2 load and about 1.25v for CB23. I read voltage from the 2nd Vcore sensor, which is slightly lower than the 1st and likely at or in the CPU, but my board offers no VR Vout sensor.

 

[DrMrLordX]

What do you expect will be different in Dragon Range or mobile Raptor Lake when they are the same CPUs as those used in desktop?

For Raptor Lake? Good luck getting a full implementation in mobile form-factor. Dragon Range remains to be seen. Phoenix is the main event (see CES), where AMD has a monolithic chip that is extremely efficient and has outstanding battery life. Raptor Lake does not stand a chance.

That aside, it appears that people are struggling to prove that Raptor Lake in general is somehow efficient, when it is not.

 

[Hulk]

Raptor Lake does not stand a chance.

Alder Lake didn't stand a chance against the last generation of AMD mobile parts, yet it is everywhere. I'm not disagreeing you with BTW. I just hope we see a lot more AMD powered mobile offerings and that they are widely available at competitive price points.

At 125W or so Raptor Lake is reasonably efficient. Not as efficient as Zen 4 of course. It doesn't get horribly inefficient until pushed pretty far up the v/f curve but it does pretty well up to about 4.5GHz. But besides Zen 4 Raptor is arguably the next most efficient x86 architecture.

 

[Carfax83]

Whoever argues in favor of the 13900K Being more efficient than the 7950X is delusional. No amount of tweaking will make the 13900K be more efficient than a similar tweaked 7950X

Well I guess der8auer is delusional then, because he tested power efficiency and performance per watt during Raptor Lake's launch and he found that Raptor Lake is generally more efficient in gaming workloads, both at stock and with a 90w power limit enforced.

 

[Markfw]

Well I guess der8auer is delusional then, because he tested power efficiency and performance per watt during Raptor Lake's launch and he found that Raptor Lake is generally more efficient in gaming workloads, both at stock and with a 90w power limit enforced.

So one youtube about ONE case (gaming) and all of a sudden that makes them more efficient overall ??

Yes, you are delusional.

 

[IEC]

That's at odds with what TPU found where the 7950X is considerably more efficient for gaming. Though with how good the 7700X is one can make the argument that the 7700X is the far superior part for gaming only:

 

[Thunder 57]

Well I guess der8auer is delusional then, because he tested power efficiency and performance per watt during Raptor Lake's launch and he found that Raptor Lake is generally more efficient in gaming workloads, both at stock and with a 90w power limit enforced.

Nice appeal to authority there. I don't give a crap who der8auer is.

 

[Khanan]

Nice appeal to authority there. I don't give a crap who der8auer is.

Typical inflated ego YouTuber, with clickbait videos to boot.

 

[Carfax83]

So one youtube about ONE case (gaming) and all of a sudden that makes them more efficient overall ??

Yes, you are delusional.

I posted another from a YouTube creator and he found the same thing so it's not just this one video. But even if I posted 10 sources we all know you still wouldn't be convinced so why bother with the act?

 

[Khanan]

I posted another from a YouTube creator and he found the same thing so it's not just this one video. But even if I posted 10 sources we all know you still wouldn't be convinced so why bother with the act?

Do YouTubers suddenly make a better job than real tech journalists? Just a question. I’m watching YouTube regularly btw, GN, LTT, HUB, nice infotainment but not on the same level as websites or magazines I used to read as a kid. The written medium can not be surpassed

 

[Carfax83]

That's at odds with what TPU found where the 7950X is considerably more efficient for gaming. Though with how good the 7700X is one can make the argument that the 7700X is the far superior part for gaming only:

TPU tested using Windows 11 21H2, which is astonishing to me. Once I found that out, I didn't look at their results as favorably as before.

 

[Khanan]

TPU tested using Windows 11 21H2, which is astonishing to me. Once I found that out, I didn't look at their results as favorably as before.

Isn’t windows 11 helping Raptor Lake? It should. It certainly doesn’t help Ryzen.

 

[Thunder 57]

TPU tested using Windows 11 21H2, which is astonishing to me. Once I found that out, I didn't look at their results as favorably as before.

That shouldn't be too shocking. The people who do these benchmarks like to keep the same image around for some time that way results are reproduceable and can be common to all hardware tested on them. You can't really expect everyone to be benchmarking using the latest Windows release, can you? Not to mention many people do not want to be those early adopters.

 

[Markfw]

I posted another from a YouTube creator and he found the same thing so it's not just this one video. But even if I posted 10 sources we all know you still wouldn't be convinced so why bother with the act?

So I agree with the majority here, but I am the one that can't be convinced ? Pretty lame insult.

 

[Carfax83]

Nice appeal to authority there. I don't give a crap who der8auer is.

That's because you're old school. Most review sites have YouTube channels these days (and the ones that don't are likely floundering), and some of these YouTube channels are better than the older review sites like Anandtech.

I shouldn't need to say that Anandtech has been going downhill for a long time now, and YouTube channels like Linus Tech Tips, Gamers Nexus, Hardware Unboxed etcetera are blowing up.

der8auer is a great technical source for people that are into tweaking and overclocking and hardware modifications and he's very competent.

 

[Carfax83]

Isn’t windows 11 helping Raptor Lake? It should. It certainly doesn’t help Ryzen.

Windows 11 is a good OS, but my point was that TPU is using a much older version of Windows 11. Windows 11 22H2 launched with a new thread scheduler that works hand in hand with Raptor Lake's new Thread Director.

So TPU using an older version of Windows is obviously not optimal for Raptor Lake compared to using Windows 11 22H2. Even if it makes no difference, it's good review practice to make sure your software is fully updated (including drivers) for benchmarks.

 

[Carfax83]

That shouldn't be too shocking. The people who do these benchmarks like to keep the same image around for some time that way results are reproduceable and can be common to all hardware tested on them. You can't really expect everyone to be benchmarking using the latest Windows release, can you? Not to mention many people do not want to be those early adopters.

Other review outlets made sure their software was up to date for new hardware reviews. If Wizzard over at TPU can't do it all himself, he needs to hire assistants. Windows 11 22H2 released one month before Raptor Lake launched, so that's enough time to update the software and run a bunch of benchmarks on older hardware in preparation for 13th gen release.

 

[Timur Born]

Yes higher temperatures decrease efficiency. Slightly, not by much.

Up to 4% higher peak power draw (230W > 240W) for CB23 40k after about 7 minutes / 10°C increase (75°C > 85°C).

 

[TESKATLIPOKA]

For Raptor Lake? Good luck getting a full implementation in mobile form-factor. Dragon Range remains to be seen. Phoenix is the main event (see CES), where AMD has a monolithic chip that is extremely efficient and has outstanding battery life. Raptor Lake does not stand a chance.

That aside, it appears that people are struggling to prove that Raptor Lake in general is somehow efficient, when it is not.

Not sure what you mean about full implementation. You mean the full chip or TDP?
8P16E32T Laptop models exist and will be available, but TDP will be basically what can be seen with Alder Lake or maybe a bit higher.
Dragon Range will be in my opinion ~105W(142W PPT) so 145-150W. Don't see much reason going any higher, and will be faster than Raptor.
I wasn't talking about Phoenix. It will be a great efficient chip, but I was talking about performance and 8C16T Phoenix can't win against 16C32T Dragon Range, unless you set both of them to 25-30W.

Raptor Lake is more efficient than Alder Lake thanks to more e-cores, but It's not as efficient as Zen4.
It can't be when It's using a much worse process and AMD has a great architecture. When Intel finally moves to a better process, we can expect some nice gain in efficiency, but until then they will have to use more E-core to try to compensate.

 

[TESKATLIPOKA]

I find this rather surprising as I see package power usage increasing slightly the higher the temperature gets with the 13900K. And with my MSI X570 (5900X) the hotter the VRMs got the higher system power usage increased under the same load (again within reason).

I was talking about Zen4. That is its default behaviour. Link
Increase clocks and voltage, which naturally increase power draw, until you don't go over a specific temperature. Something like power limit but for temperature.
Read the link.

 

[DrMrLordX]

Not sure what you mean about full implementation. You mean the full chip or TDP?

The full chip. As in, something using Raptor Cove with 2MB l2 cache per core etc.

 

[TESKATLIPOKA]

The full chip. As in, something using Raptor Cove with 2MB l2 cache per core etc.

L3 is full 36MB, but they don't mention the amount of L2 on their webpage, only for desktop part.
On the other hand, the same situation is with Alder Lake mobile vs Alder Lake desktop, L2 is mentioned only for desktop part, but both mention L3.

It looks like It's not cutdown for Alder Lake, 14MB L2 and 30MB L3. I expect the see the same for Raptor Lake. If they wanted to cut down something, It would be L3.