Geekerwan interviews Johny Srouji on A18/M4

[crazy dave]

Both things can be true. I agree very few (maybe none) do what they do. It’s also true to say their methodology is vague and results are all over the place.

It’s just the sad state of reviews.

What makes you think that? Do you think they have more money and resources than other popular YouTube tech channels like GamerNexus?


If their results are popular, it's because no one else does what they do. Even if they don't have early access to Apple hardware, their reviews become popular even if it's the umpteenth review of the same hardware.

Geekerwan may not be as good as Anandtech used to be, but the truth is that apart from Chips and Cheese, no one else is doing what they are doing.

I like Notebookcheck reviews, they publish their methodology and seem to do a pretty thorough job of testing, including power, noise, and thermals.

 

[theorist9]

Why boost only two P cores? M4 is already head and shoulders above the competition in ST speed. Boosting two cores won't yield much difference in benchmark bragging rights.

Just FYI, Intel's PL2 Alder Lake boost uses ~120w more and only provides 10-20% more performance.

I'm sure Apple has a bit more headroom since they're already so low in starting wattage, but I think to get any meaningful benchmark bragging rights, it'll be much closer to 100w than 10w.

Remember that my initial argument wasn't that they should do it for bragging rights, they should do it to improve user experience. It's gotten turned it towards a discussion about bragging rights after your argued a key reason they shouldn't implement it is for marketing reasons (an argument that, as you know, I disagreed with).

So let's return to to user experience: The overwhelming majority of programs remain single-threaded. Further, the machine's "snappiness/responsiveness" is determined by SC speed (along with GPU performance, especially if you are running multiple displays). Improving the performance on a couple of cores would help both of those.

Plus even you're at a point on the power : performance curve where you need a lot more power to get a little more performance, if you only need to boost a couple of cores that would be far less of an issue.

I.e., my question to Apple is effectively: "OK, even if you don't want to boost across the board because of the steepness of the power curve, why not at least do it for a couple of cores?" [As I speculated earlier, I suspect the answer is that they couldn't do so with the previous chip designs; perhaps M4 will be different.]

Finally, if we do consider bragging rights, while the iPad's M4 is (based on GB6 SC from primate.com) 3655/3140 = 16% faster than the i9-13900KS, there will probably be new processors coming out from Intel and/or AMD while M4 is current in the Macs.

 

[crazy dave]

Remember that my initial argument wasn't that they should do it for bragging rights, they should do it to improve user experience. You've turned it towards a discussion about bragging rights by saying the main reason they shouldn't implement it is for marketing reasons (an argument that, as you know, I disagreed with).

So let's return to to user experience: The overwhelming majority of programs remain single-threaded. Further, the machine's "snappiness/responsiveness" is determined by SC speed. Improving the performance on a couple of cores would help both of those.

Plus even you're at a point on the power : performance curve where you need a lot more power to get a little more performance, if you only need to boost a couple of cores that would be far less of an issue.

I.e., my question to Apple is effectively: "OK, even if you don't want to boost across the board because of the steepness of the power curve, why not at least do it for a couple of cores?" [As I speculated earlier, I suspect the answer is that they couldn't do so with the previous chip designs; perhaps M4 will be different.]

Finally, if we do return to bragging rights, while M4 is (based on GB6 SC from primate.com) 3655/3140 = 16% faster than the i9-13900KS, there will probably be new processors coming out from Intel and/or AMD while M4 is current.

Small disagreement: Intel is coming out with Arrow Lake, but AMD just released their latest desktop processors, which are slightly slower overall in ST than the M4 in GB (individual subtest vary obviously and lots of run-to-run variation). AMD might offer a refresh next year, but it likely won't be a new architecture and if Apple maintains its cadence, the M5 should be out by then. Also if past is prologue then the M4 in an actively cooled device should also get a few percent boost (comparing M2 in 11" iPad Air to the Mac mini M2 Pro - same clocks) so it is likely to be about 20% faster than the 13900KS in a Mac. Based on what we've seen so far from Lunar Lake and Arrow Lake leaks, I would suspect that the enthusiast Intel processors will be around that - maybe slightly faster/slower but close. So in terms of bragging rights, I don't think Apple would actually need to do much, maybe nothing, to retain "the crown".

Mostly agreement: I know Geekerwan managed to boost the iPad up to a score of 4000 in GB with liquid nitrogen and so maybe the M4 is already capable of delivering the kinds of boosts you're hoping for but I'm not going to rely on that rather specific set of circumstances for predictions. We'll just have to see. Also, it is possible that the Hidra chip, if it exists, and is desktop oriented (and isn't M5 anyway - we know from the other thread Apple already has seemingly two M5 devices planned next year which could be Airs but alternatively could be Studios), then Apple might boost M4 clocks a little like the M2 generation did for the Max/Ultra. Again, we'll just have to see. I wouldn't be averse to that as long as Apple keeps things sane.

So mostly dangerous?

Why boost only two P cores? M4 is already head and shoulders above the competition in ST speed. Boosting two cores won't yield much difference in benchmark bragging rights.

Just FYI, Intel's PL2 Alder Lake boost uses ~120w more and only provides 10-20% more performance.

I'm sure Apple has a bit more headroom since they're already so low in starting wattage, but I think to get any meaningful benchmark bragging rights, it'll be much closer to 100w than 10w.

I don't think @theorist9 wants Apple to go down that road (though to be fair I'm pretty sure the PL2 super turbo boost is transient - still it's not a good design philosophy in my very non-expert opinion). I think he's talking about just a 6-10% performance boost more along the lines of the top end M2 Max/Ultra. That still uses quite a bit more power, but on a desktop Apple could afford it. Don't get me wrong, I'm hesitant about it as well, Apple's ST power has been creeping up every generation and I wouldn't want them to fall into the MoAR powER trap that x86 did. But a small boost for desktops wouldn't necessarily go amiss.

==========

Though as I've argued before, ad nauseam I'm sure for everyone, boosting GPU clocks on a hypothetical desktop part is a lot more crucial than boosting the CPU for both user experience and bragging rights. When it comes to desktops, raw GPU performance is Apple's weakest element.

 

[theorist9]

Just FYI, Intel's PL2 Alder Lake boost uses ~120w more and only provides 10-20% more performance.

Forgot to ask about this--what's the extra wattage per boosted P core, and the attendant performance boost?

 

[theorist9]

So mostly dangerous?

LOL

I think he's talking about just a 6-10% performance boost more along the lines of the top end M2 Max/Ultra.

When I'm waiting several seconds to several minutes for Mathematica to complete a calculation I wouldn't mind more. But yes, realistically, I wouldn't expect them to offer much more than 5 GHz (11% higher than the current 4.5 GHz) => GB ≈ 4000.

Don't get me wrong, I'm hesitant about it as well, Apple's ST power has been creeping up every generation and I wouldn't want them to fall into the MoAR powER trap that x86 did. But a small boost for desktops wouldn't necessarily go amiss.

...though a way they could offer more than 5 GHz without falling into this trap is to have a selectable high-power mode that enables the user to choose to run the machine hotter temporarily when doing a demanding calculation. [Assuming the chip can handle that.]

Though as I've argued before, ad nauseam I'm sure for everyone, boosting GPU clocks on a hypothetical desktop part is a lot more crucial than boosting the CPU for both user experience and bragging rights. When it comes to desktops, raw GPU performance is Apple's weakest element.

Not for Mathematica and other ST non-graphical programs. But you're right that, more broadly, their main weakness is GPU rather than CPU.

 

[crazy dave]

LOL

When I'm waiting several seconds to several minutes for Mathematica to complete a calculation I wouldn't mind more. But yes, realistically, I wouldn't expect them to offer much more than 5 GHz (11% higher than the current 4.5 GHz) => GB ≈ 4000.


...though a way they could offer more than 5 GHz without falling into this trap is to have a selectable high-power mode that enables the user to choose to run the machine hotter temporarily when doing a demanding calculation. [Assuming the chip can handle that.]


Not for Mathematica and other ST non-graphical programs. But you're right that, more broadly, their main weakness is GPU rather than CPU.

It is interesting that when Geekerwan cooled the M4 with liquid nitrogen, that it could hit 4000 in GB with seemingly* no clock bump - and you have run-to-run variation such that I often notice that the official GB scores in their tables tend to be lower than seemingly the majority of scores that get submitted by users (though that's true for all the chips) - or at least half of them .

*I do wonder what the GHz was when it was cooled down like that? Did it stay at ~4.4GHz? They might've said in the video but I've forgotten. I also can't remember the score before cooling so it may have been an already good silicon score made better. Anyway, maybe the M4 by itself will be enough? We'll see what the November event brings but of course if no Studio's get released and Apple doesn't do a top-end Max/Ultra clock bump we still won't necessarily know if a dedicated desktop part with a different clockspeed is on the horizon or not - i.e. such a thing could still happen later. Should be an interesting year.

 

[senttoschool]

Remember that my initial argument wasn't that they should do it for bragging rights, they should do it to improve user experience. It's gotten turned it towards a discussion about bragging rights after your argued a key reason they shouldn't implement it is for marketing reasons (an argument that, as you know, I disagreed with).

Forgive me. I think I mixed up your arguments with leman's.

So let's return to to user experience: The overwhelming majority of programs remain single-threaded. Further, the machine's "snappiness/responsiveness" is determined by SC speed (along with GPU performance, especially if you are running multiple displays). Improving the performance on a couple of cores would help both of those.

There is barely any noticeable difference in overall user experience/responsiveness/snappiness between the M1 and M4 and the ST speed the the two is quite massive. If you're running something heavier like an app that runs for a few minutes, the sure. I have an M1 iPad and messed around with an M4 iPad for a while. No difference in overall snappiness or responsiveness.

I.e., my question to Apple is effectively: "OK, even if you don't want to boost across the board because of the steepness of the power curve, why not at least do it for a couple of cores?" [As I speculated earlier, I suspect the answer is that they couldn't do so with the previous chip designs; perhaps M4 will be different.]

They do. Pro/Max/Ultra have higher single core boost than the base.

What you're asking for is even more boost. For that, I'll lead you back to my original arguments: simple messaging, consistent performance/perf per watt claims, extra validation and testing for a very small market, boosting 10 more watts won't be noticeable since Intel boosts 125w to get 10-20% more perf.

I think Apple's current boost behavior is fine. Every Mac model does not spin up fans and is cool. MBPs and desktops get a slightly higher boost. I'd hate for Apple to give it a lot more power for no noticeable difference in everyday usage and likely only a few percentage more in benchmarks.

 

[theorist9]

They do. Pro/Max/Ultra have higher single core boost than the base.

What you're asking for is even more boost.

No, the desktops do not have more boost than the laptops, which is what I've been asking for from the beginning.

I think your response is disconnected from what I wrote. I said I want X (more boost for desktops over laptops), and you essentially responded 'they already do Y (more boost for Pro+ processors above base) and what you want is more Y'.

I feel like we're just not communicating here.

I'll lead you back to my original arguments: simple messaging, consistent performance/perf per watt claims, extra validation and testing for a very small market, boosting 10 more watts won't be noticeable since Intel boosts 125w to get 10-20% more perf.

And I'll lead you back to the responses I already made to the importance of consistency. So I think we're at an impasse.

 

[senttoschool]

No, the desktops do not have more boost than the laptops, which is what I've been asking for from the beginning.

They do. Mini is slightly faster than Air.

But I suspect you mean only the Studio And Mac Pro specifically?

 

[theorist9]

They do. Mini is slightly faster than Air.

But I suspect you mean only the Studio And Mac Pro specifically?

According to the info. on Primate's site, the M2 Mini, M2 Pro Mini, and M2 Air all are clocked at 3.5 GHz (there is no M3 Mini). Further, in attempting to contradict me by writing "the Mini is slightly faster than the Air", you're missing the big-picture point of what I've been saying (and it also seems like niggling). So let me be even more explicit about what I'd like to see:

With the possible exception of the 24" iMac, the Mac desktops, from the Mini up, have more thermal capacity than any of Apple's laptops; and none of them need to run on battery. Yet in no case are any of those desktops boosted to offer a max clock speed beyond the fastest seen in the laptops. I think that, in those desktops, a couple of CPU P-cores should be boosted to the maximum that (a) allows stable, reliable operation; and (b) doesn't cause those machines to be noisy (thus you might see less boost with the Mini than the Mac Pro).

Further, for those clients who are willing to trade off some noise for faster operation, there should be an option to select a high-power mode that's merely required to satisfy condition (a). Conversely, for those clients that want to focus on efficient operation, Apple could add an efficiency mode that limited max clock speed back to laptop levels.

There's precedent for both, since Apple does currrently offer High Power and Low Power modes on its MBP's (though the High Power Mode only boosts the fan curve; I don't know what the Low Power Mode does).

 

[senttoschool]

According to the info. on Primate's site, the M2 Mini, M2 Pro Mini, and M2 Air all are clocked at 3.5 GHz (there is no M3 Mini). Further, in attempting to contradict me by writing "the Mini is slightly faster than the Air", you're missing the big-picture point of what I've been saying (and it also seems like niggling). So let me be even more explicit about what I'd like to see:

I think we weren't talking about the same thing. Mini, on average, does benchmark faster than the Air. It's likely thermal headroom related.

you're missing the big-picture point of what I've been saying (and it also seems like niggling).

I'm not missing the big picture. I understood your point after reading your first post.

Yet in no case are any of those desktops boosted to offer a max clock speed beyond the fastest seen in the laptops. I think that, in those desktops, a couple of CPU P-cores should be boosted to the maximum that (a) allows stable, reliable operation; and (b) doesn't cause those machines to be noisy (thus you might see less boost with the Mini than the Mac Pro).

My explanation below stays relevant:

simple messaging, consistent performance/perf per watt claims, extra validation and testing for a very small market, boosting 10 more watts won't be noticeable since Intel boosts 125w to get 10-20% more perf.

One or more of the above is very likely to be the reason Apple doesn't do what you said.

However, my point was that that even if you ask Johny, he wouldn't say anything we don't already know.

 

[altaic]

One or more of the above is very likely to be the reason Apple doesn't do what you said.

I don’t find your arguments compelling. Messaging? Extra testing when Apple already does extensive testing on all of their products? And you’re really comparing relative power consumption between ASi-ASi and Intel-Intel? 🤨

 

[crazy dave]

According to the info. on Primate's site, the M2 Mini, M2 Pro Mini, and M2 Air all are clocked at 3.5 GHz (there is no M3 Mini). Further, in attempting to contradict me by writing "the Mini is slightly faster than the Air", you're missing the big-picture point of what I've been saying (and it also seems like niggling). So let me be even more explicit about what I'd like to see:

With the possible exception of the 24" iMac, the Mac desktops, from the Mini up, have more thermal capacity than any of Apple's laptops; and none of them need to run on battery. Yet in no case are any of those desktops boosted to offer a max clock speed beyond the fastest seen in the laptops. I think that, in those desktops, a couple of CPU P-cores should be boosted to the maximum that (a) allows stable, reliable operation; and (b) doesn't cause those machines to be noisy (thus you might see less boost with the Mini than the Mac Pro).

Further, for those clients who are willing to trade off some noise for faster operation, there should be an option to select a high-power mode that's merely required to satisfy condition (a). Conversely, for those clients that want to focus on efficient operation, Apple could add an efficiency mode that limited max clock speed back to laptop levels.

There's precedent for both, since Apple does currrently offer High Power and Low Power modes on its MBP's (though the High Power Mode only boosts the fan curve; I don't know what the Low Power Mode does).

Apple did sort of do this with the M2 Max/Ultra ... sorta. The top end M2 Max/Ultra, but not the binned variants of each, had a small clock bump but this wasn't based on device, just chip. That said these top end chips were more likely to go into the 16" MBP, Studio, and Pro (though admittedly the 14" was an option) - the most expensive and thermally expansive devices in Apple's lineup. It is not inconceivable that Apple might do such a thing again, but targeted differently, but I suspect, especially given the purported redesign of the Mini, that the Mini (and as you say the 24" iMac) would be left out if they decided to clock bump all desktop chips. It would really just be the Studio/Pro.

Low power mode I believe does throttle the device - I think I tested it myself.

I think we weren't talking about the same thing. Mini, on average, does benchmark faster than the Air. It's likely thermal headroom related.

Yes that is thermal headroom, same as Geekerwan artificially cooling the M4 iPad Pro to get it to run constantly at max. That said Apple did experiment with a small boost clock for select models of the M2 Max and Ultra as I wrote above. They did not so for the M3, but the M3 had other experiments.

I'm not missing the big picture. I understood your point after reading your first post.



My explanation below stays relevant:


One or more of the above is very likely to be the reason Apple doesn't do what you said.

I think Apple has shown a willingness to experiment each generation, including with clock bumps, so while I wouldn't definitely say that Apple will do this, add a small clock bump to Studios/Mac Pros, I wouldn't rule it out either.

However, my point was that that even if you ask Johny, he wouldn't say anything we don't already know.

That's very true and would be true of anyone at Apple. They do not talk about future products and asking would be a waste of time - unless you work for the WSJ/the Information and they're a future Apple lawsuit.

 

[Boil]

Hidra on N3X should give us the desired desktop/workstation class clocks...?

 

[MayaUser]

People, including myself hopes M4 should give us this for the Mac Studio and Mac Pro...Mac mini, since Apple seems to chose to go smaller enclosure i would not take it as a reference ...so maybe M4 Max/Ultra hopefuly will have desktop class clocks...but again the M4 Max will be on the 14 and 16" Mbp...so i dont know, chances are low IMO

 

[senttoschool]

I don’t find your arguments compelling. Messaging? Extra testing when Apple already does extensive testing on all of their products? And you’re really comparing relative power consumption between ASi-ASi and Intel-Intel? 🤨

Why do you think they don’t do it already then?

 

[MayaUser]

i wonder if the M4 Ultra can reach nvidia 4090 dGpu...2x ray tracing based on current system it will be a big deal already for people that are using pro apps and hw rt

 

[Bodhitree]

I really enjoyed the video, it was pitched at a technical level where you could follow it with a basic (software engineers) knowledge of the hardware, and get something out of the explanations in terms of Apple’s methodology and planning.

So sorry that the hardware engineering crowd didnt think there was enough detail, but for me it was fine.

 

[streetfunk]

I really enjoyed the video, ........................, and get something out of the explanations in terms of Apple’s methodology and planning.

+1, same for me. (as a total software and computer HW tech foreigner)
by all circumvention strategy, there was nontheless some interesting infos to be had.

 

[Bodhitree]

It does little for me to know that the A17 had a 9-wide decode, as opposed to some lesser number for the A16, its of passing interest at best.

It is much more interesting to me that Apple uses a data-driven internal modelling system to tune the focus on power efficiency, performance, etc during the chip design process.