AMD P-State EPP Patches Spun An 8th Time For Helping Out Linux Performance & Efficiency

[mdedetrich]

I'm still babbeling about this because I think your explanations are wrong. I don't care about any temperature difference, the newer CPUs are all in the 90 degree range anyway.

What no how? Sustained loads are particularily bad for small passive designs, that's the reason why the perf-cores are throttling to half their frequency under longer loads. Short bursts on the other hand get easily absorbed by the bigger heat capacity of the larger metal. Fan designs can handle much longer sustained loads by sacrificing noise.

Technically it actually depends. Passive cooling fundamentally works by storing thermal energy in a body of mass that can be cooled from the outside (i.e. the alumium body on a laptop). If the amount of thermal energy being stored is enough for the body to cool without the CPU going over a critical temperature (lets say 90) then passive cooling is better for sustained loads. Otherwise you are correct

A typical x86 15W CPU also gets away with only running the fans on long heavy loads, that's nothing special only Apple does. Most Laptop designers just seem to prefer lower temps or slightly higher clocks and therefore run the fans earlier.

Of course that isn't special, the special part with Apple laptops that even on heavy loads it still keeps below 90 degrees even without throttling unless you do compiling for like half an hour. Thats the critical part, the M1 is so efficient that at least in almost all cases it doesn't even need a fan to bring temps down. If you equate for performance there isn't an x86 laptop CPU out there that can do this, which is why almost every x86 laptop has fans (even thin and lights which the mac Air does not).

If you want some indication, right now my mac cores are ~35 degrees without doing anything (fans are also off). Compiling a program while also running benchmarks at the same time (taking around 10 minutes to make sure it not throttled while also connected to battery) brings it up to 60 degrees, that is no way close to 90 degrees. If I run it in a loop so its going for half an hour or more then it will start getting to ~90 degrees but the fans are still not turning on because it doesn't need to. You can verify this if you manually control the fan (with an app like TG Pro), I can manually set the fans to full throttle but it will only move from 92 degrees to 88 which basically means the fan wasn't even needed.

Heck even up until recently x86 laptops (particularly intel) got away with good performance by just bursting for an arbitrary amount of time until it got so hot that the laptop couldn't handle it, expecting fants to kick in briefly before that.

[Anux]

In any case specific point you are pushing is making a mountain out of a molehill (we are probably talking about a couple of degrees in difference between the designs)

I'm still babbeling about this because I think your explanations are wrong. I don't care about any temperature difference, the newer CPUs are all in the 90 degree range anyway.

passive cooling solution with a metalplate is slightly better when it comes to dealing with long sustained loads but is worse when it comes to cooling bursty loads

What no how? Sustained loads are particularily bad for small passive designs, that's the reason why the perf-cores are throttling to half their frequency under longer loads. Short bursts on the other hand get easily absorbed by the bigger heat capacity of the larger metal. Fan designs can handle much longer sustained loads by sacrificing noise.

A typical x86 15W CPU also gets away with only running the fans on long heavy loads, that's nothing special only Apple does. Most Laptop designers just seem to prefer lower temps or slightly higher clocks and therefore run the fans earlier.

[mdedetrich]

I didn't talk about the SOC size (AMD and Intel also use SOCs), the metalplate to get the heat away from the SOC is what I'm talkin about, nothing specific to die size. (See the photo above, the thing in the left hand)

I know you weren't talking about it but the reason why I brought it up is because its relevant. In any case specific point you are pushing is making a mountain out of a molehill (we are probably talking about a couple of degrees in difference between the designs) and its also a tradeoff which you are conveniently ignoring, i.e. passive cooling solution with a metalplate is slightly better when it comes to dealing with long sustained loads but is worse when it comes to cooling bursty loads and the most common workload for laptops in particular is bursty (people don't tend to use laptops as servers or benchmarking machines). The fact that the Mac F1 doesn't even need to use fans to deal with the extreme spike in heat that often occurs in x86 laptops when dealing with bursty loads is already indicative.

[Anux]

Yeah but that it understandable due to the sheer size of the die, its an SoC and such a design isn't specific to passive cooling.

I didn't talk about the SOC size (AMD and Intel also use SOCs), the metalplate to get the heat away from the SOC is what I'm talkin about, nothing specific to die size. (See the photo above, the thing in the left hand)

[mdedetrich]

Yes they do: 
The first one is the passive cold plate from the M1 the other is a random active cooler. Notice the massive difference in cooling surface.

Yeah but that it understandable due to the sheer size of the die, its an SoC and such a design isn't specific to passive cooling. If you have such a big die, you still need to cover it completely.

I'm just saying one is easier to be cooled passive than the other.

Possibly but all things considered the difference is going to be negligible and hence passive cooling is not going to be the main determining factor here.

[Anux]

Maybe it was a miscommunication, to summarize my point is that the M1's don't have a special chassis specifically designed for passive cooling

Yes they do: image_2231.jpg
The first one is the passive cold plate from the M1 the other is a random active cooler. Notice the massive difference in cooling surface.

I'm just saying one is easier to be cooled passive than the other.

[mdedetrich]

That's what I just said?

Maybe it was a miscommunication, to summarize my point is that the M1's don't have a special chassis specifically designed for passive cooling so you can't handwave away the efficiency of M1 chips due to the chassis.

Yes hopefully but if you take a look at their history and how they shipped generations of defect by design keyboards while blaming the user ...

I am not advocating for Apple in general, they have definitely fuked in the past with the butterfly keyboard and the touchbar keyboard. My point is with the new macbook and the new M1 chips they have really hit the nail on the head and its like the first time over a decade that I have recommended the macbook.

I am not a company fanboy, however if a company releases a good product then its a good product. In my case the performance difference compared to my thinkpad is insane, I am talking about code that compiles 7-10x faster.

[Anux]

So depending on the chip, they are rated to go as high as 90-100 degrees so this usually isn't a problem. Heck, AMD's newest 7000x generation CPU's will deliberately pull as much power as possible until they hit 90 degrees and with the M1's they are by default tuned to get away with as much passive cooling as possible and then turn on fans which means the fans will typically only kick in once the chip starts roughly hitting 90. It is also possible to tune this if you don't like the behavior.

That's what I just said?

The video in the thread you linked is also pretty old, they may have already fixed this problematic behavior (I only got m1 M1 pro like half a year ago)

Yes hopefully but if you take a look at their history and how they shipped generations of defect by design keyboards while blaming the user ...

[mdedetrich]

Maybe you're a lucky one with better thermal paste application? It seems to be the normal behavior: https://forums.macrumors.com/threads...1-max.2320100/

I don't think those temps are really a problem, Intel and AMD also go that high. More temp difference means better cooling.

So depending on the chip, they are rated to go as high as 90-100 degrees so this usually isn't a problem. Heck, AMD's newest 7000x generation CPU's will deliberately pull as much power as possible until they hit 90 degrees and with the M1's they are by default tuned to get away with as much passive cooling as possible and then turn on fans which means the fans will typically only kick in once the chip starts roughly hitting 90. It is also possible to tune this if you don't like the behavior.

Its also important to distinguish between one off quality issues (which every commodity laptop manufacturer has) compared to persistent problems. I have a 14" M1 Pro and don't have these issues at all, and I am not what is considered an average Joe user.

The video in the thread you linked is also pretty old, they may have already fixed this problematic behavior (I only got m1 M1 pro like half a year ago)

[Anux]

Yeah im gonna call BS on this, I regularly stress all cores and I have monitor that shows temps. This doesn't happen with the newest ARM based macs.

Maybe you're a lucky one with better thermal paste application? It seems to be the normal behavior: https://forums.macrumors.com/threads...1-max.2320100/

I don't think those temps are really a problem, Intel and AMD also go that high. More temp difference means better cooling.