iM 4,1-14,4 Turbo Bust Mode: why at least one Mac doesn’t ramp up to its CPU’s designed turbo clock speed

[B S Magnet]

This is one of those “this question half-belongs on the Early Intel Macs forum and half-belongs to a non-existent ‘all Intel Macs’ forum (and maybe a foot in the “iMac” forum)”:

Earlier, I posted on the “What have you done with an Early Intel (Mac) recently” the use of the Intel Power Gadget monitoring graphs and the Turbo Boost switcher utility on my Sandy Bridge MBP I picked up in March.

One thing in my mind I’d expected, intuiting (semi-incorrectly) from the long-term behaviour of my late 2013 Haswell A1418 iMac running a quad-core i5-4570S, was in how much higher the Turbo Boost mode has been on my A1278 MBP (and what makes it a toasty boi when pushing it hard). The latter is a dual-core i7 (i7-2640M rated nominally at 2.8 GHz); given the extended behaviour of that iMac I’ve had since 2020, I expected the Turbo Boost for the i7-2640M to be 3.1GHz. It’s not. It’s 3.5GHz, which explains the toasty part.

Why did I think it would be 3.1GHz? This is where it gets mystifying (to me, at least). That 3.1 figure would be 300MHz higher than its nominal clock speed.

I extrapolated that from months of letting the aforementioned iMac run a bunch of Handbrake encodes at full hilt. The iStat menubar info on the CPU clock speed would move up from 2.9GHz resting to 3.2 GHz Turbo Boost max (a 300MHz difference), bringing all four cores to nearly 100 per cent (again, it’s encoding, so that was completely expected). Which I thought 3.2GHz for Turbo Boost was as-designed, even though I know my eyes had looked over that CPU’s company specs before.

Contrary to what my mind assumed to be its designed limit, the i5-4570S is a CPU designed not for 3.2GHz boost, but a 3.6GHz boost! (Also worth keeping in mind: I have Macs Fan Controller set to run max fan speed whilst encoding, which can bring core temps to the 80–85°C range, but still well below the threshold of the CPU throttling itself to limit overheating.

So… uh… what is going on here?

Is this a firmware-based governor limiting its max turbo boost to 3.2GHz rather than 3.6GHz, or is there something else at work here? (Just a quick note: there aren’t Energy Saver settings to toggle processor performance on this model line.) I note in Intel Power Gadget that if I put a heavy load on the system (i.e., a test encode), the “Core REQ(uested)” line wants more than 3.2GHz, but isn’t being given more than 3.2GHz.

Does anyone know the ins and outs of the Core iX series CPUs to explain this discrepancy like I’m five? Whatever the case, I’m kind of meh at the realization that this CPU is meant to Turbo Boost higher than what I’ve ever seen on it.

 

[GMShadow]

Intel CPUs allow for configurable limits, in both directions (usually referred to as cTDP up or cTDP down).

Are you sure your iMac has a 4570S, not a 4570R? The 4570R does turbo to 3.2 only.

 

[B S Magnet]

Intel CPUs allow for configurable limits, in both directions (usually referred to as cTDP up or cTDP down).

Are you sure your iMac has a 4570S, not a 4570R? The 4570R does turbo to 3.2 only.

Well, per this (explicitly) and this (implicitly), there’s an i5-4570S stuffed in there. So I honestly don’t know if those were all inaccurate. Below is System Info on the iMac in question:

Curiously, the “Products Formerly Haswell” page doesn’t list the 4570R. It does list the 4570*, 4570S, and 4570T — all with the same 2.9/3.6 specs. (There’s also a 4570TE listed, but that’s another thing entirely).

* EDITED CORRECTION: the 4570 has a base clock speed of 3.2GHz with a higher TDP. h/t for the correction, @GMShadow

 

[GMShadow]

4570R was the chip in the base Late 2013 - it was pretty much Apple only because it had Iris Pro graphics, and most OEMs didn't want to pay the premium for Crystalwell. (the 'tray price' for the 4570S was $192, the 4570R was $288).

The 4570S is a bit of an odd duck anyway because it was merely a 4570 down clocked from 3.2 to 2.9GHz to keep the TDP at 65W instead of 84W.

As for this case? Turbo Boost has limitations. It does depend on the chip and generation, but in many cases the maximum Turbo Boost clock is for one core, not all of them. Running more cores at boost is going to generate heat that will reduce the actual real-world maximum. Assuming your encoding is properly multithreaded, I suspect that's the culprit.

EDIT: Just as an example, here's how the 4570 works with TB:

Default w/o turbo: 3.2 GHz
Turbo w/ 1 core loaded: 3.6 GHz
Turbo w/ 2 cores loaded: 3.5 GHz
Turbo w/ 3-4 cores loaded: 3.4 GHz

In your case with a 4570S that has a lower TDP, 3-4 cores must run at 3.2GHz.

 

[B S Magnet]

4570R was the chip in the base Late 2013 - it was pretty much Apple only because it had Iris Pro graphics, and most OEMs didn't want to pay the premium for Crystalwell. (the 'tray price' for the 4570S was $192, the 4570R was $288).

Indeed. The unit I have is equipped with the GeForce GT 750M, and I’ve been planning for swapping out the CPU at a later date (something not really feasible with the base model’s soldered CPU (unless you’re dosdude1 or someone with similar soldering chops).

The 4570S is a bit of an odd duck anyway because it was merely a 4570 down clocked from 3.2 to 2.9GHz to keep the TDP at 65W instead of 84W.

That checks out:

As for this case? Turbo Boost has limitations. It does depend on the chip and generation, but in many cases the maximum Turbo Boost clock is for one core, not all of them.

I’m aware a particular core can request a higher clock speed, as needed, and this would show up in the Intel Power Gadget window. In the case of the Sandy Bridge dual core MBP I mentioned, it doesn’t take anything special beyond throwing something processor-intensive at it (as simple as a 1080p YT clip, for example) for it to hop into Turbo Boost range to pick up the pace.

This Haswell, however, has never in three years of using it, ever climbed beyond 3.2GHz Turbo, and that’s even when I’ve had a Handbrake encode running; a Firefox browser open with 100+ tabs (yah, that’s how I do things); and watching a 1080 x265 full screen mode video in VLC, concurrently. And as noted earlier, I have Macs Fan Control ramp up the fan early and often, so I’m accustomed to seeing the max 3200rpm reading when putting the iMac under heavier load.

So either Intel fibbed on the Turbo Boost spec for this CPU, or else Apple limits it with whatever firmware updates my system is running (which I’ve only bumped to High Sierra 10.13.6, never higher… yet). Nothing else seems to add up here.

Running more cores at boost is going to generate heat that will reduce the actual real-world maximum. Assuming your encoding is properly multithreaded, I suspect that's the culprit.

Pardon the pun, but owing to all the details above, that doesn’t compute here. Owing to my fan settings, I’ve never had the system anywhere close to throttling thresholds (typically that would be somewhere north of the 90°C range, but more likely closer to the 96–100°C threshold).

So this question remains unanswered for the time being. Maybe I should ping @dosdude1 in on this one.

 

[Amethyst1]

Curiously, the “Products Formerly Haswell” page doesn’t list the 4570R.

The 4570R is Crystal Well.

This Haswell, however, has never in three years of using it, ever climbed beyond 3.2GHz Turbo […]

Try yes > /dev/null in Terminal to fully load a single core for the most demanding task ever. Still not going past 3.2 GHz?

4570R was the chip in the base Late 2013 - it was pretty much Apple only because it had Iris Pro graphics, and most OEMs didn't want to pay the premium for Crystalwell.

Gigabyte shipped BRIX barebones with the 4570R and 4770R.

The 4570S is a bit of an odd duck anyway because it was merely a 4570 down clocked from 3.2 to 2.9GHz to keep the TDP at 65W instead of 84W.

Those "S" chips have been around since the 45nm Core 2 Quads (Q8xx0/Q9xx0).

 

[B S Magnet]

The 4570R is Crystal Well.

Gotcha, cheers.

If they’re both 4th gen, are Crystal Well CPUs interchangeable with Haswell CPUs (the way, say, a Penryn CPU could replace some Santa Rosa CPUs)?

Try yes > /dev/null in Terminal to fully load a single core for the most demanding task ever. Still not going past 3.2 GHz?

Have a look here:

Interestingly, when I quit that command, the “Core Max” reading on the Power Gadget briefly blipped up to to 3.6GHz, but the “Core Avg” remained at its usual 3.2 to 3.22 figure (when under load), which comports with what iStat Menus reports.

 

[Amethyst1]

If they’re both 4th gen, are Crystal Well CPUs interchangeable with Haswell CPUs (the way, say, a Penryn CPU could replace some Santa Rosa CPUs)?

All Crystal Well CPUs are soldered and use BGA-1364, which is only shared with some mobile Haswell CPUs, and their Broadwell successors.

Intel CPUs allow for configurable limits, in both directions (usually referred to as cTDP up or cTDP down).

The 4570S doesn't though.

 

[Amethyst1]

Just tested my rig with the CrystalBroadwell i5-5675R. Full throttle under single-core load:

 

[B S Magnet]

Just tested my rig with the CrystalBroadwell i5-5675R. Full throttle under single-core load:

View attachment 2199681

Wow, it goes all the way to 3.6GHz. Nice.

Also, you might know the answer to this oddball question: is it still Haswell if it’s that unicorn called Devil’s Canyon (i.e., can the i7-4790K run in Haswell Macs from 2013, such as the iMac)?

 

[Amethyst1]

is it still Haswell if it’s that unicorn called Devil’s Canyon

Yup.

(i.e., can the i7-4790K run in Haswell Macs from 2013, such as the iMac?

The 2014 iMac 5K has it. It should™ work, but like PC mainboards requiring BIOS updates for "official" support, there may be pitfalls.