Apple confirms MacBook Pro thermal throttling, software fix coming today

Photo by Vjeran Pavic / The Verge
If you buy something from a Verge link, Vox Media may earn a commission. See our ethics statement.

For a week, we have been seeing reports that the newly released MacBook Pros run hot, which all kicked off after this video by Dave Lee. They run so hot, in fact, that the very fancy 8th Gen Intel Core processors inside them were throttled down to below their base speed. Apple has acknowledged that thermal throttling is a real issue caused by a software bug, and it’s issuing a software update today that is designed to address it.

The company also apologized, writing, “We apologize to any customer who has experienced less than optimal performance on their new systems.”

Apple claims that it discovered the issue after further testing in the wake of Lee’s video, which showed results that Apple hasn’t seen in its own testing. In a call with The Verge, representatives said that the throttling was only exhibited under fairly specific, highly intense workloads, which is why the company didn’t catch the bug before release. The bug affects every new generation of the MacBook Pro, including both the 13-inch and 15-inch sizes and all of the Intel processor configurations. It does not affect previous generations.

Here’s the company’s statement, in full:

Following extensive performance testing under numerous workloads, we’ve identified that there is a missing digital key in the firmware that impacts the thermal management system and could drive clock speeds down under heavy thermal loads on the new MacBook Pro. A bug fix is included in today’s macOS High Sierra 10.13.6 Supplemental Update and is recommended. We apologize to any customer who has experienced less than optimal performance on their new systems. Customers can expect the new 15-inch MacBook Pro to be up to 70% faster, and the 13-inch MacBook Pro with Touch Bar to be up to 2X faster, as shown in the performance results on our website.

Apple declined to provide more detail on what precisely this “missing digital key” is — beyond the fact that its lack impacts the thermal management system. The company is sticking to its stated performance claims on the new machines, and it will add one more benchmark graph to its official MacBook Pro page on its website to reflect more recent tests.

The admission caps off a full week of drama over these new MacBook Pros. Surprisingly enough, thermal throttling is not the main storyline in this drama. That role falls to the keyboard. Apple persists in insisting that the redesign of the keyboard to include a silicone barrier was to make the keyboard “quieter.” It does so in the face of its own internal documents that state plainly how the new design should help its keyboards be more reliable when crumbs or grit make their way inside.

As for the throttling drama, making sense of the various articles and videos benchmarking these machines has been nothing less than head-spinning, and making sense of all the differing results has been a challenge. Here’s a couple, though: Jonathan Morrison ran an extensive battery of tests that show the i9 MacBook Pros to run faster than the i7. On the other hand, the founder of Geekbench created some custom tests that showed that the top-end i9 processors performed slower:

Now that there’s a software fix that puts the so-called “missing digital key” back in to better manage the temperature of the processors, it looks like everybody will need to run those tests all over again. That’s precisely what we’re going to do ahead of our review.

We’ve also reached out to Intel and will report back if they have comment.

Comments

Sucks to be an early adopter, as ever. Good of them to apologise, they’ve been lax with those recently.

Even with the fix its still slower than the I7…

Still technically slower than the i7, but close enough that, practically, it doesn’t matter.

What doesn’t matter? That it’s slower than the i7? Really?

I think he’s saying for all intents and purposes they perform identically rather than caring about how much of a rip off the thing is.

Still a rip off.

Doesn’t performing identically absolutely solidify the 300 dollar up-sell i9 being a ripoff?

Do they perform identically in every scenario or just in a few tests? You don’t have to get back to me. I already know the answer.

Still sounds like most things this is tailored for would hit the identical speeds rather than single core or burst speeds. Large compiles, renders, fluid dynamics, etc. So again, the i9 doesn’t seem worth the upsell for most people with the right compute load to want an i9.

The good news is for all of those jobs you don’t need to feel like you didn’t get the top end with the midrange i7…

Because Apple essentially lowered the max Turbo Boost speed the Core i9 can hit to keep the temperatures down. The actual problem is VRMs overheating trying to power the chip at higher clock speeds and that’s a hardware flaw that can’t be fixed without re-engineer the cooling and motherboard.

The upside of Apple’s fix is that your MacBook would at least live it’s normal life. Previously, overheating the VRMs meant it would most definitely fail sooner rather than later. Still a massive rip-off paying for the Core i9 since you’re not actually getting full performance of that chip.

Everybody’s repeating the VRM bit as if it were a 100% certainty. There is a lot of evidence out there that suggests it might not be true after all, eg the processors in many of these tests actually stick to the 100 degrees limit suggesting thermals of the processor itself are the problem, not those of the power delivery.

Nobody’s getting full performance on any mobile chip these days on thin, portable laptops. All of them throttle. And there are many use cases where, even throttled, the i9 pull ahead with a comfortable margin.

Doesn’t disprove that both could have been the issue, i.e the VRMs causing the severe dips when the CPU wasn’t at the 100 degree tjunction max temperature. Still needs more investigation.

I didn’t say it proves anything. I’m merely saying that that claim might not be entirely true and as you said, this needs further analysis.

I don’t think you understand how VRMs actually work, and unfortunately, explaining that on a Verge comment isn’t really possible.

A processor can hit Tj(max), which 100°C for current chips without actually drawing a lot of power simply because the heat sink on the processor is not dissipating it efficiently. Conversely, it is also possible for a processor to overload and trip the VRMs without hitting Tj(max) itself if it’s cooling is better than the cooling on the VRMs.

Different workloads will show different behavior since not all components on a CPU behave in the same way. Integer ops for instance require much less power than floating point operations. And even beyond the CPU, in the MacBook Pro case, the VRMs are also responsible for supplying power to the other components like the GPU etc. so a sufficient load from the GPU + sufficient load from the CPU would still hit the VRM limits without either being at 100%.

Edit: I’m not only pissed at Apple with this either. Most i9 based notebooks have this problem right now where they aren’t essentially fully utilizing the chip and we should be speaking against that since we’re paying for it.

A processor can hit Tj(max), which 100°C for current chips without actually drawing a lot of power simply because the heat sink on the processor is not dissipating it efficiently. Conversely, it is also possible for a processor to overload and trip the VRMs without hitting Tj(max) itself if it’s cooling is better than the cooling on the VRMs.

I feel like we’re saying the same thing here. That was my idea. Processor at 100° likely not drawing enough power to cripple the VRMs meaning any throttling seen is likely solely due to processor temperature because of poor cooling and not VRM.

Sorry if I wasn’t clear. What I said was both are happening based on what kind of workloads are being run. It’s fairly easy to tell if the throttle is because of VRMs not supplying enough power or because the processor itself is hitting Tj(max).

Throttle because of CPU temps has a smoother slope since it’s gradual. Throttle because of VRMs not supplying enough power is quite abrupt. Most of the charts and graphs I’ve seen point to a VRM issue (also confirmed by several hardware experts which had the laptop on hand).

Throttle because of CPU temps has a smoother slope since it’s gradual. Throttle because of VRMs not supplying enough power is quite abrupt.

Didn’t know this, gotcha!

it doesn’t matter.

It does matter when the i9 cost more than $300 over the i7.

Nah, any difference lower than $500 is negligible for them.

Sucks to be an early adopter, as ever.

There is a very simple solution, though… don’t be one.

In the vast majority of cases, it really wouldn’t hurt to hold on to your cash for a few days, weeks.

I mean, I’m more than happy that early adopters exist, so the rest of us can avoid most of the pain, but they should be fully aware of the risks they are taking, so my sympathy is limited.

WTF is a "digital key"?

Also, apparently the same person that wrote "teather" instead of "tether" in Alien Colonial Marines is now working at Apple.

Digital Key: Code for "oh shit, turn fans to overdrive"

Digital keys commonly refer to one of two things in the software world.

1) A "digital key" can refer to a "signature" that "signs" a piece of functionality and that signature is required to verify it as authentic in order to be run. Think of it as a signature on a check. No signature, the check cannot be exchanged for currency. It is possible that some functionality with the thermals was not being run, because it had no signature.
2) It could refer to a key that is paired with another to enable functionality. For instance, with web protocols, a key, or certificate is presented from a client to a server. The server validates the certificate, and if valid, allows traffic. In essence, just like #1, lack of it blocks functionality.

They accidentally pulled the wrong motherboard MSR register value for platform power.

-Dan Matte

The VRMs were being utilitized incorrectly it seems.

View All Comments
Back to top ↑