Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Optimal CPU Tuning settings for i9 MBP to stop VRM throttling / Explanation of Apple's Engineering F
- Thread starter winterny
- Start date
- Sort by reaction score
I updated the original post.
IMHO, for 99.9% of users, it's no longer worth messing around with this, and apple's update fixes this properly [within the constraints of the physical limitations] for most use cases.
There are still limits to the power delivery system, so of course you will experience some amount of throttling if you run things to 100%, but, it is far better than it was before, and probably better than a 2017 i7 MBP in any use case.
IMHO, for 99.9% of users, it's no longer worth messing around with this, and apple's update fixes this properly [within the constraints of the physical limitations] for most use cases.
There are still limits to the power delivery system, so of course you will experience some amount of throttling if you run things to 100%, but, it is far better than it was before, and probably better than a 2017 i7 MBP in any use case.
This is particularly interesting. This means Apple has some deeper level management / regulation where a firmware change is possible. I wonder if we have sensor data close enough to the VRM chips to watch temperature of its proximity pre and post "patch".
Didn't think to look there. Saw it now - great stuff. @winterny I dub thee hero of macrumours from my position of zero authority
Apple have typically had sensors all over the place, and the SMC, or now, T2, will be what actually manages thermals.
Though I find it curious that neither bootrom nor T2 firmware version seem to show as changed in system report? Update is clearly in place as my graphs in cinebench now look closer to what manually tweaking before did, though clearly with an extra layer of refinement over what the raw cpu management allow.
Everyone who bought a 2018, or was thinking of buying one, owes you a lot of thanks. Great job.
It still seems to be quite uneven in windows, though, sadly.
stil dropping to 800mhz and bouncing back, so, this is still not a complete solution if it's only active in mac os, not bootcamp.
stil dropping to 800mhz and bouncing back, so, this is still not a complete solution if it's only active in mac os, not bootcamp.
You can use Intel XTU under Boot Camp to implement something similar.
I can, but I should not /have/ to, and the apple solution seems finer grained than what intel XTU is doing.
Apple after all officially support Windows via boot camp, so this fix /should/ be working there too.
I’d agree, though Apple does not appear to. Most of my support interactions around Boot Camp end up being polite ways of telling me Apple chooses not to support it.
That's weird. I would have expected this to be handled by the T2, which would make it work in Windows too.
Unless the T2 has a separate configuration for bootcamp, which they didn't touch.
Pretty sure there is an EFI update, but I didn't take it completely apart yet.
Wonder if it just didn't take properly for me, but both shows same builds before and after update for me on 2.6 i7 ( bootrom 16.16.334.5.5.0 T2 16P50334e )
[doublepost=1532460291][/doublepost]
The firmware update parts are about 250mb, then there's 400mb each of OS/recovery updates, and some other stuff.
[doublepost=1532461919][/doublepost]Attached cinebench under current post update bootcamp first on default settings, then manually setting 39W throttle profile, which seems to roughtly match what macos is setting it to when running cinebench there. Daylight difference in throttline, and extra 100 points, as usual.
Attachments
Actually, testing this under mac, I'm still not totally sold. Performance looks hugely improved under a load test, but only for the first couple minutes. On a Prime95 "max power" torture test (which isn't great since it's loading the CPU in a way where you'll never get 2.9GHz continuous at 45W), it pretty quickly hits the point of VRM throttling, oscillating from 800MHz to 2.9GHz. On the "balanced" test, it climbs up to and settles at an operating point of ~40W, 2.9GHz, and ~95C, and stays there. But then after maybe 5m, the VRM throttling kicks in again.
I'd suggest testing with a more realistic workload at this point. Prime95 was useful in diagnosing the problem initially, as it was a very easy way to generate a heavy load, but I do not think it's particularly useful at this stage.
It looks like before there was absolutely no management of the power delivery at all, and now it is actively managed. This means that different workloads will result in different power delivery -- and that's probably a good thing for several reasons, including both longevity of the system and better handling of GPU workloads.
As an example, I would think doing 1080p or 4K video encoding is a much better way of testing at this point.
[doublepost=1532467901][/doublepost]
I just noticed that my system did go into what appeared to be a firmware or EFI update as it installed -- screen turned black for what seemed like a minute. I didn't look too closely at the installer package.
Last edited:
But still, no true cap on max power draw? While it's doing something now (whatever that is exactly), it still overheats the VRMs when people run very heavy loads, which doesn't seem to happen when people manually cap power draw. Sure, you get throttling, but not the below-idle panic throttle clock of 800MHz!
The question is how realistic these heavy tasks are? Running Prime95 torture test doesn't really qualify as doing useful work...