I'm having the same issue on a 2017 macbook pro with the 7700hq, anyone been able to resolve the issue? Would like to undervolt the cpu and gpu.
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'm having the same issue on a 2017 macbook pro with the 7700hq, anyone been able to resolve the issue? Would like to undervolt the cpu and gpu.
Its realistic to some people. For example I might want to run an MCMC algorithm across all cores for 24 hours (or all cores -1 so I can do something else with the last core
)
Its realistic to some people. For example I might want to run an MCMC algorithm across all cores for 24 hours (or all cores -1 so I can do something else with the last core
I don't think you can fairly compare Prime95 and your everyday's MCMC algorithms... Prime95 was specifically designed to push the CPU very very hard by loading as many ports on the CPU as possible, without giving the system a break. I doubt it very much that an MCMC algorithm will reach that level of CPU utilisation.
Given how slow and the number of reboots I endured for such a small update, I'd bet there was an EFI update.
The average sustained Cinebench R15 multi-core CPU scores that notebookcheck.net has measured for other i9-8950HK implementations can be found in the screenshot below (source: https://www.notebookcheck.net/Intel-Core-i9-8950HK-SoC.279270.0.html). Four achieved 1300+, but they are all 8-10 lb. gaming laptops. The only one in the 15" MBP's weight class (it weighs 4.0 lbs.) is the 4.1 lb. Asus Zenbook Pro 15, which averaged 1074.
Also in the MBP's weight class is the 4.6 lb. Dell XPS 15 i9, whose stock performance averaged ~1000 but, with undervolting, averaged ~1150 (https://www.notebookcheck.net/Dell-XPS-15-9570-15-more-performance-by-undervolting.317738.0.html ).
So it would seem that if the post-fix i9 MBP could, under the same test conditions, average in the 1000 - 1100 range, it would be in the ballpark for multi-core thermal performance within its class—at least for the use cases approximated by this benchmark. [I am assuming that the OS doesn't significantly affect Cinebench CPU performance.]
Also in the MBP's weight class is the 4.6 lb. Dell XPS 15 i9, whose stock performance averaged ~1000 but, with undervolting, averaged ~1150 (https://www.notebookcheck.net/Dell-XPS-15-9570-15-more-performance-by-undervolting.317738.0.html ).
So it would seem that if the post-fix i9 MBP could, under the same test conditions, average in the 1000 - 1100 range, it would be in the ballpark for multi-core thermal performance within its class—at least for the use cases approximated by this benchmark. [I am assuming that the OS doesn't significantly affect Cinebench CPU performance.]
Last edited:
"Not useful work" is a really bad excuse, especially in a $3k laptop. Any workload doing the kind of work Prime95 does would hit similar temps and thus severe throttling, and what Prime95 does (matrix multiplications) IS a pretty common workload in scientific applications.
If you load up the AVX units - especially AVX512 units which I believe these CPUs support - with work you're likely to be grilling those VRMs, as long as there isn't a proper current draw limit set.
The update definitely contained t2 and boot firmware update files, but why has the version number not changed.
The problem with tuning for P95 is you have to lower max wattage by circa 10%, which directly affects the max performance everywhere else.
Alternatively, maybe they need to adjust the AVX offset.
Best solution would be a dynamic system which modulates the current thresholds to avoid sustained throttling.
“Oh, it’s VRM or thermal throttling, let’s lower sustained wattage a bit for a time”
“Hasn’t been throttling for a bit, let’s restore...”
Sounds a bit like what a T2 chip should be doing, and also sounds like something that can be done in future updates.
[doublepost=1532486569][/doublepost]
Maybe the files were always there... but not signed right, so the T2 rejected them
Alternatively, maybe they need to adjust the AVX offset.
Best solution would be a dynamic system which modulates the current thresholds to avoid sustained throttling.
“Oh, it’s VRM or thermal throttling, let’s lower sustained wattage a bit for a time”
“Hasn’t been throttling for a bit, let’s restore...”
Sounds a bit like what a T2 chip should be doing, and also sounds like something that can be done in future updates.
[doublepost=1532486569][/doublepost]
Maybe the files were always there... but not signed right, so the T2 rejected them
It's certainly plausible.
A theory I've seen floated is that something from the power management system was signed with a key which was only valid on internal/prototype machines, and therefore would not run on production machines.
I spent about a half hour looking through the upgrade package, and with the sheer quantity of files included, it's a bit difficult to pinpoint exactly what they changed.
[doublepost=1532488696][/doublepost]
True, but those kinds of scientific workloads, if they are long-running, are typically run on servers, not laptops.
[doublepost=1532488783][/doublepost]
Perhaps because the version didn't change, and only the cryptographic signing of those things changed.
Possible, but it would be more reassuring to see a version number change! And still like to know why the updated profiling only works on macOS, This is why I kind of wonder if update took properly, if it's only being managed in macOS.
I'd rather not have to manually manage the profile under Windows, I hope people keep up at Apple till they fix it in both 'supported' environments.
Well, I can confirm that the fix seems to work well in all my testing scenarios. It doesn't quite get the same sustained performance as my hand tuned settings (which are specifically tuned for my laptop), but does get better instantaneous boost... which is more important to me.
I definitely won't be bothering with any hand-tweaked boost parameters.
Now, it'd be nice if they fixed it in windows... at least with some sortof daemon... I'm guessing if the problem is fixed by a soft-firmware update, loaded by macos, then Linux and Windows are going to have issues on T2 macs.
I definitely won't be bothering with any hand-tweaked boost parameters.
Now, it'd be nice if they fixed it in windows... at least with some sortof daemon... I'm guessing if the problem is fixed by a soft-firmware update, loaded by macos, then Linux and Windows are going to have issues on T2 macs.
You could say that, and also, you could say Apple brags about how fast this thing is - and they certainly charge enough money for it - so going like, "this is good enough" and adding a slapdash power regulation feature to a premium laptop doesn't actually seem good enough.
People pay for the little extra when they buy Apple - they deserve getting it, not a spit in the face.
The missing digital key is definitely plausible: a configuration management and testing issue. But given the size and breadth of the update, it is clear that more than just correctly signing an extension was involved.
I'm not certain of this but I believe if you go under About this Mac -> System Report...:
- Boot ROM Version:
- 15.16.6703.0.0,0 (patched)
- 15.16.6613.0.0,0
Lots of people these days doing Bayesian MCMC algorithms on laptops. Maybe not quite as demanding as Prime95, but its not unusual to run MCMC chains for hours/days on a laptop. Not everyone has access to a server!
A spit in the face? Excuse me, I just spent several hours running various numerical workflows on the new machine (i9 CPU post patch). Not only was I unable to trigger any VRM throttling, but this thing also consistently hits higher frequencies on each of its cores on heavy loads than my i7-6820HQ manages on a single core under max boost! This is a massive performance increase to anyone who runs scientific or engineering workflows on a laptop.
Yes, larger and heavier laptops will perform better. Need a truly performant machine? Get one of those. The MBP is still a great solution for folks that need to be mobile.
Guys, anyone tried Fortnite after the update? I am not a fan of Fortnite and do not believe really in gaming on a Mac but Apple itself advertises gaming as one of the features of the new MacBook Pro and they even have a screenshot of Fortnite on their website.
Fortnite it totally free if someone wants to give their CPU/GPU combo one last torture test. The game used to stutter a lot pre-patch but someone posted in another topic that they found no difference after patch too.
Fortnite is a Mac game and optimized on Metal (was announced on stage in one of the WWDCs in the past few years).
Fortnite it totally free if someone wants to give their CPU/GPU combo one last torture test. The game used to stutter a lot pre-patch but someone posted in another topic that they found no difference after patch too.
Fortnite is a Mac game and optimized on Metal (was announced on stage in one of the WWDCs in the past few years).
Apple's Mac division, while it may not be their largest, is enormous in absolute terms. I once estimated that, by itself, it would rank as a Fortune 100 company—about the same size as Gilead Biosciences.
Given the substantial resources that come with this, it seems they could devote just one SKU to a more pro-focused MacBook Pro—something that would be to the MacBook Pro what the iMac Pro is to the iMac (maybe call it the MacBook Pro^2?). [And yes, parenthetically, update the Mac Mini!]
This would be targeted towards creatives, developers, scientists, and engineers. I'm not suggesting they do anything so non-Apple as an 8–10 lb. gaming laptop. But it would be nice if they could, with somewhat more weight and thickness, tilt the engineering optimization of the product towards better thermals and thus higher performance (and throw in an FAA-maximum 100 Whr battery, and a greater variety of ports, as well).
Consider, for instance, the Gigabyte Aero 15X v8. With just an extra 0.5 lbs. of weight, and 0.1" of thickness (vs. the 15" MBP), and using just a 2.2 GHz i7-8750H, its thermals enable it to turn in multicore Cinebench R15 CPU scores ~10% higher than what I've seen posted for the post-fix 2.9 GHz i9-8950HK MBP.
This would be a neat engineering exercise for Apple—what's the lightest, thinnest laptop they could make that would perhaps not meet, but at least approach, the CPU performance of the fastest i9 gaming laptops? And if they succeeded in making a relatively light, thin device that met those parameters, it would be classic Apple—engineering hardware that no one else has.
As one app developer plaintively posted a while back (I paraphrase): 'I understand that we are only a small part of Apple's market. But we are the ones that make the software that keeps the rest of Apple's market happy....'
Given the substantial resources that come with this, it seems they could devote just one SKU to a more pro-focused MacBook Pro—something that would be to the MacBook Pro what the iMac Pro is to the iMac (maybe call it the MacBook Pro^2?). [And yes, parenthetically, update the Mac Mini!]
This would be targeted towards creatives, developers, scientists, and engineers. I'm not suggesting they do anything so non-Apple as an 8–10 lb. gaming laptop. But it would be nice if they could, with somewhat more weight and thickness, tilt the engineering optimization of the product towards better thermals and thus higher performance (and throw in an FAA-maximum 100 Whr battery, and a greater variety of ports, as well).
Consider, for instance, the Gigabyte Aero 15X v8. With just an extra 0.5 lbs. of weight, and 0.1" of thickness (vs. the 15" MBP), and using just a 2.2 GHz i7-8750H, its thermals enable it to turn in multicore Cinebench R15 CPU scores ~10% higher than what I've seen posted for the post-fix 2.9 GHz i9-8950HK MBP.
This would be a neat engineering exercise for Apple—what's the lightest, thinnest laptop they could make that would perhaps not meet, but at least approach, the CPU performance of the fastest i9 gaming laptops? And if they succeeded in making a relatively light, thin device that met those parameters, it would be classic Apple—engineering hardware that no one else has.
As one app developer plaintively posted a while back (I paraphrase): 'I understand that we are only a small part of Apple's market. But we are the ones that make the software that keeps the rest of Apple's market happy....'
Last edited:
Hear Hear. Here's my list :
- a couple mm thicker and re-design thermal protection for faster CPU/GPUs
- thicker means better keyboard - sorry but classic 2014 style rmbp keys still rules
- thicker means better battery life
- thicker means better built-in graphics
- expandable ram and SSD (may have to re-think T2) to give us better short-term price options
- bring back at least micro SD slot and USB3+ - gives mobile photographers/videographers better options
- OLED screen
- get rid of touch bar - sorry good try but still a solution searching for a problem
Just to start. At this point, the design is really a constraint pros can live without...Those who want trendy thin and cool can opt for the air.
Thanks for the upvote—we're thinking along the same lines! I agree with everything you wrote, except for the very last—the current MBP is a very popular product, and provides a great trade-off for consumers/prosumers that need more power than the Air or MacBook, but still want thin/light/quiet. Thus I think it would be a better business decision to continue to offer the current MBP design, but then add a more pro-focused laptop like we've both described (akin, as I mentioned above, to Apple's decision to add the iMac Pro to the iMac line).
I also like the basic design - but I think they could come up with slight variations with better ventilation in the rear. Just sayin'. In fact, I think they already have design studies galore - management just needs to set a priority.
No question, seeing the temps easily spike up to 100c on my 2.2, shows me that apple dropped the ball
Yes, I did.
I'm the guy who posted the video of Fortnite stuttering on Twitter the performance has been fantastic since the update.
560x + 2.6ghz
Last edited: