Max Tech confirmed that M1 Ultra's GPU isn't using more than 60W or way lower than that. Blender used only up to 30W for example. GFXbench worked well but I have no idea why it's happening. Can anyone test this with M1 Ultra and let us know since nobody but only Max Tech mentioned? If it's true, then I guess MCM isn't working well? Does Apple aware of this issue?
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.
M1 Ultra's GPU issue?
- Thread starter sunny5
- Start date
- Sort by reaction score
Max Tech confirmed that M1 Ultra's GPU isn't using more than 60W or way lower than that. Blender used only up to 30W for example. GFXbench worked well but I have no idea why it's happening. Can anyone test this with M1 Ultra and let us know since nobody but only Max Tech mentioned? If it's true, then I guess MCM isn't working well? Does Apple aware of this issue?
Blender is not working well on M1 Max either, this has nothing to do with MCM but a blender specific issue. Apple's approach of "MCM" is totally transparent to software and the dirty parts is handled by hardware themselves. Good news is that Apple is spending money on blender to improve its support on Apple silicon, the question is when the improvements will land.
Or it has to be using METAL, which apple have suggested Mac developers to use for like… 10 years now?
Metal will handle this for you transparently supposedly. Apple are pouring time/money into blender right now, so maybe a future version of 3.x will run better, when there’s an engine for it that actually runs metal.
When programming for GPUs the number of threads is almost always based exclusively on the size of the input data. I don't think there's a plausible scenario where someone would hardcode something like that. Threadgroup size maybe, but that would likely affect the M1 Max too, not just the Ultra.
Maybe the problems will be solved with the next Mac OS update. The word is it’s because they want to hide new products from being discovered before they’re released.
It makes no sense, if the OS release it comes with does supports the GPU then there is nothing hidden.
Even M1 Max doesn't work well on Blender. I have it and it performs just like 1660TI.
? Yes, you are right.
Anyway, I am puzzled why the apps don't use the double encoder or the full GPU power. Apps should need no change to use more GPU capacity, shouldn't they?
In other words, you may spend 2000 more on a machine that promises double power, but you will be rarely using it.
And BTW if I want these power to be used is in video editors as FCP X.
I was laaning towards the Ultra but now I think it is a waste of money.
I wish someone more knowledgeable than me would address this. Regardless of what benchmark I've tried I can't get the fans to spin up more than a ridiculous 20 rpm to 1345 max in my Ultra because the GPU just isn't running up to full power as shown in the first post of this thread. It seems that either app software needs major revisions to utilize the Ultra's larger GPU (but Apple said the software could treat the Max and Ultra the same without revision), or Apple's Mac OS or Firmware is limiting the GPU far below its thermal capacity. Seems it has to be the latter, since the M1 Max is not running up its fan speed either in benchmarks, which implies it's not app software just underutilizing the Ultra GPU.
So when are we going to hear from Apple about what's going on here? If it's turns out to be a hardware problem with the SoC's (which seems doubtful) I'd want to return my Ultra within the 14 day window. If its just a software/firmware issue then I'd assume it will be addressed by Apple.
Open a case with Apple Support and you might get a response from engineering.
Applications could be hardcoded for the individual SoC in the M1 family. We don't know but something is definitely amiss.
Apple tested Final Cut Pro with a prerelease version that still haven't been released.
I did some experiments on my M1 Max, summary is here: #74
Bottomline is that it takes A LOT OF WORK for the GPU to trigger it's maximal power mode, which is likely the problem with Geekbench and friends. It does not explain Blender and other results of course, that is likely due to imperfect software.
Bottomline is that it takes A LOT OF WORK for the GPU to trigger it's maximal power mode, which is likely the problem with Geekbench and friends. It does not explain Blender and other results of course, that is likely due to imperfect software.
Or does it? I scored around 3x the score on an M1 iPad on my 24 GPU M1 Max. The score at least scales and making sense. If you don't use the Unlimited mode, then you may have some problem that doest ramp the GPU as the workload is too light.
I mean, the scores I’ve seen are not bad at all. But according to videos the benchmark is fine very quickly. It would be curious to see the results of the stress test mode.
Were you able to trigger the fans to speed up? I still haven't seen the fans in my Ultra ramp up more than 25 rpm, to 1350 rpm, with any benchmark I've tried. Which begs the question why isn't the GPU running at a higher clock rate with so much additional cooling available unless their clock rate is constrained by factors other than heat.
I don’t own a Studio. My 16” MBP does get a bit louder under full GPU load. And how do you know what clock the GPU is running at? The only reliable tool is the Metal trace profiler.
I don't know what its clock rate is. But if we never see the GPU consume enough power to raise the temperature enough to increase the fan speed at all, then either the GPU clock rate is constrained by factors other than heat (and that could very well be), or why not allow the clock rate to increase? So far no one has reported the fans EVER leaving the idle speed with any benchmark or actual use, and that seems very odd.