M1 Ultra & Under Utilization of the CPU and GPU

[hopkins802]

Hey all,

I know there is a similar thread out there but I'm curious if anyone has any ideas as to why the M1 Ultra seems to be throttling both it's CPU and GPU power consumption and core clock frequency, when the temps are clearly under control? It appeared based on Max Tech's tests, that the CPU and GPU both have a bunch of headroom. Cinebench R23 always will push your CPU to the max, yet this 3.23Ghz processor is only being utilized at 3.00 Ghz. And the GPU frequency is under as well. Not to mention, the GPU is barely hitting 60W when it should hit 100W.

The main question is, do you all think this was a mistake and that Apple will fix it in a software update? It's kind of crazy that the CPU and GPU are barely hitting 60C when fully pressed. It should be able to crank much higher.

 

[ct2k7]

My initial thought is that these are mobile first chips. But who knows.

 

[hopkins802]

My initial thought is that these are mobile first chips. But who knows.

Yeah, even if that's true though, what prevents them from reaching their advertised core frequency specs? Is it because they're designed to be on the safer size due to the limited thermal control in a laptop? My M1 Max GPU when fully pressed (in Davinci Resolve) hits 85-90C and my CPU cores hit 90-100C...

 

[diamond.g]

Yeah, even if that's true though, what prevents them from reaching their advertised core frequency specs? Is it because they're designed to be on the safer size due to the limited thermal control in a laptop? My M1 Max GPU when fully pressed (in Davinci Resolve) hits 85-90C and my CPU cores hit 90-100C...

Does Apple advertise core frequencies?

Per Anandtech M1 Max review:

The CPU cores clock up to 3228MHz peak, however vary in frequency depending on how many cores are active within a cluster, clocking down to 3132 at 2, and 3036 MHz at 3 and 4 cores active. I say “per cluster”, because the 8 performance cores in the M1 Pro and M1 Max are indeed consisting of two 4-core clusters, both with their own 12MB L2 caches, and each being able to clock their CPUs independently from each other, so it’s actually possible to have four active cores in one cluster at 3036MHz and one active core in the other cluster running at 3.23GHz.

So what is seen may in fact be normal when the die is fully loaded.

 

[hopkins802]

Does Apple advertise core frequencies?

Per Anandtech M1 Max review:

So what is seen may in fact be normal when the die is fully loaded.

This test is what bums me out -

It shows the GPU on the 64core scaling down to 750Mhz when fully loaded, and scaling its power consumption way, way down.

 

[leman]

Cinebench R23 always will push your CPU to the max, yet this 3.23Ghz processor is only being utilized at 3.00 Ghz.

M1 Firestorm clusters top out at 3.00ghz if more than one core is active, thermal headroom or not. Why, only Apple knows. Could be some sort of design limitation, could be the lowest common denominator to ensure highest possible yields.

And the GPU frequency is under as well. Not to mention, the GPU is barely hitting 60W when it should hit 100W.

It very well may be that the software used simply is unable to properly utilise that humongous GPU. One would need to run some sustained ALU throughput tests to see where there is a problem or not. As I wrote in one of the other threads, if the GPU kernel execution time is under 10ms, the GPU won't enter the high power state.

 

[hopkins802]

M1 Firestorm clusters top out at 3.00ghz if more than one core is active, thermal headroom or not. Why, only Apple knows. Could be some sort of design limitation, could be the lowest common denominator to ensure highest possible yields.




It very well may be that the software used simply is unable to properly utilise that humongous GPU. One would need to run some sustained ALU throughput tests to see where there is a problem or not. As I wrote in one of the other threads, if the GPU kernel execution time is under 10ms, the GPU won't enter the high power state.

The tests are done in a program notorious for hammering the GPU and hammering it well, Davinci Resolve. It’s extremely optimized for the M1 chip. Is there a possibility that it can be optimized more to kick the GPU into higher gear when needed?

 

[Pressure]

This test is what bums me out -

It shows the GPU on the 64core scaling down to 750Mhz when fully loaded, and scaling its power consumption way, way down.

It fells premature before applications have been updated specifically for the M1 Ultra.

 

[hopkins802]

It fells premature before applications have been updated specifically for the M1 Ultra.

True. I decided to cancel my pre-order until it’s proven that the Ultra can actually perform at a significantly higher level than the Max.

 

[leman]

The tests are done in a program notorious for hammering the GPU and hammering it well, Davinci Resolve. It’s extremely optimized for the M1 chip. Is there a possibility that it can be optimized more to kick the GPU into higher gear when needed?

Difficult to say without having any debug info. Resolve is a video editing suite, does it actually submit enough work to properly utilise the Ultra? Hard to tell.

 

[hopkins802]

Difficult to say without having any debug info. Resolve is a video editing suite, does it actually submit enough work to properly utilise the Ultra? Hard to tell.

It's actually a color grading suite that has gotten better at editing over the years. It fully presses the GPU and has been widely known as the king of GPU utilization among the top three NLEs (FCP, Resolve and Premiere). Premiere is now using the GPU more, but Resolve decodes and encodes heavily with the GPU. Anyway, I guess we'll find out once more updates are done.

 

[bobcomer]

This test is what bums me out -

It shows the GPU on the 64core scaling down to 750Mhz when fully loaded, and scaling its power consumption way, way down.

I sure hope it's a software issue. If it's power distribution, that's a lot harder to fix. I'm only getting the max, and I don't have much use for the GPU, so I expect I wont see any throttling at all, but for those with the Ultra, ...

I'm guessing it's just software and Apple decided they'd play it safe for some period after release.