2023 Mac Pro M2 Ultra: Sleep/Wakeup Causes Drastic Loss of Performance for both CPU and GPU

2023-07-18 • SEND FEEDBACK |
Related: Apple, Apple 2023 Mac Pro M2 Ultra, Apple Core Rot, Apple macOS, Apple macOS Ventura, Apple Silicon, GPU, Mac Pro, Photoshop, video tech

re: Apple Core Rot

UPDATE 2023-08-02: this bug appears to be fixed in macOS Ventura 13.5.

...

macOS Ventura 13.4.1 (c) (22F770830e)

Sleeping the 2023 Mac Pro M2 Ultra can result in a “permanent” loss of both CPU and GPU performance, that is, until reboot.

My first puzzler in running tests to evaluate the 2023 Mac Pro M2 Ultra were some weird discontinuities in performance; things were either fast or slow with no in-between... huh? When in doubt, reboot, which I did. Now it all makes sense! I had found the cure for a nasty problem without realizing it!

MPG previously reported upon severe performance degradation of both CPU and GPU with the 2023 Mac Pro M2 Ultra:

2023 Mac Pro M2 Ultra: Large Losses in CPU and GPU Performance; Rebooting Recovers temporarily
2023 Mac Pro M2 Ultra: Large Losses in GPU Performance Require Reboot

I can repeat the problem at will 100% of the time. There are no extra cards installed and nothing special installed, indeed my install was manual with no Migration Assisant—as clean as one can possibly get.

This should be easy enough to check by anyone else who has a 2023 Mac Pro M2 Ultra. But of course it is possible that there is some provoking factor including things like rapid security updates. Hopefully no one “proves” me wrong with logical fallacies like “works for me”.

My GUESS is a BUG that downclocks the CPU core clock frequencies, this lowering performance no matter how many cores are used

As yet, I have not observed anything like this ever before including on my 2021 MacBook Pro M1 Max. It is likely specific to the M2 Ultra and perhaps even to the Mac Pro only.

How to provoke the issue: sleep and wake

I went for a walk, sleeping the Mac Pro prior. Upon return, it was in degraded “Slow Mode” again. It made me wonder: does sleeping the Mac Pro cause it to enter Slow Mode? Indeed it does.

As yet, MPG has found no way to restore normal performance short of rebooting.

The only workaround seems to be to never allow the Mac Pro to sleep. That means not sleeping it explicitly, and enabling Prevent automatic sleeping when the display is off.

UPDATE, July 22: in 4 hours, Slow Mode has not returned, a strong semi-proof that it indeed relates to system sleep/wake, which I disabled.

BEFORE sleep:

SHA-512: 1513 MiB/sec for 1250 iterations over 1.00 MiB lasting 826ms size 1250 MiB
SHA-512: 1570 MiB/sec for 1250 iterations over 1.00 MiB lasting 796ms size 1250 MiB
SHA-512: 1504 MiB/sec for 1250 iterations over 1.00 MiB lasting 831ms size 1250 MiB
SHA-512: 1504 MiB/sec for 1250 iterations over 1.00 MiB lasting 831ms size 1250 MiB
SHA-512: 1547 MiB/sec for 1250 iterations over 1.00 MiB lasting 808ms size 1250 MiB
SHA-512: 1511 MiB/sec for 1250 iterations over 1.00 MiB lasting 827ms size 1250 MiB
SHA-512: 1580 MiB/sec for 1250 iterations over 1.00 MiB lasting 791ms size 1250 MiB
SHA-512: 1578 MiB/sec for 1250 iterations over 1.00 MiB lasting 792ms size 1250 MiB
SHA-512: 1578 MiB/sec for 1250 iterations over 1.00 MiB lasting 792ms size 1250 MiB

AFTER sleep (permanent reduction in performance, REBOOT required to fix):

“Permanent” loss of performance. Testing over and over shows a major loss of speed. Rebooting seems to be the only cure.

SHA-512: 1002 MiB/sec for 1250 iterations over 1.00 MiB lasting 1247ms size 1250 MiB
SHA-512: 1033 MiB/sec for 1250 iterations over 1.00 MiB lasting 1210ms size 1250 MiB
SHA-512: 997 MiB/sec for 1250 iterations over 1.00 MiB lasting 1254ms size 1250 MiB
SHA-512: 1062 MiB/sec for 1250 iterations over 1.00 MiB lasting 1177ms size 1250 MiB
SHA-512: 1341 MiB/sec for 1250 iterations over 1.00 MiB lasting 932ms size 1250 MiB
SHA-512: 1248 MiB/sec for 1250 iterations over 1.00 MiB lasting 1002ms size 1250 MiB
SHA-512: 1119 MiB/sec for 1250 iterations over 1.00 MiB lasting 1117ms size 1250 MiB
SHA-512: 1065 MiB/sec for 1250 iterations over 1.00 MiB lasting 1174ms size 1250 MiB
SHA-512: 1089 MiB/sec for 1250 iterations over 1.00 MiB lasting 1148ms size 1250 MiB
SHA-512: 1192 MiB/sec for 1250 iterations over 1.00 MiB lasting 1049ms size 1250 MiB

Below, visually:

macOS Ventura: wildly differing CPU performance, degraded vs rebooted

Below, the same issue afflicts GPU-intensive tasks in Photoshop.

macOS Ventura: degraded Photoshop performance

The world’s most profitable company cannot test properly?

In MPG’s opinion, Apple rushed the Mac Pro out the door with poorly tested software. How can a bug of this magnitude ever see the light of day for end users?!!!

It’s not just this issue, but the guaranteed kernel panic with NVMe blades.

By definition, if an end user can find such severe problems within 24 hours with no special tools or software, the software quality assurance was sloppy, at best.