diglloyd Mac Performance Guide

Up to 8TB of Thunderbolt Storage!

100% Kona, 100% Family Owned

Mac Pro Nehalem Power Usage Spike Playing Music

Last updated February 04, 2010 - Send Feedback

In late 2009, a reader brought a bizarre problem to my attention. It seemed innocuous enough, and I could not pay much attention at the time, due to other pressing projects. That same reader sent me more detailed information in late January which showed just how irksome the problem actually was: a massive power usage and very hot CPUs. Of special concern, at least long term, is very hot CPUs with no apparent attempt by the CPU fans to cool them.

Playing a sound track consumes 143 watts!

The graph below shows just how bizarre the problem is. My fully-loaded 8-core 2.93GHz Mac Pro Nehalem consumes 202 watts at idle, as measured by a WattsUp! power meter.

Launch iTunes and play a track. Mozart did the trick for me, but it doesn’t matter. I played through the awful built-in speaker; that doesn’t matter either.

Power usage at idle vs maxed-out vs full CPU usage

To determine how much power fully-loaded CPUs could consume, the MemoryTester program was used with the command 'mt compute'.

The 'mt compute' command maxes-out all 16 virtual cores, using all available computing power (1600% CPU usage). Yet power usage peaks at 393 watts, only 14% more than when playing music.

Mac Pro Nehalem Photoshop speed
At idle, 202 watts
Playing a track in iTunes with 2.5% CPU usage ramps power usage by 143 watts!
At 1600% CPU usage, power usage is 393 watts

* Mac Pro Nehalem 8-core 2.93GHz, 64GB memory, 4 X 2TB Hitachi 7K2000, SSD, 2 X FirmTek SeriTek 2ME4-E, 2 X NVidia GeForce GT120 video cards.

Other circumstances cause the issue

Some users report that merely having sound-related equipment attached via USB or Firewire can cause similar symptoms. I did not investigate this.

Performance impairment?

Some users have reported impaired performance in some programs while playing music.

Using MemoryTester with 'mt compute', I detected no impact on compute performance while playing music, as measured by the aggregate speed reported.

Using MemoryTester with 'mt stress', I detected no impact on memory bandwidth.

Those results show no impairment, but they also use all 16 virtual cores. simultaneously.

I then ran XBench and it showed a substantial degradation in performance, but XBench is almost entirely single-threaded and thus a very different test.

I haven’t yet run real-world tests, such as RAW-file conversion in a program like Lightroom.

Conclusions

Substantial performance impairment while playing music looks to be a serious concern, which I'll be looking into more.

Consuming 143 watts to play a music track is ridiculous. On less “loaded” machines, it’s likely that the power usage would double over the idle usage (my machine is stuffed full with memory, drives and cards).

Perhaps it’s just a bizarre side effect that is “normal”, but it’s offensive to the conscience to consume 143 additional watts to listen to one music track, let alone the extra heat it dumps into my office, and my higher power bills (my bill hits the $.44/kilowatt hour tier).

 


Up to 8TB of Thunderbolt Storage!

diglloyd.com | Terms of Use | PRIVACY POLICY
Contact | About Lloyd Chambers | Consulting | Photo Tours
Mailing Lists | RSS Feeds | Twitter
Copyright © 2008-2014 diglloyd Inc, all rights reserved.