My Hacintosh Experiment

Previous: Performance Benchmarking...

Performance Tweaks

Memory Tweaks

Curious to increase the performance a little bit, another 2GB stick of DDR3 matched memory was installed into the motherboard's second DDR3 slot, bringing total memory to 4GB and enabling Dual Channel Interleaving. Xbench was run seven consecutive times on each of the 2GB and 4GB setups, the results averaged for each, and compared:
MicroMini 2 vs 4 GB DDR3
(Cilck image for larger version)
The average Memory Test score increased 15%, while the Thread Test score remained essentially unchanged due to offsetting Lock Contention and Computation scores. The Overall Score only increased slightly, suggesting the Memory Test is not a large contributor.

Memory Speed Issue

After adding the second memory stick and running the above test, I noticed Apple's System Profiler reporting "4GB of 1066MHz DDR3" memory instead of 1333MHz memory. This was confirmed by Gigabyte's BIOS settings in the "MB Intelligent Tweaker" panel: The "Memory Frequency" setting displayed "1333 1066", indicating the 133MHz memory was being run at 1066MHz.

Wanting to have the system run the memory at it's rated speed of 1333MHz, I used the tweaker panel to vary the Performance Enhancement, (G)MCH Frequency Latch, and System Memory Multiplier (SPD) values, re-ran Xbench, and documented the change in performance for a number of combinations of these three parameters:
MicroMini BIOS DRAM Tweaks
(Cilck image for larger version)
For each configuration tested, Xbench was run seven consecutive times and the results averaged. Of the configurations tested, and assuming larger values are always better, the best configuration was "Ext-200-4.00C-1333", corresponding to:
  • Performance Enhance = Extreme
  • (G)MCH Frequency Latch = 200MHz
  • System Memory Multiplier (SPD) = 4.00A (or, 4.00 @ 266MHz FSB)
  • Memory Frequency (MHz) = 1333 1333 (or, the DDR3 RAM rated at 1333Mhz was being run at 1333MHz)
These improvements are in addition to the gains for using Dual Channel Interleaving discussed above. Although the "MB Intelligent Tweaker" panel allows for a variety of other BIOS adjustments affecting the DRAM memory, I've done enough tinkering for now: Anyone with further tweak results on this motherboard is welcome to contact me.

PCI-Express Frequency Increase

The video card installed in the motherboard's PCI-Express slot runs at a default frequency of 100MHz.  Increasing the frequency may provide faster performance, but with a greater risk of video instability:  Eventually, the video will start to exhibit artifacts, and in the extreme the computer may not startup.

The video tests within Xbench were run seven consecutive times on the default setting and the results averaged to establish baseline performance. Randomly increasing the PCIe Frequency to 108MHz via the "MB Intelligent Tweaker" panel resulted in a computer that failed to boot. After waiting for the BIOS to automatically recover and reboot, the PCIe Frequency was lowered to 104MHz and the computer booted normally.

Xbench testing was repeated for this new setup and then compared to the baseline:
PCIe Freq Tweak
(Cilck image for larger version)
The Overall Result improved over 3%, driven by a 7% increase in the OpenGL Graphics Test score: Modest, but I'll take it. Note that Geekbench won't see this performance increase as it does not perform video testing.

Xbench Score Variability

Of note, during the Xbench testing the Thread Test scores seemed to vary greatly from run to run despite the same hardware and BIOS setup. The results from the baseline 4GB memory configuration, discussed above, were typical of those seen during all Xbench testing:

(Cilck image for larger version)
Within the group of Computation scores, two were roughly double the values of their collegues: This behavior was always observed for consecutive runs of Xbench on the same setup. Whether this is actual computer behavior or due to Xbench is unknown.

Geekbench

Satisfied with the MicroMini setup for now, I re-ran Geekbench:
MicroMini Hackintosh: Geekbench 2 Summary
This score is 2.5% higher than the baseline MicroMini setup. Compared to Geekbench's historical reported score for Apple's Intel-based Mac Mini models, the MicroMini (dark green bar) now scored 60% higher than the Mid-2010 Mac Mini (light blue bar):

Apple Mac Mini: Historical Geekbench score
NOTE: In the above chart, the Mid-2010 Mac Mini 2.40 Ghz ("M10 2.40") is assigned the baseline score of 100%

Next: Stress Tests