I couldn’t find much on the interwebs about overclocking Zotac’s Ion motherboards. This was especially true with systems running Ubuntu Linux. Since I couldn’t find many in-depth results of such experimentation, I’ve decided to post some of my own. This isn’t the most accurate or objective test, but maybe these results will help someone else with their own experimentations.

First, the baseline system:

• Zotac IonITX-L-E motherboard – Intel Atom N330 dual-core CPU @ 1.6Ghz
• 2x 2Gb sticks of Crucial DDR3 1.5V 1066Mhz memory
• Intel X25-V 40Gb boot drive, formatted ext4
• WD3200AAJS 320Gb media drive, formatted ext4
• FSP 220-60LE PSU

All benchmarking and stress-testing was done with both the D-Sub and DVI video ports outputting @ 1920×1080. The operating system is Ubuntu 10.04, single-boot. Benchmarking was performed by running Benchmarks Module 0.5.1 inside Hardinfo 0.5.1. Short-term stress-testing was done by concurrently opening 8 tabs each in Firefox and Chromium (including one each playing flash video), 4 separate 3840×1080 images in GIMP, Rythymbox, and System Monitor. I would then perform a series of random transformations on one of the open images and spin Compiz through it’s paces. The stress-test is fairly unscientific, but accurately simulates the worst real-world condition I’d ever expect to personally encounter.

To be realistic, I wasn’t ever expecting huge performance gains. These systems aren’t that powerful to begin with; their appeal is in their small size, low power consumption, cool running, and simple set-up. I think of them as being laptops unencumbered by laptop packaging.

First, the baseline results:

All BIOS settings stock: FSB 533Mhz, 1.6Ghz clock, everything “Auto”

CPU Blowfish – 8.036
CPU CryptoHash – 103.652
CPU Fibonacci – 8.254
CPU N-Queens – 44.511
FPU FFT – 7.763
FPU Raytracing – 73.719

Stable and cool. CPU cores at 30-32 Celcius, GPU at 45 Celcius.

Now, some incremental little bump-ups. Except where noted, for all the tests “CPU-BSEL Select” was left on FSB-667,the System clock was Unlinked, and the voltages were unchanged. Percentage gains are relative to stock.

FSB=550Mhz, 1.65Ghz

CPU Blowfish – 7.694 (+4.26%)
CPU CryptoHash – 118.895 (+14.71%)
CPU Fibonacci – 8.008 (+2.98%)
CPU N-Queens – 43.464 (+2.35%)
FPU FFT – 7.533 (+2.96%)
FPU Raytracing – 71.861 (+2.52%)

Stable and cool. CPU cores 30-32C, GPU 45C. No perceptible real-world performance increases.

FSB=600Mhz, 1.8Ghz

CPU Blowfish – 7.036 (+12.44%)
CPU CryptoHash – 128.359 (+23.84%)
CPU Fibonacci – 7.349 (+10.96%)
CPU N-Queens – 39.064 (+12.24%)
FPU FFT – 6.896 (+11.17%)
FPU Raytracing – 66.873 (+9.29%)

Stable and cooler. CPU cores 20-28C, GPU <45C. Not much perceptible real-world performance increase.

FSB=630Mhz, Memory Voltage 1.55, 1.889Ghz

CPU Blowfish – 6.764 (+15.83%)
CPU CryptoHash – 128.659 (+24.13%)
CPU Fibonacci – 7.000 (+15.19%)
CPU N-Queens – 37.637 (+15.44%)
FPU FFT – 6.896 (+11.17%)
FPU Raytracing – 66.873 (+9.29%)

Stable and even cooler. CPU cores 18-25C, GPU <45C. Feels a little “snappier”, but so hard to quantify. This is where I am set as I write this, a fall-back position from the next configuration listed.

FSB=640Mhz, 1.921Ghz

CPU Blowfish – 6.660 (+17.12%)
CPU CryptoHash – 133.800 (+29.09%)
CPU Fibonacci – 6.919 (+16.17%)
CPU N-Queens – 37.667 (+15.38%)
FPU FFT – 6.470 (+16.66%)
FPU Raytracing – 62.602 (+15.08%)

Benchmarked and stressed okay, but spontaneously crashed 12 hrs later under lighter load. CPU cores 18-25C, GPU 45C. I would like to return to this setting and toy with voltages to see if I can get that long-term stability back.

FSB=645Mhz, 1.935Ghz

CPU Blowfish – 6.579 (+18.13%)
CPU CryptoHash – 133.600 (+28.89%)
CPU Fibonacci – 6.819 (+17.39%)
CPU N-Queens – 37.258 (+16.29%)
FPU FFT – 6.414 (+17.38%)
FPU Raytracing – 63.506 (+13.85%)

Benchmarked but failed on stress, just as the second browser was opened. Notice the drop in Raytracing improvement… seems like we’re pushing some limit here. Still remarkably cool; CPU <30C.

FSB=645Mhz, CPU voltage +0.0125V, Memory 1.55V, 1.935Ghz

CPU Blowfish – 6.760 (+15.88%)
CPU CryptoHash – 131.793 (+27.15%)
CPU Fibonacci – 6.843 (+17.09%)
CPU N-Queens – 36.836 (+17.24%)
FPU FFT – 6.413 (+17.39%)
FPU Raytracing – 62.311 (+15.47%)

Benchmarked okay, then failed under stress as the large GIMP transformation was being initiated. Raised CPU and memory voltages in search of stability, but notice that 4 of the 6 benchmarks actually worsened. Still running cool.

FSB=648Mhz, 1.943Ghz

CPU Blowfish – 6.581 (+18.11%)
CPU CryptoHash – 135.166 (+30.40%)
CPU Fibonacci – 6.813 (+17.46%)
CPU N-Queens – 37.446 (+15.87%)
FPU FFT – 6.426 (+17.22%)
FPU Raytracing – 61.938 (+15.98%)

Benchmarked okay, but failed under stress. Benchmark increases across the board, but no useful stability. Still cool.

FSB=650Mhz, 1.95Ghz

Failed to complete the benchmark; froze on FPU FFT, then crashed.

FSB=650Mhz, CPU Voltage +0.0125V, Memory 1.55V, 1.95Ghz

CPU Blowfish – 6.906 (+14.06%)
CPU CryptoHash – 127.076 (+22.60%)
CPU Fibonacci – 6.763 (+18.06%)
CPU N-Queens – 37.551 (+15.64%)
FPU FFT – 6.372 (+17.92%)
FPU Raytracing – 63.266 (+14.18%)

Benchmarked okay, then failed on stress. Notice the performance losses relative to stable slower speeds.

FSB=650Mhz, CPU Voltage +0.0375V, Memory 1.55V, 1.95Ghz

CPU Blowfish – 6.912 (+13.99%)
CPU CryptoHash – 121.692 (+17.40%)
CPU Fibonacci – 6.789 (+17.75%)
CPU N-Queens – 36.865 (+17.18%)
FPU FFT – 6.400 (+17.56%)
FPU Raytracing – 63.534 (+13.82%)

Benchmarked okay, but failed on stress into an uninterrupted boot loop. Some benchmarks better, some worse, and no stability. I went on to try a 660Mhz and 667Mhz FSB, with a few modest combinations of CPU and memory voltage, but all failed to boot, or rapidly failed during benchmarking.

There are probably some sorts of accompanying memory overclocks, along with all the latency settings, etc. to be tried. What I’ve read tends to suggest that such things are just trading performance for stability, when it seems like sticking to a “safe” speed in the first place is wiser. I may have to invest in two sticks of DDR3-1333, which this motherboard purports to support, and see what that brings about.

Overall, I have to conclude that 1.9Ghz really is the sweet spot here. I want to keep testing in the 635-640Mhz FSB range, playing with voltage and memory controls until I have the stability I require. While other testers with more skill than I are reporting useful 2.0Ghz clock speeds and even 2.2Ghz with nitrogen-cooled cores, it seems like speeds above 1.9Ghz offer diminishing returns and decreased stability for the real-world air-cooled average user. In any event, nobody’s saying exactly how they’re getting these “stable speeds”.

The real sweet surprise has been the low operating temps! Having run at 1.889Ghz for 24 hours now, my cores report 18-22C, and the GPU remains planted at 43-45C. I can’t help but believe that the superior airflow management of my unusual casemod is helping here.

Has anyone else out there been playing with these boards and the current Zotac BIOS? I’ll continue to test and post my results; I’d sure like to hear from anyone else working on the same.