Shuttle SG33G5: Black Beauty or Black Sheep?

Overclocking

Remember that small margin of overhead discussed in the Power section? It's about to be tossed out the window. To recap, a few issues hinder performance and overclocking capability in this small form factor system. The primary limitation is the 250W power supply. Even at stock speeds, we've already seen that the system is practically at maximum specified draw.

A secondary limitation is heat. It's not practical to use alternate cooling solutions in a system like this (though if you're determined it's possible). The temperature measurements don't seem to indicate that will be a huge problem. The third potential limitation is the lack of control over the memory divider, which might artificially limit the system. If you're utilizing DDR2-667, it's already at full speed at a 266MHz FSB with the 4:5 divider.

For overclocking, the first step was to reduce the multiplier and see how high we could push the FSB.

The maximum we achieved was 340MHz. We could run the memory in either 4:5 or 2:3 in this configuration and it was stable. Thus a 510MHz threshold was established for the memory, which removes it from the equation (at a 4:5 setting, the memory could surpass a 400MHz FSB). 350MHz failed to pass the Windows boot screen at stock voltages.

Next, we dropped the FSB to 320MHz and raised the multiplier back to 9. This 2.88GHz speed was stable at stock voltages. We ran the UT3 CTF benchmark, Suspense, at this speed.

This mild overclock yields a solid 27 FPS performance increase in UT3. As previously mentioned, however, we saw no improvement at all in Crysis, as the GPU is clearly the bottleneck for that title.

So what about that margin we had? We monitored power during the UT3 run, and it spiked as high as 231W. We then stressed the system using CINEBENCH multi-render and 3DMark06. The result was a 260W average draw. Danger, Will Robinson! At this clock speed with the 8800 GT, we're exceeding the maximum power rating of the supply - at least at the outlet. Hmm… maybe just a little more peril?

We raised the FSB to 340MHz with a multiplier of 9, for a resulting clock speed of 3.06GHz. The system would boot into Windows and run Prime95 for a short period, but CINEBENCH and 3DMark06 caused a fatal crash. Increasing the CPU voltage by 100mV stabilized the system. However, when CINEBENCH and 3DMark06 stressed the system to 100%, the power draw immediately spiked to 286W. If we figure efficiency at around 80%, the 250W PSU is putting out 230W, but without precise control over where this power is going, we felt it was prudent to stop the test.

All the components (CPU, memory, and motherboard) were capable of stability at this speed and likely higher, but running a 250W power supply at such a load for prolonged periods is unlikely to prolong component life. In fact, our experience with many previously tested SFFs is that motherboard and PSU problems are common even without overclocking after 12-18 months. We certainly don't recommend overclocking on SFFs unless you're willing to risk failures - and such failures appear to be common given enough time, unlike many desktop systems that we've run overclocked for years on end without difficulty.