Gigabyte GA-8N-SLI Quad Royal Motherboard

System Stability Holding up under pressure

Placing four full PCI-e x16 slots on one board is no simple feat of engineering, it requires the ability to manage and power up to four or five power-hungry devices and keep them from conflicting with each other.  Hardware and BIOS options allow the user to define their setup, but running them all could be another thing entirely.  As such, the first thing that needs to be checked is the Power Supply Unit.  The manual suggests a PSU that supplies at least 500W, but our gut feeling is that you'll probably need more than that since SLI configurations alone call for that spec.  For our testing, we tried setting up a four GPU configuration consisting of two 7800GTs in SLI, a Radeon X1800XL and an additional X800XL (yes, you can mix and match NVIDIA and ATi in the same case!)

The results were a little puzzling; we could install and use the three higher-end cards requiring external power connections, but not the bus-powered X800XL.  Our first inclination was that there was a power problem, but we found that we couldn't run it as the single-card in the system either.  As such, we were left with running the system with just the two 7800GTs in SLI and the X1800XL.  We did not run into any problems running all three cards with a 500W Antec PSU in general operations.

We did, however, have some issues with the system stability overall that may or may not be solely Gigabyte's problem.  First, while we had one of the 7800GTs in PEG Slot 2, we found some instability running 3DMark05 that wasn't present had we installed the same card and drivers in Slot 1.  The scores obviously reflected the difference in performance, shifting from 16 lanes to 1, but while it was rock solid at the slower rate, the benchmark typically crashed at the faster rate.  Reinstalling the application, and even reinstalling the OS and 3DMark05 by itself, did not correct the issue.  On a somewhat less concerning note, we also ran into problems when switching between systems while using a KVM box - a problem that has almost never reared it's head before.  Often, switching to another system and returning would leave the GA-8N-SLI completely frozen and unrecoverable.  We tried isolating the problem by limiting the connection to just the video card, and this seemed to help to a degree, with far fewer lockups.

Overclocking Tools Time for a tune-up

One of the bundled applications that came with the GA-8N-SLI is a comprehensive utility dubbed Easy Tune 5.  It supports many of the same features that we first encountered in the BIOS, including C.I.A.2, PC Health, and M.I.T., and gives the user a Windows-friendly graphical view of these settings.  The first time we used it, we didn't make any changes and clicked on the 'GO' button, which promptly caused Windows to crash.  Subsequent usage didn't prove to be as problematic, but while great for viewing information, we didn't have the greatest overclocking results, so we headed back to the BIOS for another round.

Our first step was to unlink the CPU and Memory, to limit any problems that overclocking the DDR2 RAM would have on the system.  We are quite convinced that these sticks are able to get higher than their original speeds, getting as high as 798MHz effective, but we wanted to remain on the safe side and left it at 667MHz.  The CPU was another story: we've typically taken this P4 550 CPU from the stock speed of 3.4GHz and reached just over 4.0GHz using a stock Intel cooler.  With the various voltage options on the GA-8N-SLI, including CPU, memory, NB, SB, and FSB voltage settings, we had hoped for roughly the same results...

SANDRA CPU Results Overclocked to 3.84GHz	PCMark05 Details with CPU Overclocked to 3.84GHz

The final overclock, however, was well lower than what we had expected.  Initially, we shot forward, reaching an FSB of 220MHz without any problems whatsoever.  At this point, we needed to raise the CPU voltage up to 1.425V to ensure some stability.  Moving up from that speed to the 225-230 range proved perilous.  225MHz (900MHz effective for a quad-pumped P4) was attainable by putting the CPUVolt at 1.45V without changing any other voltages, but getting any higher resulted in an unusable system.  We were able to boot into Windows at 910MHz, but could not run any benchmarks, even after raising NB, SB, and FSB voltages up 0.2V each.  Thus, we were left with a small overall overclock of only 25MHz for the FSB, with the P4 running at 3.84GHz.  There were obvious benefits, as seen in the jumps in the SANDRA and PCMark05 scores, but the feeling was that we left some performance out there on the playing field.