Testing Methods:

After a recent editorial, we have decided to change the way we review motherboards here at bit-tech. We've moved our focus away from synthetic benchmarks that provide meaningless numbers from scenarios the consumer rarely finds themselves in.

With the exception of SiSoft Sandra's unbuffered memory bandwidth benchmark - which, incidentally, measures real memory bandwidth when you need it most - all of our benchmarks have been engineered to give you numbers that you are likely to find useful when actually using the products we have evaluated in the real world. There are plans to increase the number of benchmarks over time and we're running additional tests that will not be published until we have enough products to make a reasonable comparison.

We are also focusing a lot more of our time on evaluating the stability of the motherboards (and platforms) using a stress test designed to highlight any of the potential weaknesses that the product may have. That involves a gradually increasing amount of stress starting with Prime95 and expanding to IOMeter and 3DMark03 if all is well. This is to ensure that all parts of the system are stressed simultaneously over a period of time.

We believe that the consumer is never likely to subject their platform to this level of stress and we are not expecting every product to complete an entire extended stress test. However, most poorly engineered products fail within the first couple of hours, or even minutes, allowing us to make a conscious decision on whether a motherboard (or platform) is worth your money, regardless of how well it performs in our benchmarks.

Test Setup:

The Epox EP-8NPA SLI is somewhat unique at the moment, so we have tested the motherboard in a slightly different way to how we normally would. The board is an SLI solution for the mainstream, so we have configured a mainstream system for testing the motherboard. We included results from our normal high-end test system to give you an idea of how much performance you get for your money.

AMD Sempron 3300+ (operating at 2000MHz - 10x200); Epox EP-8NPA SLI; 2 x 512MB Corsair Value Select PC3200 (operating in single channel with 2.5-3-3-7-1T timings); BFGTech GeForce 7800 GT OC (operating at 425/1050MHz); Maxtor DiamondMax 10 250GB 7,200RPM SATA 150 Hard disk drive; Windows XP Professional Service Pack 2; DirectX 9.0c; NVIDIA ForceWare 81.94 WHQL.

AMD Sempron 3300+ (operating at 2000MHz - 10x200); Epox EP-8NPA SLI; 2 x 512MB Corsair Value Select PC3200 (operating in single channel with 2.5-3-3-7-1T timings); XFX GeForce 6600 DDR2 (operating in both single card and SLI modes at clock speeds of 400/800MHz); Maxtor DiamondMax 10 250GB 7,200RPM SATA 150 Hard disk drive; Windows XP Professional Service Pack 2; DirectX 9.0c; NVIDIA ForceWare 81.94 WHQL.

AMD Athlon 64 FX-57 (operating at 2800MHz - 14x200); VIA K8T900 Reference Motherboard; 2 x 512MB Corsair 3200XL Pro (operating in dual channel with 2.0-2-2-7-1T timings); BFGTech GeForce 7800 GT OC (operating at 425/1050MHz); Maxtor DiamondMax 10 250GB 7,200RPM SATA 150 hard disk drive; Windows XP Professional Service Pack 2; DirectX 9.0c; NVIDIA ForceWare 81.94.

AMD Athlon 64 FX-57 (operating at 2800MHz - 14x200); DFI LANParty nF4 SLI-DR (nForce 4 SLI); 2 x 512MB Corsair 3200XL Pro (operating in dual channel with 2.0-2-2-7-1T timings); BFGTech GeForce 7800 GT OC (operating at 425/1050MHz); Maxtor DiamondMax 10 250GB 7,200RPM SATA 150 hard disk drive; Windows XP Professional Service Pack 2; DirectX 9.0c; NVIDIA ForceWare 81.94

Memory Performance:

Obviously, the Epox motherboard only supports single channel memory, by virtue of using Socket 754. The board appeared to have slightly more memory bandwidth when a single video card was used, around 1% more in real terms. You're not going to see dramatic performance losses when using two 6600 DDR2's in SLI.

Stability & Overclocking:

The board was very stable throughout our testing, even when we stress tested the motherboard with Prime95, I/OMeter and 3DMark03 simultaneously using two GeForce 6600 DDR2 in SLI mode. Interestingly though, we found that the board would consistently fail our stress testing after the 21 hour mark. We ran our stress test before updating to the latest available BIOS, and then after updating. On both occasions, we found that 3DMark dropped back to the desktop with an error message after just over 20 hours 50 minutes.

Regardless of this problem, the motherboard was very stable. Our stress test is designed to be exactly that and there are very few enthusiasts who are likely to put their motherboard through the same level of abuse for a continuous period of time. Although the motherboard is stable, it is not quite as stable as some of the motherboards we have seen from the usual suspects.

We overclocked the motherboard to 290MHz FSB when setting the CPU multiplier to 8x and dropping the memory back to a 3:2 memory divider to ensure that it didn't become an issue when overclocking. The board was very stable at 290MHz FSB, and we suspect there might be more in it using a CPU with better memory controller than the one on ours.