Testbed and Methods

To make our review complete we tested the performance of the Gigabyte GA-K8NXP-SLI in our standard set of benchmarks. However, I have to point out that these tests hardly make that much sense. The memory controller of Socket 939 systems is integrated into the CPU, so the mainboard makers can hardly seriously affect the system performance. In fact, the only thing the memory makers can really do is slow down their product by setting the wrong memory timings in the BIOS. Gigabyte GA-K8NXP-SLI didn’t have this issue, so I don’t think its performance is going to be very much different from that of most other NVIDIA nForce4 based products.

Nevertheless, we may still detect certain performance difference during overclocking. As we have already mentioned before, Gigabyte GA-K8NXP-SLI is not very stable with Command Rate set to 1T. As a result we have to set this parameter to 2T in order to achieve the same clock rates as the other mainboards allow. So, our hero will be slightly slower during overclocking than some of its competitors.

To illustrate what I have just tried to explain we decided to compare the performance of Gigabyte GA-K8NXP-SLI against that of another Socket 939 mainboard that has already become a reference point in our performance analysis of Socket 939 platforms. I am talking about DFI LanParty UT NF4-Ultra-D mainboard. For our tests we assembled the following testbeds:

• CPU: AMD Athlon 64 3800+ (Socket 939, 2.4GHz, 512KB L2, Venice);

• Mainboards:
• DFI LanParty UT NF4 Ultra-D (Socket 939, NVIDIA nForce4 Ultra);
• Gigabyte GA-K8NXP-SLI (Socket 939, NVIDIA nForce4 SLI);
• Memory: Corsair Twinx CMX1024-3200XL (2x512 Мбайт, 2-2-2-10);
• Graphics card: PowerColor RADEON X800 XT (PCI-E x16).
• Storage subsystem: Maxtor MaXLine III 250GB (SATA150).
• OS: Microsoft Windows XP SP2.

In the overclocked platforms the CPU frequency was increased to 2.7GHz. The clock generator frequency was set to 270MHz, and the clock multiplier – to 10x. The processor Vcore was set to 1.5V on both mainboards. As for the memory frequency, it was set to 192MHz, because lower memory latency is much more important for Athlon 64 systems than high memory bandwidth. Thus we assigned the most aggressive values to all memory timings. Nevertheless, even with these memory settings we couldn’t use Gigabyte GA-K8NXP-SLI with Command Rate 1T.

In the nominal mode we ran all the tests with the Command Rate 1T.