Our tests have proven the results we have obtained on Asus Striker Extreme mainboard: the maximum performance is provide at 400MHz FSB, then the performance drops and gets back to the same level only at 460MHz. In other words, this behavior is typical not of a particular mainboard, but of all mainboards on Nvidia nForce 680i SLI chipset: it doesn’t make sense to overclock beyond 400MHz FSB, if we can’t hit 460MHz. Unfortunately, I cannot check how the boards would behave at higher frequencies because of the limitations set by our today’s hero, abit IN9 32X-MAX Wi-Fi. In conclusion I would only like to add that you should try to avoid not only the FSB frequencies between 400 and 450MHz, but also 325MHz FSB, and 360MHz-380MHz interval. Too many limitations for the top-of-the-line chipset, don’t you think so?

When we overclock CPUs on Intel P965 Express based mainboards we have to use the minimal 1:1 divider for the memory frequency, and the memory is working synchronously with the FSB in this case. However, Nvidia nForce 680i SLI chipset has one interesting peculiarity: it allows clocking memory almost independently of the FSB. I am sure that a lot of users will take advantage of this opportunity, so how will the performance change in this case?

To check this out I performed another series of tests in EVEREST at the same FSB frequencies, but the memory frequency in this case was locked at 800 and 1000MHz. I selected these particular values for a reason. According to our tests, even PC2-4200 memory intended for work at 533MHz can overclock to 800MHz, not to mention PC2-5300 working at 667MHz. as for PC2-6400 modules, not all of them can hit 1000MHz, but the best ones can.

Since the memory frequency was fixed at a certain value, and we were measuring its performance, we were expecting the graph to be more like a straight horizontal line, because only FSB frequency was changing. Of course, if there would be no additional surprises from Nvidia nForce 680i SLI chipset.

There were hardly any surprises this time, only at FSB frequencies below 300MHz the memory running at 800MHz works faster than the memory running at 1000MHz. The dividers for 1000MHz could be not the best ones in this case. The difference could actually be even greater if we had set more aggressive timings for 800MHz memory, which was actually quite possible to do.

We again see a smaller performance drop that is smoothed out by the constant memory frequency. It occurs at 325MHz FSB. The second time performance drop is detected is at frequencies around 400MHz FSB. This drop is much more noticeable for memory running at 800MHz, but it is obviously explained by the fact that the mainboard cannot always detect the memory frequency we are setting: sometimes it is a little lower, and sometimes a little higher than the desired, which is shown on the graph.

You can easily see that the performance drops only at those FSB frequencies when the memory frequency is lower than the set one, and the least successful frequency in this case is 380MHz: the memory frequency is minimal in this case and equals 760MHz instead of 800MHz.

I would like to say once again that the obtained results do not demonstrate the actual system performance because the memory timings are fixed. These results are only interesting for the sake of studying the chipset behavior at different frequencies. If we consider the mainboards on Nvidia nForce 680i SLI chipset in real working conditions, then synchronous memory clocking will look much more preferable. During our tests the CPU was overclocked without losing stability to the maximum that abit IN9 32X-MAX Wi-Fi allowed, i.e. to 400MHz FSB. The memory was working synchronously at 800MHz with 4-4-4-12-1T timings. If we increased the memory frequency, the results worsened, for instance at 854MHz with the same timings. Asynchronous mode requires additional coordination and increases latencies. We only managed to hit the same performance at 1100MHz with the timings set to 5-5-5-15-2T.

As a result, I would advise Nvidia nForce 680i SLI based mainboard owners to perform a number of additional tests with different FSB and memory frequencies once the CPU has been overclocked to its alleged maximum. It is quite possible that you may get higher level of performance with different combinations of FSB and memory frequencies even though formally the bus or memory frequencies are lower.