CPU Overclocking

We performed out overclocking experiments in an open testbed built with the following components:

• Intel DX38BT mainboard;
• 2x1024MB Super Talent W1600UX2G7 DDR3-1600 SDRAM;
• NVIDIA GeForce 8800 GTS 320MB graphics card;
• Seagate Barracuda 7200.10 ST3320620AS HDD (7200rpm, 16MB, SATA 320GB);
• Zalman CNPS9700 LED CPU cooler;
• Antec NeoPower HE 550 PSU (550W).

According to the package, Intel DX38BT mainboard is first of all designed for Intel Core 2 Extreme processors that is why we chose this particular CPU for our overclocking experiments: Intel Core 2 Extreme QX6700 (2.66GHz, 266MHz FSB, 8MB, Kentsfield rev.B3). This particular sample can overclock to 3.4GHz with Vcore increased to 1.44V, but if we use x10 clock frequency multiplier, we will be able to achieve this speed at 340MHz FSB already. This is actually a very simple task, even non-overclocker mainboards can do it. To make things a little more interesting, we dropped the processor clock frequency multiplier to x7. So, we reduced the clock multiplier, set the minimal memory frequency of 800MHz, increased processor Vcore, Vmem, Vchipset and FSB bus voltage, set the FSB at 450MHz. However, Intel DX38BT mainboard refused to boot in this case.

I have to say that Failsafe WatchDog function works perfectly well. After a few failed booting attempts, Intel DX38BT mainboard restarts on its own with safe BIOS settings, stops after POST and offers to access BIOS Setup to correct the parameters. In this case the BIOS settings to not get reset to their nominal values. This work mode is very convenient when searching for the most optimal parameter combinations during overclocking. Sometimes the board would start looping the failed boot-ups, but to resolve this issue we could simply reset the config-jumper to access the BIOS and make all necessary changes without losing the previously set parameter values. It is much more convenient that using a Clear CMOS jumper offered by many other mainboards out there.

It is interesting that if you access the BIOS after a failed boot-up and just confirm the settings without changing anything, Intel DX38BT mainboard will boot and load Windows normally. However, the FSB frequency will for some reason drop down to 322-325MHz. And after restart, the board will not be able to boot again. This phenomenon got my attention, so I lowered the FSB speed to 325MHz and started the OS normally. I was very surprised to see that the memory frequency equaled 650 (1300) MHz. Of course, the mainboard couldn’t boot with higher FSB frequencies because the memory speed was way too high. It was my honest mistake: I set the memory frequency to its minimum of 800MHz. However, the CPU with default 266MHz FSB uses an increasing divider, so memory would turn into a limiting factor preventing further overclocking because it simply cannot work at frequencies beyond 1600MHz. So, actually, the mainboard has nothing to do with it.

Now let’s leave the memory speed at 800MHz, but set the Reference Frequency parameter to 400MHz. In this case the FSB:Mem ratio should be 1:1. To check it out we restart at a guaranteed operational FSB frequency of 325MHz. The result is that the memory frequency equals 325 (650) MHz. Now it is working the right way. So, we change the FSB frequency to 450MHz and the board would boot again! Yes! Unfortunately, we were celebrating too soon: it turned out that the mainboard dropped the FSB speed down to 325MHz again.

I don’t think it makes a lot of sense to go into details regarding every failed overclocking attempt we undertook. We reflashed the BIOS with the newest 1217.1401 version, played with processor clock frequency multipliers, checked out different Memory Frequency and Reference Frequency combinations, but we could only get one of the following: failed boot-up and an offer to correct the BIOS settings or system boot-up at the mainboard’s favorite 322-325MHz FSB. Surprisingly, but Intel DX38BT mainboard could still load Windows Vista OS even with the processor’s nominal x10 clock frequency multiplier and relatively low 340MHz FSB frequency, although the errors popped up immediately when we launched Prime95 utility for stability run. We have recently criticized ASRock 4Core1333-eSATA2 mainboard for the budget segment for its poor overclocking friendly features, although, it could easily clock the CPU to 340MHz FSB. However, our today’s overclocker solution - Intel DX38BT – cannot do even that.

Ok, if the board likes FSB frequencies in the interval from 322 to 325MHz, then be it. Let’s try to make its life as simple as possible. Our Intel Core 2 Extreme QX6700 processor can work at 325MHz FSB with its nominal x10 clock multiplier without increasing the core voltage above the nominal 1.35V. We have already seen many times that Intel DX38BT mainboard can boot at this speed without any problems. This time it also did and even passed the stability check with Prime95 utility. However to ensure acceptable performance we have to use increasing multipliers even for DDR2 SDRAM at this low FSB frequency, not to mention the DDR3 SDRAM, where only high frequencies can make up for high latencies. Unfortunately, all our attempts to get the memory to work at high frequencies failed.

In this case we assumed that Intel DX38BT mainboard cannot succeed at overclocking because we are using an engineering sample of Intel Core 2 Extreme QX6700 processor. So we decided to repeat our experiments with a regular Intel Core 2 Duo E6300 (1.86GHz, 266MHz FSB, 2MB, Conroe-2M, rev.B2). This CPU can work stably at 490MHz FSB with the Vcore increased to 1.45V. However, all our attempts to get it to run at this speed on Intel DX38BT mainboard were vain.

Overclocking, like any other good occupation, should be fun. Unfortunately, it causes a lot of frustration when you try to do it on Intel DX38BT mainboard. After a number of failures you start perceiving the logo from the mainboard package and numerous marketing materials differently. Maybe it says something like that: Intel DX38BT mainboard will eat your brain and if you don’t give up overclocking you may end up dead :)

However, we still had a little chance to rehabilitate the board a little bit by trying to overclock a CPU from Windows using some of the available software tools. We are going to discuss this topic in the ongoing chapter of our article.