Quad core performance

After our little dual core adventure, it's time to have a look at Quad core performance. For this test we plugged in our trusty QX9650, a Yorkfield CPU, the first based on the 45nm manufacturer process. The 45nm quad cores are notoriously harder to overclock to higher FSB values than their 65nm counterparts (like the QX6700/6800). What's more, reaching 500FSB on any quad core has proven to be almost impossible, except maybe on the 790i ultra chipset from nVidia. After the rather disappointing dual core score we started off with modest expectations. We used the same "829" bios that we used on the dual core (a more recent bios has become available lately, but according to early feedback it hasn't changed a lot).

Anyway, have a look at the maximum FSB we reached :



Another rather disappointing result you might say. In a way, yes, but this cpu has never been a real "FSB King" on any motherboard. I actually struggled hard to get it past 440FSB on my Asus Blitz too, so the result isn't that bad. Luckily this extreme edition cpu has an unlocked multiplier, and with more extreme cooling we might squeeze out a little higher FSB. This does mean however that a lower multi quad core might not be the best investment to plug into this board. If you want to hit 4Ghz, you'll need at least a Q9650.

One reason why overclocking the quad core went so difficult is the fact that we had to start from a blank sheet. Until now, and as far as I know, no review or test has been published using a quad core cpu and this exact board. Therefore, no settings and especially no GTL-ref settings were available to us when testing. Thus it might be very well possible that better GTL-ref values are available than the one I used, giving way to a little more headroom in the FSB department. The values I used were 96/93/75, which are the same actually than the ones for the E8600. This is very strange, as quad cores tend to want higher values (dual cores : ±63% of VTT vs quad cores ±67-69%). The exact table for GTL-ref voltages on the DFI X48 board can be found here (.xls). I tried all kind of GTL settings, but these were the best for me. This is a typical "trial and error" setting so you'll have to try your own luck.

As far as the other settings go, I tuned especially at 1.34 to 1.35vtt, Northbridge voltage I kept below 1.65Volts, but above 1.55V. Southbridge (PLL) also up a notch. This way I could work my up to 430FSB, but had to tune it down a little to get the combo completely stable. Another interesting setting was the Performance level (tRD), because the motherboard responded a little different to it depending on the CPU multiplier. Below are some screens of PC-Mark Vantage and Superpi 32M, the last one on two different tRD settings depending on the multiplier.


The Quad core in action, click to open


Memory performance

Time for my favorite part in any motherboard review: the memory section. As a bit of a memory enthusiast, I've been doing many DDR2 reviews over the last couple of years, and my first DDR3 review just recently. I made a promise in that last review to see what these modules are capable of on this board, so here we are.

Without loosing too much time, let's see what we got :



Ah, the sweetness! 500FSB and 2Ghz on the memory. Watching it in a simple CPU-Z screen makes it look so easy doesn't it. Unfortunately, things are never as easy as they look, and you'll probably know what I mean when you've read the previous page. Although the 1:2 memory dividers was working great, I struggled for several weeks to get this setting to any kind of stability. The above screenshot actually is one dating back to one of the earlier biosses, but I couldn't get the setup stable at that time. The next cpu-z screen was completely stable though:



Again, the bios "825" improved things considerably, and we finally got reasonable stability as the following screenshots show:


memory at 2ghz, click to open

The last thumbnail shows the problem though, as I couldn't get the setup Orthos-stable at 2ghz, no matter what I tried. This is also the reason we couldn't clock our dual core higher than 500FSB on the 1:2 setting, although it was easier to work with than 3:5. These were the best settings for me :

memory timings : 8-8-8-30-2T or 9-9-9-30-1T

vdimm : 1.91V (real)

vCore : 1.30V (up to multi of 8)

vNB : 1.63V

VTT : 1.35V (E8600)

gtl-ref enabled at 96/93/75

performance level (tRD) : 8 or 9 (slightly lower vNB)

clock drive strength : 3 or 6

data drive strength : 8

skew model CH1+CH2 : 5 (Samsung based memory)

I also tried some micron based memory in the form of OCZ's PC3-12800 EB kit, which is a 1600Mhz kit with standard timings of 7-6-6-22. For this memory, the skew model of 5 results in a simple "C1" aka "no boot" situation, but with a skew model of 3 they work perfectly. They also scale up to about the same level as the PC3-16000 Flex kit, albeit they needed a little more voltage.

Last but not least, let's look at 4-module operation. Unfortunately I didn't have any 4Gb DDR3 kits available, so 4x 1Gb will have to do (which obviously isn't the most taxing setup that is possible, but the most taxing I could apply) :



Again, at first only cas 9 seemed to be stable, with a performance level (tRD) of 9 as well. In the end, even a tRD of 9 was not stable enough to run four instances of prime 95, so I had to relax to a rather high tRD of 10 to get the four sticks completely stable. I could tighten the timings a bit to 8-8-8-24 though. In the thumbnails below you can see a superpi 32M test, and over an hour of prime95 on these settings. As a final note, a command rate of 1T was not possible when the board was equipped with 4 memory modules. This is not entirely surprising, but I did manage this with DDR2 on the nVidia 680i not long ago.


4 sticks of memory at 1600mhz, click to open


Let's wrap it all up on the last page ->