The real deal

Time to have a look at the pages where the action is going to be, and as always the action starts in the bios page with the very cryptic name of "genie bios setting". No need to worry though, there's no ghost involved, but you do need some magic from time to time to get the most out of it. This page looks the same by the way on most current DFI motherboards, so if you're familiar with a recent DFI board, this page will look familiar as well.



The settings you're looking at are the main OC settings, such as front side bus settings ("cpu clock"), the memory divider (in this case 400/1333 or 3/5) and the speed of the PCIe slots. On the above screen you can also find some more exotic settings, which are not as common on non-enthusiast motherboards. The CPU clock amplitude is one of them, but more important maybe is the clock skew. Tuning the clock skew can definitely increase the FSB range of your CPU a bit, but it needs a little trial and error to get it right. I personally didn't touch the amplitude, but a clock skew of 100 to 200ps gave me another 5 to 10Mhz on the FSB.

The above menu is also the place where the other CPU, memory and voltage submenu's can be accessed. The CPU menu is pictured in the thumbnail below, to the right. Here you can enable or disable the power settings for the CPU, and shut down on or more of its cores if you like.



The middle thumbnail above is the "CMOS reloaded" screen, another "golden oldie" from DFI. This gives you the possibility to save the settings of your overclock, to return to a certain set of bios options quickly, for example when you run into trouble. Running into OC trouble on this board isn't too hard, honestly, so you'll find yourself returning to this menu maybe more often than you'd like ;-)


Some memory settings...

The memory screen is where the hard work starts. With dual core processors, getting to 400 or even 450FSB nowadays isn't too hard. In some cases, typing the number into the appropriate box, together with lowering the cpu multiplier a notch, should be enough to get things right very quickly. It's taking your CPU beyond 450FSB that takes a little time on most boards, and this board certainly is no exception. With this board release, another thing happened though: DDR3. DFI had to develop a different memory tuning section for this release, as it's the first DDR3 board in their lineup. DDR3 is a quite different animal when it comes to overclocking, as chipsets react in a different way to DDR3 than DDR2. First of all, frequencies are up, and latencies as well. Second: dividers are different to match the higher frequencies. Third: calculation of the memory sub timings (which most of the time are set to auto, or are not present in the bios and thus non-adjustable) is different. All of this makes for a very different overclocking experience, something I already encountered when reviewing the OCZ PC16000 kit lately (on the Asus P5Q3 that was).

In my opinion, DFI made two choices on the bios of this board that could explain some of the difficulties that we encountered:

1. They gave us (the user) too many options to play with
2. They didn't differentiate the bios release enough from the DDR2 counterparts.

The first point actually is something that DFI (and other manufacturers) have been doing for a while: with every major bios generation, options multiply. While enthusiasts (at first) loved this, it became increasingly difficult to understand all the different settings. More importantly, when you offer an option to tune a certain setting, the possibility of mistakes for the end user increases. While you could leave a lot of the settings alone, my guess (or educated feeling) is that some of today's new settings actually are not optimized at all when left to their "auto" setting. Being an enthusiast myself, it falls hard on me to say that DFI should, in my opinion, take things out of the user’s hands again and tune them on their own. Right now, DFI leave too much to the user and in some cases fail to deliver a working OC solution themselves. Of course, let me stress to say that on standard or stock levels (in other words: out of the box), everything works fine, and I didn't encounter any problem on a first boot with optimized defaults. The thing is : DFI's are not meant for people that run "optimized defaults", they are meant for much more, meant for overclocking. And this board makes that very hard.

The second point is a thing that struck me only recently when looking at the bios settings of the DDR2 variant of the X48 board. On the website OCXtreme, a guy nicknamed "Praz" has been doing a great job tuning DFI boards, while at the same time educating the rest of us. When looking at his results with both boards, which share the same bios for about 95%, one can only conclude that the DDR2 board is by far the better board, if you're judging by overclock performance of course. My guess is DFI took the DDR2 bios and adapted it for DDR3 memory. While they arguably did a great job already, the bios does not feel like it's a finished product. And indeed, our motherboard actually has gotten better with every bios release, albeit with some serious drawbacks.

So let's have a look at the memory settings:




second half of the memory screen - click to open

The important settings are on the top half of the memory screen, which is in the big picture. Of course we have the usual: CAS, RAS to CAS, RAS Precharge and TRAS, and right below we have the performance level or tRD setting. The "performance level" setting is actually a rather new setting, introduced on mainboards not much longer than a year ago, with the introduction of the P35/X38 chipset. The first time I encountered it was in the Memset application, where it has been present for a while. This "performance level" is a rather correct description of the thing it represents : PL determines the actual timings and latencies of the chipset, so by lowering the timings (and the performance level) you actually get better performance, by relaxing the timings (and increasing the performance level) you loose a little performance. When increasing the front side bus you can use two methods for maintaining system stability: increase the chipset voltage, or increase the performance level. The more relaxed the chipset timings, the higher it will scale front-side-bus wise. In other words you have to balance the chipset voltage, the front side bus and the tRD setting to get a successful overclock. A higher voltage for the northbridge can make a lower tRD stable, resulting in better performance, but as more voltage means more heat, you have to make sure you have adequate cooling for the chipset as well.

On this page Oscar has introduced some more settings though: we're talking about the "Enhance data transmitting", the "Enhance addressing" and the "T2 dispatch" settings. These have a slight influence on your memory performance, and in the DDR3 case, they can mean the difference between 1950Mhz or 2000Mhz. In the case of the 3:5 divider, the pictured settings are the only ones that work by the way. Another setting is the MCH ODT latency ("on die termination latency"). For some this actually gives the chipset a little more headroom, but honestly I haven't been able to tune it in any way. Finally, I found that on this board the tRFC setting and the "all precharge to active" setting can make quite a big difference too.


...and some more memory settings...

When you look at the memory screen above, you'll notice that there's another two subscreens of the memory screen. Have a look at the "clock setting fine delay" screen first:



As this is a bios screen I hadn't seen on any board so far, this was the first time I had to work with it. Had to is the right word here, as this screen prooved to be very important to DDR3 clocking. The first 4 lines are the most important:

dram clk drive strength or CDS

dram data drive strength or DDS

channel 1 DLL default skew model

channel 2 DLL default skew model

I'll start with the bottom two, as they are the straightest forward: these two settings set the skew type for the two different memory channels. When the memory modules are the same in both channels, these settings will mostly be alike. The skew model is different for different types of memory modules, for example Samsung based memory will be different from Micron based memory, and not all of the Micron-based memory uses the same Micron chips. Oscar has implemented 8 different skew settings, you'll have to find out which one's best for your memory. Usually, a setting of 0 to 4 is suited for Micron based memory, 5, 6 and 7 are for Samsung based memory. I tried both, for me 3 (OCZ 1600Mhz EB, Micron-based) and 5 (OCZ 2000Mhz Flex II, Samsung-based) worked best. A word of warning though: the wrong skew model will really mess up your memory overclock, and with it your FSB overclock (more on that later), so you'd better tune this right.

The clock and data drive strengths are another new setting, and these I found out the hard way. One golden rule: CDS has to be lower than DDS. Through trial and error guys like OPB (on ocxtreme), praz and some others found that a 6/8 combination works quite well. I found a 3/8 combination to work best for me, but again you'll have to find out yourself, and your mileage no doubt will vary.

The rest of the screen gives us some fine delay settings. When using the 1:2 memory dividers, I like to put this to the "more aggressive" setting, but on the 3:5 memory dividers, they had to be on auto or I got a no boot "C1" error.

Finally, on the bottom of this bios screen (see thumbnail below) we got the "command per clock" (called command to CS here) setting. This is a well known setting and can be set to 1 or 2 cycles. There's also an "auto" setting available but be careful, this one can set 1T without you knowing, resulting in another "C1". I might have liked to have this setting on the main memory bios screen, but as it's related to the fine delay settings, I guess Oscar put it in the right place.




As a last screen in the memory department, the thumbnail above to the right. This shows some "pull in" settings to the read delay or tRD setting. Simply put it's a way of fine-tuning the tRD setting, when your "performance level" is somewhere in between two values for one channel or the other. I mostly left this alone, but it could give you some stability on the edge of FSB tuning.


Voltage tuning

The last important subscreen of the "genie bios" screen is of course the Voltage department. It looks like this:



First up are the main settings: vcore (CPU VID and VID special add), dram voltage, Northbridge and Southbridge (+PLL) voltages. DFI still swears by the basic CPU voltage controls, enlarged by a "special add" control which increases vcore percentage wise. I actually forgot about the exact reason they started with it back in the days of nForce 4, but I believe it had to do with some kind of power saving (or speedstep) mode (when the multi is lowered the cpu can/could decrease its voltage to the basic CPU VID, dropping the VID special add). I don not know if this is still the case today. Last but not least you can set VTT here for added FSB headroom. Careful with your 45nm CPUs though: settings over 1.35V will harm your CPU, certainly in the long run.

After the basics come some more important features:



As you might remember, one of the most brilliant settings signed "Oscar Wu" has been the GTL-ref voltage settings. Most important with quad core cpu's, this settings have been a revolution for FSB-tuning. For dual cores GTL-ref also will make a difference, but it won't be huge. For me (and I'm not alone this time) the 96/93/75 settings worked best, even for my 45nm QX9650.

Further on this screen are some more fine-tune settings for the MCH (memory controller hub, or northbridge), but in my and others opinion these settings would best be ignored, giving too many openings for a lot of trouble. Oscar should, in my humble opinion, not offer any control over these features to the end user, but then again, I could be proven wrong in the long run.

Finally, we've got the obligatory vdroop control, which after Kris Boughton's excellent article on Anandtech I'd rather leave alone, and we've got some more goodies in the form of a clockgen voltage control (never touched it) and the "host slew rate", which sometimes made a small difference when maxing out the front side bus.

Last but not least, I don't want to leave out a significant document made by DFI, something they didn't do in the past: DFI's own
X48-T3RS bios guide (.pdf). It could be a little more elaborate, doesn't tell you everything, but it is a must have to get started. Thanks to RGone and the DFI Club forum to point me in this direction.

Have a look how far we pushed this board on the next page >>>