Tweaking
I was originally going to have "Overclocking and Tweaking" as one section, but it ended up being a massive page, so I've separated them. Overclocking's on the next page.

I should mention first-up that I am using a beta bios for this review - the KA7_42 revision available from bxboards.com. This adds a few worthwhile features including memory interleaving and, most interestingly, the ability to change L2 cache ratio in bios. This is fantastic, as previously you had to use a GFD with L2 option, which required soldering the PCB of your Athlon. The L2 cache seems to often be the limiting factor with Athlon overclocking. For example, if you were to try to run your Athlon 750 at 1GHz with the default L2 cache ratio of 2/5, you'd be trying to run the L2 cache at 400MHz - good luck! Changing the ratio to 1/3 makes it 333MHz, of course. There's a performance hit for slowing the L2 cache, but if it gets you stable at a higher core speed you may be able to achieve more speed overall. Anyway, the ability to change the ratio in BIOS is a very cool addition for overclockers.

The KA7 BIOS really lives up to Abit's reputation of supporting overclockers and tweakers. There are zillions of possible combinations of settings, everything from "CPU Fast CMD Decode" through to "CPU Pull-Up Strength", "Fast R-W Turnaround" and quite a few other ones. There's also a lot of settings relating to PCI and AGP timing. Fairly typically (for everyone, not just Abit), the more obscure ones are not explained very well in the otherwise excellent manual. I experimented with a few but couldn't see any difference, perhaps the persistent tweaker will find a few FPS here and there. There is a "Load Optimized Defaults" option for lazy folks, as well as a "Load Fail-Safe Defaults" for when things go horribly wrong. I note that the "Optimized" setting leaves the RAM at the slowest timing, so it's hardly ideal.

One place where this board really shines (with this beta bios, anyway) is in memory tweaking. There are no less than 6 settings for each slot, from "SDRAM 10ns" right through to "Turbo". These DO make a very noticeable difference. You of course have the ability to run your memory at FSB+PCI, FSB, or FSB-PCI (so, if your CPU is at 100MHz, you can run your memory at 133MHz, 100MHz or 66MHz). You also get memory interleaving, which is like RAID for memory. Given that you have 2 banks per stick of memory, usually, you can read/write from alternate banks instead of each bank in turn. This also produced a noticeable difference with my single 128MB KingMax PC133.

I spent a couple of hours twiddling with memory and FSB settings with my Athlon750 on the KA7, to see what kind of Sandra memory benchmarks I could get out of it. This is Ath750 @ 112FSB/38, Mem FSB+PCI, CAS3, Mem timing "Turbo", L2 @ 2/5, interleave X4. Single 128MB KingMax CAS3 PC133.

If that doesn't impress you for a CAS3 memory benchmark, you haven't been paying attention. Admittedly, I was getting some "stuttering" in games at that speed, but I suspect it's the memory failing to cope, not the board. It may even be related to the compatability issue in the next paragraph. Regardless, it's a dramatic demonstration of what the board's capable of in the hands of a determined tweaker.

One thing I will mention - the board seems to have an odd compatability issue with the KingMax RAM. This is a v1.1 stick, 128MB of CAS3 PC133. It works flawlessly at some speeds, as you can see above, but at other speeds it seems to be extremely sensitive to BIOS changes and will hang during the POST. This sends you into a frustrating loop of rebooting, powering off etc, in an attempt to be able to get back into the BIOS to fix whatever you did. Oddly, it doesn't even seem to be fixed by clearing the CMOS or flashing to a previous version of the BIOS. Very odd. It's not common enough for me to say "Don't try to use KingMax with this board", but when it does happen, it's really frustrating. Trying to run the stick at CAS2 at 100MHz, "Turbo" timing, did it every time - this is not a particularly optimistic setting. I also found the board less stable when overclocking (anything over 110FSB would lock up from time to time). These problems all went away when I swapped the memory for a HTL PC100 CAS2 stick, which does 133MHz @ CAS3 fine on this board. Speaking of overclocking..