At the Limit
How far is it possible to
take these boards? All boards were stable for at least 1 hour
at 1570 MHz, provided a powerful HSF like the Taisol with a 92
mm fan or the Swiftech MC426 is on duty. As already explained
a PAL 6035 or a FOP32, even with a Delta 60mm fan, will not do
the job, unless you drop your room temperature to less than 15C.
How much beyond 1570 MHz is it possible to take these boards?
At that point most A7M266 owners would have to leave the party,
because without the possibility to change the clock multiplier
our A7M266 would have been at the end with FSB 157 MHz. As our
A7M266 was modified, we tried various FSB x multiplier combinations
and settled finally for a non-stable maximum CPU speed of 1584
MHz achieved at 12 x 132Mhz. At that speed we could boot into
WinME, but after a few minutes in 3DMark 2001, the system would
lock. We experienced unpleasant data corruptions and bad sectors
on our HDD when trying to go higher than 1584 MHz The A7M266 would
post up to 1620MHz, but did not boot anymore.

We managed to take the Epox to 10 x 160 MHz = 1600 MHz, and the Gigabyte to 9.5 x 169Mhz = 1606 MHz. Looking for the highest achievable FSB frequency, we could push the Epox as high as 9 x 163 MHz. The Gigabyte went all the way up to 9.5 x 169 MHz. That was a record for us: we never achieved a higher FSB MHz with this CPU and this memory. It has to be noted that the Epox is somewhat crippled by the fact that the BIOS allows "only" for a maximum of 166 FSB. Epox should revise this, because we think this board has more potential than FSB 166.
It's always interesting to have a look at the system stability under extreme conditions, because under stress problems become more apparent than under regular conditions. The 3 boards were a bit different when pushed to their limits: after an unsuccessful overclocking attempt, the A7M266 several times simply refused to post. The BIOS was not accessible anymore, because it did not fall back to the defaults settings, as it is supposed to do. Even clearing the CMOS did not help. Additionally the DIP switch for the FSB had to be set to the default frequency of 100 MHz, only then would the A7M266 come back to life. The Epox had the tendency to show "check-sum errors" already above 1570 MHz, and we frequently had to clear the CMOS and reset the BIOS to default. The onboard LED monitor then displayed the POST code "7B". That didn't help, because this code was not explained in the manual. The board posted (without booting) up to 1670Mhz. The Gigabyte showed the most stable behavior under stress. We experienced only above 1600 MHz the dreaded "check-sum errors" and we had to clear the CMOS to regain posting capability. The board was posting (without booting) up to 1760 MHz.

The Verdict
The Asus A7M266 was quite a surprise in this test. Performance-wise
this board was almost on par with his younger siblings and could
even slightly outpace the competition in some tests. The Asus
engineers did a good job, when they designed the A7M266. At least
from the pure performance point of view, Epox and Gigabyte couldn't
take advantage of the fact that they had half a year more of experience
with DDR technology. Asus revised the A7M266 BIOS already several
times. We assume therefore that the performance of this board
is now at the maximum. At the feature side the A7M266 can less
convince: the A7M266 lacks the possibility to change the multiplier,
an advanced power supply, has only 2 DIMM sockets, and is not
available with RAID.
The Epox EP-8K7A showed excellent performance and the highest CPU scores of the 3 competitors. The board has high overclocking potential. It is the only one among the tested boards with 6 PCI slots, and comes with a unique on-board LED monitoring device and a 2-phase power supply. The "+" version of the EP-8K7A is available with a High Point Raid controller. The Epox lacks a quality sound device and has only 2 DIMM sockets. STR capability and AGP voltage setting were missing at our board, that might be revised with a BIOS update. With the somewhat-reduced features, yet excellent performance, the Epox is obviously targeting the more budget-oriented costumer and offers excellent value for money.
The Gigabyte GA-7DXR is a top performer and highly overclockable: a memory score of 927/704 and an FSB frequency of 169 MHz were the highest in this round. It is also the best equipped board in this round: It features a 4-channel Creative sound chip, 3 DIMM sockets, 4+1 fan headers, Promise Fasttrak Raid, 2 BIOS chips, a precise CPU temperature sensor, a 3-phase power supply, AGP Pro and several other useful hardware and software items like an online flash and overclocking utility. The GA-7DXR is not only technically top, but also esthetically appealing.. The equipment comes however at a price: the Gigabyte is around 15% more expensive than the Epox. The only drawback we see with this board is the missing 6th PCI slot.
It is fair to say that the performance of these 3 boards is very close. The Epox and the Gigabyte are brand-new. We expect further BIOS tweaking and performance enhancing. But currently no A7M266 owner has reason to upgrade to the DDR boards from Epox or Gigabyte. On the other hand: somebody who is now considering moving from a PIII board or an older VIA 133 board to DDR should see the Epox EP-8K7A and the Gigabyte GA-7DXR as better choices because of their better equipment and slightly higher overclocking potential.