Related Articles

Overclocking

There is no mystery here ... as you saw on the (many) charts, the AM2-5000+ overclocked VERY well.

We reached ~3.0GHz from 2.6 GHz using air cooling in our fairly warm lab environment, which is no mean feat -- especially with a new die.

It's still worth discussing how I got the results, as the initial attempts at overclocking were less stellar.

The ASUS M2N32-SLI we used is an early pre-retail sample with a BIOS that did not allow us a great deal of leeway in overclocking the system. The biggest concern was that the board simply didn't offer enough headroom and had limited tweakable settings. Fortunately, late last week we received an updated BIOS (revision 0402) from ASUS that made an astounding difference. Retail boards that are being made available will have the same if not newer BIOS.

How big a difference did the BIOS make?

With the original BIOS I was unable to stably run at HyperTransport speeds of over 233MHz. With the new BIOS I was able to stably run at HT speeds of 333MHz. You be the judge.

That's a huge difference if you ask me, and made all the difference in pushing the 5000+.

The new BIOS also added several pages of additional memory parameters that could be tweaked, which I plan to take a closer look at - but with the launch date I was trying to meet I did not have the luxury of trying as many things as I otherwise might.

Basically there were three factors to the successful overclock - four if you count the updated BIOS.

1. Figuring out the fastest stable memory speed
2. Figuring out the fastest stable memory timing
3. Setting the HTT speed and multiplier
4. Getting the right Vcore

I wish the BIOS just let me choose the divisor manually, I could then better control the memory overclocking and get even better memory performance figures.

The memory frequency is determined as follows:

(HTT speed) x (cpu multiplier) / (memory divisor) = memory clock

... And the DDR2-800 memory (by definition, since it is Double Data Rate) actually uses a 400 MHz clock.

Therefore, to reach the 333 x 9 final overclocked speed, the following steps were taken:

• - Set the FSB (HTT) speed to 333MHz
• - Configure the HTT multipliers to 3x
• - Raise HTT voltages to 1.25V
• - Raise Vcore to 1.5V
• - Set the memory speed to "DDR2-533"
• - Set the DDR2 memory voltage to 2.2V (as specified for the modules)
• - Set the memory timing to 4-4-4-12/2T (as 1T was not stable at that speed)

Now why was not 1T stable? Simple! According to CPU-Z the memory was running at DDR2-876!

A Sample of Rightmark Benchmarks During Overclocking

I thought you might find a quick table interesting - the complete set of experimental results for the Rightmark Read benchmark. Note that some of these results were obtained at what were ultimately unstable settings, which is why I ended up choosing the 333x9 setting for the maximum stable overclocked results.

As you can see the 275x11 setting achieved somewhat higher transfer rates, but unfortunately the system was not totally stable at that setting.

I really hope future BIOS's will allow me to manually set the memory divider, and give me more than four choices - I suspect there is extra bandwidth to be had by using non-standard settings. I'm also cautiously optimistic that the Foxconn C51XEM2AA that we're also reviewing today might give me more headroom, or that our full retail M2N32-SLI (in the mail) might also offer some additional headroom -- so stay tuned.