AOpen AK89 Max: From Ugly Duckling to Swan
by Wesley Fink on April 21, 2004 4:16 PM EST- Posted in
- Motherboards
AOpen AK89 Max: Extreme Overclocking 1
With a working AGP lock and working multipliers, we finally get a chance to see how far we can take this Athlon 64 3200+.
Click to Enlarge.
At default voltage and the stock multiplier of 10, we achieved a stable OC of 233FSB or 2.333GHz on a CPU rated at 2.0GHz. This is the most common method of overclocking where the ratio is fixed and the memory speed is 1:1 and increases with the FSB. It is also the only method of overclocking available on Intel Pentium 4 processors (unless you are lucky enough to have an ES or Engineering Sample that is unlocked).
The Athlon 64 family offers additional methods of overclocking. If the board supports it, ratios or multipliers on the Athlon 64 Socket 754 can be adjusted - but only downward. A 3200+, for instance, runs at 2.0GHz or 10x200. Ratios can be adjusted from 10 down on some boards. The Athlon 64 FX has a completely unlocked multiplier, and can be adjusted either lower or higher than the "assigned" multiplier.
Click to Enlarge.
By using this technique and dropping multipliers, we easily reached 250x9. Why would anyone want to run at 250x9 instead of 233x10? First, to use fast memory at the fastest speeds possible. In this case, 250 allows us to run our memory at DDR500, which is significantly faster than DDR400. Second, a higher bus speed, even with the same resulting frequency, generally performs better. We have seen this in the Intel Pentium 4 when the FSB was increased from 533 to 800. Even at the same resulting frequency (2400), 800FSB (12x200) was faster than 533FSB (18x133).
Another way to illustrate this is to compare the same CPU speed at both a 200 and 250 FSB. Here, we are running the AK89 Max at stock 200x10 (2.0GHz or 3200+) using OCZ3500EB memory at 2.5-2-3-9 timings. We have included 3DMark2001SE scores to show the performance difference.
Click to Enlarge.
The same tests were run at the same 2.0GHz but with settings of 250x8. While we used the same OCZ3500EB memory, we had to lower CAS timings to 3 from 2.5.
Click to Enlarge.
Even though we are running slower memory timings and the exact same CPU frequency, the 3DMark increased. Imagine the differences in 200FSB and 350FSB in this benchmark.
Facebook
Twitter
RSS
23 Comments
View All Comments
GooFy - Sunday, May 30, 2004 - link
Could you somehow test some other memory modules with this board, i have a pair of Adata PC4000 that will not go over 230mhz @ original (Cas 3 8-4-4) or any other timings either yet they are stable @ 230 mhz fsb and Cas 2 6-3-3.I'm going to buy some new memory modules and i would like to know what options there are as it seems there are some other memories that will not go to 250+ fsb.
Anyway i'm sure that it is the memory that's stopping me because i ran alot of tests with different timings and lower multiplier and so on.
TrogdorJW - Tuesday, April 27, 2004 - link
What the crap... lost my post. Grrrr.... Okay, here's a recap of what was supposed to be in that last one:You're cruel, showing us scores of 200 * 10 and 250 * 8 and then holding out on the scores for the "Extreme Overclocking" setups. I don't want a screenshot, but I would have been curious to hear what sort of performance difference there was between the 233 * 10 and the 345 * 6.5 setups. Do the asynchronous RAM timings hurt performance a lot, or can the integrated memory controller deal with that okay? Would be nice to find out. Oh, well...
TrogdorJW - Tuesday, April 27, 2004 - link
MadAd - Sunday, April 25, 2004 - link
thanks wesley, they did look nice and clear in this one, thats why I asked - ill wait for your review.NegativeEntropy - Saturday, April 24, 2004 - link
cnq,I know if I had to program Cool'n'Quiet there would be, at minimum, a sanity check to make sure the Hyper Transport (~FSB) frequency was 200 MHz and the multiplier is where it should be for that model CPU prior to doing anything. It's what Cool'n'Quiet would do after that that is up in the air (or so goes my reasoning):
1) abort any attempt to engage Cool'n'Quiet (most likely given ease of programmability and official disdain for overclocking)
2) override any current settings, do its thing, and not reimpliment the pre Cool'n'Quiet settings upon an increase in CPU usage (thus effectively 'undoing' any overclock)
3) note the current state, engage Cool'n'Quiet, and then return to the pre Cool'n'Quiet state when CPU usage demands it (this is the scenario we dream about :)
As long as I get confirmation Cool'n'Quiet works on this mobo (even w/o overclocking), I may get it and answer the other questions myself :)
cnq - Saturday, April 24, 2004 - link
NegativeEntropy,Weird, I was about to ask the *exact* same question!
***This should be a FAQ.***
Everyone likes to overclock the snot out of their boards, and everyone likes cool 'n' quiet to kick in when they are just web browsing. (Naturally, when we stop web browsing and resume gaming, we want the system to return, automatically, to the fully overclocked settings we were using before.)
But is it technically possible for Cool'n'Quiet to coexist with overclocking? Just by reading AMD's technical docs (BIOS Guide), there is no answer on whether this will work. My fear is that overclocking can confuse the hell out of cool'n'quiet. Cool'n'quiet works by switching between P-states, which are combos of CPU frequency and CPU voltage. Problems could be:
a) Cool'n'Quiet is ignorant of overclocking in general. The possible P-state transition values are supplied by the BIOS, who in turn gets them from AMD documentation. Naturally, there won't be a P-state setting for an overclocked system in the official tables. But it would be possible for this to work: the BIOS could use your overclocked settings to create an additional P-state that isn't part of the official docs.
b) Even if this works, underclocking via Cool'n'Quiet is limited. P-states just define CPU speed and CPU voltage. There's nothing about FSB speeds in a P-state (and there should be, imo, along with RAM voltage). So, for example, when Cool'n'Quiet ramps down your CPU, does it also ramp down the FSB speed? If not, then it's pretty easy to see how things could crash here: a drop to (for example) 1000 MhZ might be done by setting the CPU multiplier to 5, on the (incorrect) assumption that your FSB is 200MHz, the "usual" value. But if you overclocked by lowering the CPU multiplier (already) and jacking up the FSB to outrageous amounts, then this CPU multiplier might actually _increase_ CPU speed rather than decreasing it. Not so much with a multiplier of just 5, but you get the idea.
c) Even if this works (the CPU multiplier is set to 5 in the above example, your FSB is overclocked to, say, 250 instead of 200, so you get 1250MHz instead of 1000MHz)...fine, except that the P-state contained a CPU voltage that is only guaranteed to work well with 1000MHz. Such a low voltage (whatever it is) might not be enough to power 1250MHz, causing a crash. If it is enough to power 1250MhZ, it's just blind luck, and not very stable.
Naturally, I hope I'm wrong on all of these points, as the combo of serious overclocking and cool'n'quiet would be fantastic.
Can someone speak from experience on this? In general, can you enable cool'n'quiet yet overclock huge amounts? Wes?
NegativeEntropy - Friday, April 23, 2004 - link
Wesley,Does Cool 'n' Quiet work on this MoBo?
If it works, does it still work when overclocking (or is it borked because of the lower than stock multiplier or for some other reason)?
The reason I ask: looking to build an HTPC -- CnQ would come in handy to reduce power usage, and this board would be more fun than the K8V :)
Wesley Fink - Friday, April 23, 2004 - link
#13 and #14 -Just skip Extreme Overclocking 1 and 2, then it's a standard motherboard review. ALL performance charts, like always, are at default speed and default ratios. Since this is the first A64 board we have tested with a working AGP/PCI lock and working ratios that are available on the A64, it was the perfect opportunity for a "how-to" and why of high FSB overclocking. These features will be the most important thing about this board for many Enthusiasts, and of little interest to others.
#15 -
I normally use a Pentax *ist D digital SLR with a 50f1.4 or 28-105f2.8 lens - at 800 ISO, no flash, and adjusted color temperature. However, most of the shots in this review were with a $300 Kodak 5 megapixel digital we will be reviewing on AnandTech in the next few weeks.
MadAd - Friday, April 23, 2004 - link
Its a bit of an OT question, sorry Wesley, what kinda of camera do you use to take the pictures of the board layouts/chips to use in the reviews please?retrospooty - Friday, April 23, 2004 - link
By that I mean, there are so many levels of OC, so many different ways to look at it. I am unclear what was screenshot stable, and benchmark stable, and I am unclear exactly what speed the final tests were run at. Maybe its just too early in the morning for me. :D