Originally Posted by Fatal05
I built a bunch of 939 systems back in the day, and AM2 systems aren't much different. Increasing the hypertransport speed (by using quicker memory accordingly), does virtually nothing in terms of performance. You're better off tightening your memories timings.
While I have had no experience in overclocking an AM2 platform, I assume that at your memories stock settings (DDR2 667), you can reach a "FSB" of 333 in your bios. Which is more than enough for pretty much any AMD system.
There is no "FSB" for current AMD systems. You're referring to the reference clock, but NO DATA FLOWS ON THE REFERENCE CLOCK. It is a number generated for reference by the CPU, HT Link, and memory. The CPU and HT Link multiply the reference clock by a certain number to determine their speed. The memory references the CPU speed to determine it's speed. There's therefore no advantage to running a reference clock at 333MHz and your memory at 667MHz effective (or 333MHz in reality). This 1:1 concept doesn't apply to current AMD systems as far as I can tell, unless someone can prove otherwise.
The only thing that truly matters when overclocking an AM2 socket AMD system is CPU speed, assuming you don't completely butcher the HT Link or memory. Max out your CPU speed, keep your HT Link around 1000MHz, and your memory between ~650-1200MHz. It doesn't really matter which speed you run your memory at, as long as you run the most aggressive timings possible at that memory speed. For example, if you run your memory around 650MHz, but can nlower the timings to 3-3-3-8-1T, you'll probably get similar performance to running your memory around 850MHz and 4-4-4-12-2T timings. And both would probably be similar to say 1080MHz at 5-5-5-15-2T timings. There will be slight fluctuations in performance, but we're talking like 1-4%. Bandwidth and latency depend on BOTH speed and timings, and usually any boost you get from speed is offset by the increase in timings. Unless you're completely noob, it's pretty hard to mess up the memory settings haha. Just choose a speed and tighten the timings.
Now this doesn't mean go buy some cheap DDR2 667. You have to look at the speed of the memory and the timings it's rated for. A quality set of DDR2 800 rated for 4-4-4-12 means it can run those timings at 800MHz. That means you can probably run 3-3-3-8-1T if you lower the speed to 600-700MHz or so. But that cheap set of DDR2 667 probably has bad timings, like 4-4-4-12 or 5-5-5-15 at 667MHz, meaning it will probably need 5-5-5-15-2T if you ran it at 800MHz, and probably is incapable of running 3-3-3-8-1T at 600-700MHz.
So the point is comparing DDR2 667 to DDR2 800 is pointless unless you also take into account the timings at those speeds. If you can run DDR2 800 at 4-4-4-12-2T, and you're unable to tighten the timings any further at DDR2 667, then DDR2 800 is the better choice for sure. But if you were able to tighten the timings to say 3-3-3-8-1T, then it's probably the same performance at both settings.
Originally Posted by MiniTuckin
so how much more headroom would u guys say it would give you? i just ordered me 2x1gb sticks of g.skill upgrading from my 1gb samsung 667
sorry for trying to steal thread... just figured this would help us both
See above. However, your situation is a bit different as you're talking about upgrading the amount of memory as well. Most systems today use more than 1GB of memory in a lot of applications (or combinations of applications) so you'll likely see a nice performance increase based on having more memory.