I'm planning to soon upgrade my PC to this:

CPU: AMD Athlon 64 3000+
CPU Cooler: Thermaltake Silent Boost K8
Mainboard: Albatron K8X800 Pro II
RAM: 512 MB

1. I'm lost in the RAM jungle. I gather that I need DDR 400/PC3200 RAM which normally comes at a 3-3-3 timing. Is there any real gain investing into modules with better timings? Has anyone seen benchmarks of RAM latency impact on Athlon 64 systems? Since all current Athlon 64 boards are single channeled, I can buy one 512 MB module (and not 2x256 MB), right?

2. Does anyone have experience with Albatron as a company? How is their service (frequency and usability of BIOS upgrades, tech support etc.)? This board comes closest to what I want: fanless northbridge, firewire, sata, no performance problems with gigabit lan or sata raid, ability to turn off sata when not in use (otherwise expect a long delay before booting) and on top of that is rather cheap (130€ - but hard to come by)

3. See any trouble with the above setup?

btw: I don't plan to overclock this system.

1) I would actually expect the A64 platform to derive the greatest gains from low latency memory relative to other current platforms since the A64 has integrated its memory controller on die specifically in order to reduce latencies. The on die controller is a bit picky about which modules it can use however. You should be able to buy 512MB DIMMs for the system, provided they are properly compatible.

2) I've not used an Albatron board but I have not heard horrible or great things about their service. Here's a review of that board if you haven't seen it yet.

3) No, provided the RAM isn't using Winbond CH-5 chips.

Thx for the reply. I went ahead and bought the setup listed above. I opted for Corsair CMX512-3200C2 RAM (see here for a review). The tested sample, when powered with 2.8V, allows for 2-2-2-6 timings at 206 MHz. I hope that at standard 200MHz I don't need to "overvolt" to get this performance.

Once I get the system (It'll take a while to ship since not all parts are readily available) I'll run a few benchmarks at slower timings to evaluate the influence on performance since I'm very interested in this.

edit: fixed link