Test Setup:

JMke's Test Setup
CPU	Pentium 4 "C" 2400Mhz
Mainboard	Asus P4C800
Cooling	Zalman CNPS7000A-Cu
Memory	2 * OCZ EL PC3700 Gold Rev.2 at 2.85v vdimm
Video	nVidia Geforce4 Ti4600

Benchmarks

I set out to find out how low I could go with the memory timings; I set the FSB speed to 200Mhz using 1:1 divider. My first attempts were unsuccessful, no matter what timings I chose the system refused to POST or would hang during boot. But I found out (thanks Bigtoe@Xtremesystems) what I was doing wrong. The Asus P4C enables PAT automatically when the FSB <= 200Mhz, this will set the RAS to CAS at 2. OCZ PC3700 did not work with this setting. So putting the PAT at standard instead of auto/turbo solved this issue, choosing memory timings CL2 2-3-5 the system booted up without problem and was stable up to 208Mhz.

Earlier PC3700+ memory had problems running at CL2 with such tight timings, but these sticks can handle it as long as you put RAS to CAS at 3.

Now that I found the maximum speed at tight timings, I went for the highest speed. At CL2.5 3-3-7 the system was ROCK stable at 250Mhz! That's PC4000 performance right there! In order to reach 255Mhz the timings went to CL3 4-4-8, this was the max speed with this memory at 2.85v vdimm voltage.

So let's find out what the sweet spot is of this system, my P4 2.4 "C" is no "high flyer", the CPU can reach 3200Mhz at default voltage, and 3270Mhz when I push 1.7v through the core (not something I like to do). This translates into an FSB of 260-265, the ideal memory setup would be of course ram doing ~265Mhz at CL2 2-2-5, there is no ram out there on the market that qualifies for these extreme conditions. Sure I've seen memory run at these speeds, but those are very rare finds and you have to increase the memory voltage up to dangerous heights to obtain these speeds (3.3+v vdimm).

So now that we have an idea of what we want to obtain, let's see what memory configuration delivers the best performance.

While we can get a general idea of the memory benchmark by using synthetic benchmarks, I find it is of value to include a game-benchmark which is known to be memory bandwidth sensitive: Quake 3 Arena.

All bars in BLUE: divider set to 1:1
All bars in RED: divider set to 5:4

So let’s start of with the classic one: Sisoft Sandra 2004



Higher Memory speed equals higher bandwidth in this benchmark, although the 255Mhz results are with very relax timings, they turn in the best results.

And now a pretty unknown benchmark called "Particle Fury". This little 3D application stresses the system and memory bus by drawing a fountain of particles on your screen as fast as possible. It provides a quick and easy way to verify if the changes you made to you memory subsystem have a positive or negative impact.





They reflect the results I got from Sisoft almost identically; the highest memory speed again takes the top performance crown. Almost I said, as you can see that CL2 @ 208Mhz using a 5:4 divider gives a higher result then CL2.5 @ 250Mz using 1:1 divider, the CPU does have a 120Mhz advantage in the CL2 config adding to the equation as to why it gave better results in this test.


Enough with the synthetic benchmarks, here some real-life application action: Quake 3 Arena




Test with both resolutions give the same results in the end.

1. CL2 2-3-5 @ 208
   
2. CL2.5 3-3-7 @ 250
   
3. CL3 4-4-8 @ 255
   

Let's wrap it all up in the conclusion ->