Short intro...

When you try overclocking and you post your results on a forum, you'll get a reaction like this:



Seen this conversation before? Well, as a matter of fact, there are very few sticks that can do CL2 2-2-5 timings. Most ram even won't do "Ras To Cas Delay : 2" and "Ras Precharge : 2". Mostly it's the Ras to Cas that's holding you back, set it to 3, or higher, and problem solved. Most of these sticks have the famous Winbond 5ns chips. It really are wonderful chips, can't deny that. They go over 250MHz at these timings, at a nice voltage of 3.1 Vddr that is. You’ll rarely find sticks that match these high specifications.

Let's take TwinMos for example.
TwinMos once had a series of pc3200 memory with such chips, hard to find right now. Most of their memory modules you find currently have "50D" or "50B" chips. One letter difference = a whole other world of performance. The first are terrible, barely working at the rated 200MHz. The 50B chips on the other hand can do almost 225MHz on a healthy voltage of 2.85 Vddr. But what about their timings ? Well, I happen to have such sticks, on 225MHz they can run at CL2 2-4-11. Not bad, but no matter what I do, they won't boot at 2-2-5...

Recently, Winbond stopped producing the BH-5 chips, but because of the popularity they started to make them again, and now they stopped again, don't really know it for sure, but that's not the point.
The only memory modules that use such chips now are Mushkin Modules. “Yippiekayee” you would say? Not really, they sport a price tag almost the double of their TwinMos 50B brother...

So is it really worth spending that much money for these slightly different timings? I tested the most common memory timings and compared them against each other using a small range of benchmarks. Thank you RichBa5tard, who borrowed me his stick of memory.

As currently Nforce2 based mother boards are the most popular for AMD XP systems, I decided to use such a setup to test it. So the results could slightly differ on other chipset based boards.

Don't know what all these terms mean? Ras / Cas / Tras ? More and detailed information concerning this can be found here.

The test setup :

The test setup !
CPU	AMD Athlon XP 1800+ "DLT3C"
Mainboard	Abit NF7-s Rev 2.0
Memory	1 * 256 Mb pc3200 Winbond BH-5
VGA	Club3d GeForce FX5900 XT 128 Mb DDR

The CPU was running at 2.2GHz (11 * 200 fsb), with the memory on a 1:1 divider. Meaning it was running also on 200 Mhz = synchronous.

The different memory timings tested:

CL2 2-2-5

CL2 2-2-11

CL2 2-4-11

CL2.5 3-3-7

CL2.5 4-4-8

I also wanted to test CL3 4-4-8, but nor the BH-5 stick, nor the 50B stick wanted to boot at these timings. So I used my OCZ pc3200 Revision 2. Worked flawlessly on these timings, but gave a performance that wasn't linear compared to the BH-5 stick. It was in most cases even faster at less aggressive timings. Weird! I asked around on Xtremesystems.org forums and there they told me that it was the OCZ behaving specially, not the timings. So these results were not included.

To know how to change your memory timings take a look at our “AMD Bios settings howto”


Here are the different memory timings, you can verify them in windows by using CPU-Z

Sisoft Sandra - Memory Bandwidth Benchmark

Benchmark used: Sisoft Sandra 2004

Theoretical Memory Bandwidth: here the difference is smaller than 100 MB/sec between the two edge values.



It shows you that there IS a difference, but to better get an idea how this affects real world applications we included some other benchmarks, read on ->

SuperPi

Benchmark used: SuperPi

The difference between the fastest and the slowest timings results in a score difference of 6 seconds. Gaining 6 seconds in this benchmark requires the overall CPU speed to almost increase by 200MHz.
So the timings are important here...2-2-5 isn't the fastest, 2-2-11 is!

3DMark 2001Se

Benchmark used: Futuremarks’ 3DMark2001SE

Timings matter, clearly. 700 3DMarks could mean a big difference on the ORB (Online Result Browser – 3DMark score stats page)

Particle Fury

Benchmark used: Particle Fury



Once again, the winner remains the same, but like seen before, 2-2-11 only beats 2-2-5 by a small margin.

Quake 3 Arena

Benchmark used: Quake 3 Arena

By setting the resolution to 640*480, this game becomes less video card dependable. Now it's more sensitive to CPU or Memory changes. The most remarkable is that the very famous 2-2-5 seems not to be the fasted on an AMD Athlon XP & NForce² platform....
And, CL2 2-4-11 isn't much slower either...

Conclusion

What memory to buy?

If you're not aiming towards any world records, I wouldn't buy high performance memory, but if you want to squeeze every last 3DMark out of your system then you might take it into consideration.

For the average overclocker and normal user it's of course nice to have it, but far from a must-have. The increased performance margin is too small to justify the remarkably higher asking price.



Cheap but decent performance = TwinMos "50B" or better.
Best performance = Winbond’s BH-5 chips, for example Muskin Level 2 Memory sticks.

/Piotke out.

Comments or questions can be added to this thread @ the forums (no registr. required)