AMD memory timings

[malcore]

There has been a lot of talk about a strange dicovery concerning RAM timings on AMD systems.

To clarify what the following numbers refer to:
2-2-2-5

1st number = CAS Latency
2nd number = RAS Precharge Delay
3rd number = RAS to CAS Delay
4th number = Active Precharge Delay

Most know that the ultimate goal for RAM timings is the tightest you can get (2-2-2-5). While this remains very true on Pentium 4 systems, it has been discovered that the last number (Active Precharge Delay) has had some unexpected results on memory performance in AMD based systems. It seems an Active Precharge setting of between 9 and 11 produces better results.

I tested this using memtest86+ DOS based memory tester, which will give your (unbuffered) memory bandwidth in MB/s. The following were my results with various timings:

2-2-2-5 1343 MB/s
2-2-2-6 1363
2-2-2-7 1363
2-2-2-8 1363
2-2-2-9 1363
2-2-2-10 1363
2-2-2-11 1384
2-2-2-12 1384
2-2-2-13 1343

It would seem that these claims are true. On my system, A7N8X, Athlon XP2800 (12 x 189FSB) 1:1 FSB/RAM timings, the optimal timings were either 2-2-2-11 or 2-2-2-12. In fact, the tightest timimgs of 2-2-2-5 gave the worst results.

While it is not a huge difference, it is interesting for those who try to get the most they can from their system.

If you have an AMD system, you can stop trying for that Precharge delay of 5 and relax it a bit.

[pave_spectre]

I have no idea what any of that really means, and it will probably be of no use to me, but strangely enough it's nice to know anyway.:)

Maybe one day when I dont have cheap crappy RAM it will come in handy.:D :D
:cool: :cool:

[Deagle]

Hey, nice observation Malcore. So does that work for Cas 2.5-2-2-11? I'd love to get me some Winbond and do 2-2-2-11 @ 230fsb.:D
:cool:

[malcore]

Not my observation really Deagle. There's been a lot of talk about this on the web. I first picked up on it at Anandtech.

It should work with the timings you mentioned, although every system is different. Some find a tRas of 9 best, some 10, some 11, some 12. It all depends on the system. One thing is for certain though, across the board on any AMD system (64 bit included) a tRas timing of 9 to 12 produces a higher bandwidth than than a tRas of 5 or 6.

You will have to test what's best for your system. Some use SiSoft Sandra to test, but I prefer memtest86. I don't have to keep booting into Windows and because it's DOS based, there are no variables. The measurements are always the same.

Here's some tests from another system: ABIT NF7-S, XP2800, (11.5 x 203 FSB) 1:1 timings,

2-2-2-5 1508
2-2-2-6 1508
2-2-2-7 1534
2-2-2-8 1534
2-2-2-9 1534
2-2-2-10 1562
2-2-2-11 1534
2-2-2-12 1508
2-2-2-13 1457

It's not much, but it's interesting nonetheless. To me anyway.;)

[saphalline]

It's not much, but it's interesting nonetheless. To me anyway.;)Hehe, me, too. But still, these findings are bizarre to say the least. Are there any theories as to why this is?

[malcore]

I need to make a correction here. This anomaly appears to be restricted to nForce motherboards, both 2 and 3.

It seems easy to understand why a timing of 5 is not optimal, pages being closed too fast. As to why such a high tRas is better on nForce boards still remains a mystery to me. Manufacturers are beginning to discover (especially now that RAM speeds are heading beyond DDR500:eek: ) that balanced timings are producing better results than tight timings.

I saw an article about the new technology head of OCZ (creator of LostCircuits website) where he discusses this in detail. Of course, I can't dig it up just now.

Edit- I found it! Where else?.....Anandtech (http://www.anandtech.com/memory/showdoc.html?i=2019) :p

There's a link on the second page to OCZ's Enhanced Bandwidth whitepaper.

[Deagle]

Hmm, is that why all the pc4000 and above run at 3-4-4-8 timing? I still prefer 2-2-2-11 since I'm using AMD.
:cool: