I managed to build my 1st computer a few weeks back (specs below) and have successfully pushed the CPU up to 3.6GHz (so far!) without any changes to my default memory config. And the reason I haven't touched my memory config is simple: I have no clue what to do.

Can someone summarize the basics of tweaking memory settings?

I've heard about the 'big 4' numbers but have no clue what to do with them (or what they really mean). I know I can also tweak the voltage to the memory, but have no clue how much or under what conditions. Finally, I know there's a ratio between the CPU and memory (which I currently have set to 1:1) but no clue when to modify it. (I kinda expected http://www.tomshardware.com/forum/ [...] ad-posting to cover this but, sadly, it did not.)

Where do ya begin with all this junk?

OK, perhaps I was a bit...hasty...in posting my question. While I didn't find my answer on my favorite overclocking site (Tom's Hardware), I did find a great one elsewhere: http://www.hardwaresecrets.com/article/152. If you are new to overclocking, this is the one article about memory worth reading (preferably before you buy your motherboard and memory).

On an Intel (Front Side Buss) based system, the answer is "Don't Bother 'cos it doesn't make much difference in overall performance, since Intel cache the hell out of everything."

scotteq, could you explain what you mean by 'Don't Bother 'cos it doesn't make much difference in overall performance, since Intel cache the hell out of everything.'?

What would the Intel cache (I am assuming you mean the cache on the chip itself) have to do with the performance of memory that is serving my operating system and the memory-intensive applications I might be running on it?

he means intel preloads everything from the memory in to the cache right away instead of storing it in the memory and having the cpu read off the memory and than caching it.

Intel chips communicate with the memory through the FSB. Because the FSB is slower then the direct HTT link that AMD chips have, Intel chips have large L2 caches to compensate. This is why AMD NEEDS fast DDR2 memory, while Intel doesn't really.

The "big 4" I'm assuming you are talking about the latency numbers. (5-5-5-15 for example.) Latency is just that, a certain number of clock cycles that MUST pass before something else can happen. The lower the number, the less time ram spends waiting. 4-4-4-12 is faster then 5-5-5-15, because you only need to spend 4 clock cycles waiting instead of 5. You can lower these numbers by either downclocking the memory (running DDR2-800 possibly at 667, or more like 533.) or by overvolting it. (running at 2.1-2.3, instead of 1.8v)

I think Scotteq was trying to say it doesn't matter much what you do with your ram, because Intel chips aren't sensitive to RAM settings. As long as your RAM is as fast as your FSB, there aren't a lot of farther gains to be had.

You have a 8500. 3160 / 333MHz = 9.5 CPU multiplier. 3600 / 9.5 = 379MHz actual FSB. 379 * 2 = 758 for your ram speed. Seeing as your ram is validated at 1066, you might be able to lower the latencies without having to bump up the voltage.

Or just listen to Scott, don't worry about it.

actually, we don't really know if the memory multiplier is 2, as stated above. It depends on how the NB stap is set. For instance, if you strap NB as 1066FSB (4*266), and use DDR2 800, then it will set the multiplier to 3 (as 3*266 = 800). If you raise the FSB later, then the memory will follow.
I believe you are using 1516 FSB (4*379), and your memory is working at 379 times whatever the multiplier is set. it could be 2, 2.5, 3, I can't really tell.

You might want to read the first post again.

Assuming the OP is telling the truth, he is running the ram at a 1:1 ratio, not 4:5, or 3:2, etc.

Chris - It means pretty much what I said: On current Intel based systems with a Front Side Buss, the actual clock speed of your memory doesn't matter too much. The standard to shoot for is a 1:1 ratio (ex - DDR2 667 for a 1300Mhz FSB), as this provides the best theoretical efficiency. In essence, since the two are synchronised you generally don't have data sitting in a queue for few clocks before being sent.

In absolute terms, you can achieve more memory throughput by running faster RAM, but the gains are generally a few %, and you can achieve similar results by running tighter timings. (Timings control how many clock cycles it takes to move data - "Cas 5" timings of 5-5-5-* are 'slower' than Cas 4 4-4-4-*, given the same clock speed) The additional throughput will certainly show up in benchmarks, but generally won't be noticable in normal usage because there are very few situations where you are acually limited in terms of overall bandwidth. Tighter timings with a "slower" clock speed, on the other hand, can make your system feel more responsive in general.

You can almost think of it like a sports car versus a motorcycle: (using imaginary numbers here, and only to illustrate a point) You can have a hot Corvette capable of running 190 MPH, because it has more horsepower/gearing/aerodynamics that enable it to reach that top speed. You can also have a Sportbike that "only" goes 175 MPH - Less overall power from the smaller engine, and aerodynamics roughly equivalent to a brick. The Car's better, right?? Not so fast: At the speeds people actually drive, the bike will rip the bloody lungs out of the 'Vette cos it's so small and light compared to the power it makes. The bike is "quick" (tighter timings), and the car is "fast" (higher clock speed).

Back to computer land: A reason to buy premium RAM, even if you aren't running it at rated speed, is that at lower clocks you generally have more performance left over to run tighter timings.

Hope the explanation helped.

This is all very helpful advice! I think I get the gist of CAS settings, bus speeds, etc. But I can't say I know enough to answer my next questions:

1. is 3-4-3-9 DDR2 800 better or worse than 5-5-5-15 DDR2 1066 for an Intel chip/motherboard? (I can't seem to find ANY memory that has both CL < 5 and DDR2 > 900.)

2. if my memory is rated at '5-5-5-15', does that mean I can't tell it to operate at faster speeds? Is this why my current memory doesn't boot after I try something like 4-4-4-12?

Unless you are going for overclocks involving a Front Side Buss speed in excess of 400 (1600 Mhz), there is no reason to buy faster/more expensive memory on an Intel system.

Yes, you can tell your memory to run tighter timings, or clock it faster, or whatever... But understand that memory is MUCH less amenable to overclocking attempts than (current Intel) processors. You can get 30% out of a Q6600 practically for free, more than that if you have a little knowledge and patience. 30% overclock on memory?? Don't count on it. Not at all.

Generally speaking, the safest path if you want to overclock an Intel box is to UNDERclock the RAM until you have the processor running where you want it. Then go back and set your memory accordingly - knowing that a 1:1 FSB/RAM ratio is generally most efficient.

The maximum DDR2 memory standard is DDR2-800, so DDR2-1066 modules etc are really essentially factory overclocked modules. Often, you can clock back the modules and tighten the timings. This can provide similar performance to the higher-frequency, looser timings combo. In general, it won't make a huge amount of difference in your system performance, though.