dpawl31

So here is the deal.
I have been told running your RAM faster than CPU is pointless and just takes away from 1:1 stability. I have seen this on many a website as well as this one. (The guide)

Well I saw no reason to not run it as fast as it could, so it could take whatever the CPU could throw at it (was running 250:400)

So here is the thing: I tried underclocking my RAM to 250Mhz to achieve 1:1, which also gave me the ability to drop my timings to 3-4-4-10 from 5-5-5-12.

Then I ran Aquamark and SuperPI... both got lower scores? 2 seconds slower with PI and about 3k points less in Aquamark.

So... what is the deal? Supposedly the stuff going through the CPU is at the fastest 250Mhz... and the RAM is running just as fast as the CPU. Why would I lose points and time when supposedly it's not good to run your RAM faster than your CPU? My ram is rated for 400Mhz and 5-5-5-12. I was running a 5:8 divider.

Anyone know what gives?

dpawl31

BUMP! Please help.

Wincet

Running the ram faster than 1:1 gives a performance boost. It is most noticeable in artificial benchmarks like SuperPi. Generally for Intel, it gives better performance to run CAS 4 or 5 and max out the bandwidth for best performance. You should experiment to see what is fastest for you.

dpawl31

Sweet, one bump and a reply minutes later hah.
Thanks for the reply!

Can you help me understand why? I can see running it faster would allow it to take whatever the CPU dishes out. But if data coming out of the CPU is 250Mhz at most, how can that happen? Can data sometimes get through faster than the FSB? I don't get it lol.

sccr64472

If you understand memory latencies, then you'll have a better picture of why. Remember, latencies are time delays and are inversely proportional to clock speed. If you take a 2T Column Address Strobe, for instance, that is imposing a 2 clock cycle delay (T is a period or 1 complete cycle). If you reduce the value of T (increasing clock speed), then you are lowering the delay.

dpawl31

Yeah I really don't get latencies lol I will have to research that...
Thanks for the help guys.

sccr64472

Just imagine a grocery store and you're asked to get an item from the 3rd column about 4 rows down. It's very similar to your cpu asking for data from a memory address. If you were told that you had to pause before moving from row to row and also when moving from column to column, think of those pauses as the memory delays required for RAM to read/write data. If you lower those delays, you will speed up the process and you lower those delays by increasing clock speed. Does that help somewhat?

1c0n

+rep. well said

__________________
Meet Mr. Blobby. He's a Blobfish, yes he's real, so dont be mean to him. He's my friend, he makes me smile, and no, I'm not taking him down, I'm lonely

Say Hi to Blobby, he likes it

3.33

kc-tr

Its a bit off from what sccr is talking about but the faster your memory is running the better performance you will have. for example I'm running pc3200 1GB of Kingston HyperX, normally this will be @ stock speeds 200Mhz via cpu-z but 400MHz actual speed. Now right now Im running @ 240MHz or.. 480Mhz actual, which increases your memory bandwidth, as in data transfer rates, which first boost performance, and also with the higher frequency it provides a faster response. only drawback of running high speeds like that is that you have to increase voltage and your memory timings, however the higher memory timings are compensated by the additional speed. For example my cpu @ 2.6 with memory running at 225Mhz will give me a super pi time of about 32 seconds... now with my memory running at 240Mhz I can now hit 30 seconds with the same clock speed.

Timings @ 225MHz
2.5,3,3,6

Timings @ 240MHz
3,3,3,8

__________________

dpawl31

Ah, I was under the impression to GET to the RAM, it had to go through the CPU first. Apparently my understanding of RAM is way off! REP to you guys for all the helpful info!