I'm wondering what benefits there are to running RAM at parity with the FSB (1:1) when overclocking. It has seemed to me that some articles recommend doing that, but I don't recall anyone ever specifically describing what the benefit might be.
Is it really better than, say, running the RAM at a faster speed than the CPU?
I have my RAM at 1:1 right now with my OC'ed E4300. (FSB 333x9; RAM is at 667). But my RAM will also run happily at 833MHz at the same 333FSB. Which is better?
What I think I'd like to do is drop the multiplier on my E4300 and run the NB at 400MHz so that I can run my PC6400 RAM at 1:1 on it's stock instead of downclocking it. But I have concerns about NB heat--I've posed another thread regarding this issue, so if anyone wants to chime in please search my posts (there aren't many yet!) and help me out with that issue.
Eh, the answer to this question isn't nearly as simple as it used to be with IMC, Bus Clock Locks, HyperTransport, and even Memory Clock Locks now available but I'll try.
"what is going on at 1:1 ratios that isn't going on at other ratios?" Absolutely nothing. Which can be good. Theoretically anything it did would slow it down. At 1:1 the pulses should basically go in one side of the chip and get spit right out the other with minimal interferance.
There used to be some chipsets that suffered horribly from using Memory divisors of any kind. Running at 1:1 would, in fact, give better performance than running the memory twice as fast on some motherboards. This has lead to people recomending 1:1 as the "best performance". This is not nearly as much of a problem as it once was though.
Running the memory slower would, of course, reduce memory bandwidth and lower performance for any application that was sensative to memory bandwidth but it could allow a higher CPU OC or the use of crappier RAM (theoretically saving money if you just need crunching power and not bandwidth).
Running the memory faster than the FSB has very limited gains anyway since it doesn't increase FSB bandwidth. It can reduce latencies though CAS4 @400mhz is lower latency than CAS4 at 333mhz which can speed up appz that are sensative to latency and slightly increase bandwidth through lower latency (but not increasing bandwidth from the extra 66mhz as it is bottlenecked at the NB chipset). AMD's CPUs don't really have a FSB and use an IMC and I'm not pulling any good explanations as to how that relates to your question off the top of my head. Perhaps someone else will.
You could, of course, always bench it both ways and see what impact it has on the appz that matter to you. Definately don't relax the timings to run the RAM faster though. Almost positive you would get less performance as a result (and spend more energy generating more heat in the process). In general 1:1 with tightest timings possible is regarded to be the fastest configuration. CL4 @ 333mhz is lower latency than CL5 at 400mhz for example.
My PCMark05 score with CPU the same (9x300) only changing the memory ratio was....DDR600 = 1:1 = 6615marks. At DDR800 = 2:3? = 6650. Less than 1% gain from running memory that was 30% higher speed.
All memory timings were kept the same at 4-4-4-12. I think the reason people recommend 1:1 is that there is little difference for most applications. You may get better results from running DDR600 3-4-4-10 than running DDR800 4-4-4-12. And its easier a bit simpler to leave it 1:1 when OC'ing
Wow! Thanks for all of the info, folks. I think I'm understanding things a bit better now.
Here's another question. I'm not sure if the answer is "readily apparent," and I may just have to try it myself to see, but what do you think of this: Right now I run at FSB 333 with 1:1 RAM at 667DDR2 4-4-4-12. Would I be better off dropping the multiplier on my E4300 from 9 to 7 and clocking my FSB at 400 while running the RAM 1:1 at 800 4-4-4-12 (yes, it's rated for that).
I'd lose about 200MHz in CPU speed (I'm at 2.99 right now), but would the FSB and memory performance increase offset it?
Also there are claims that there are issues with lowering multi on Intel CPU that might make it harder to get to 400mhz FSB with a lowered multi then it would with the full multi.
I would try to tighten my ram timings first if I were you. 4-4-4-11 is still within spec for your RAM @ 667. 3-3-3-9 @ 667 is only 10% faster than what it's rated for. Something inbetween those two ought to be possible. Once you figure out the tightest timings it can do at 667 vs. 800 then you will have a more informed question to ask.
But what you really REALLY ought to do is leave the timings alone, leave it at 1:1, and increase the FSB. You should be able to hit 350mhz for DDR2-700 and that should be faster than any of the other configs. If you write all your settings down for the various configs once you've verified they are stable you can quickly switch between them to do comparative benchmarking.
Oh, and I just looked at your sig again and you should be able to lower your RAM voltage if you're going to run lower frequency than DDR2-800. It'll save you some electricity and help system run cooler.
No that would be worse I think. You would gain 1% if your lucky from the ram increade and lose 6-7% performance from CPU decrease. Now if you can run your FSB at 400 and use a 8X miltiplier then your stylin!