Newbie to Overclocking

[DarkRogue]

Hello,

I'm in need of upgrading my PC, as it just passed its 5th birthday. Its a good mchine but it just doesn't have the power to run recent games like TF2 very well anymore. Since the 8800GT was released, I've been waiting for the next chipsets and procs to come out. The 780i is out, and the Wolfdales are almost out, so I have started putting together a list of hardware to buy. One thing though, is I'd like to save a bit of money and overclock (mildly! I don't want to push anything to the extreme.) a few components to gain some extra performance out of it, as likely I'd depend on the new system for another 5 years or so.

The video card seems fairly easy to do, I can get RivaTuner it seems and just OC the core and mem clocks. The other two components are the CPU and RAM and that's where I get stuck in the mass of information. As I understand it, the CPU overclocks via the FSB setting x the multiplier, correct? The E8400 Wolfdale is a 3.0GHz proc with 1333FSB so I think it's safe to assume the default settings will be 333FSB with a multiplier of 9. I'm not sure if I can go above that multiplier, so the next available thing is to increase the FSB, but then I read that it also affects the PCI/PCI-E slots, and I have no idea what to do about those..

Then there's a few articles saying it's best to match the DDR FSB rating with the CPU's core FSB (So with 333 I'd get DDR 667, and DDR 800 if I OC CPU to 400 FSB) and other articles saying it's better to have the RAM speed one 'notch' higher (DDR 800 with 333FSB, DDR 1066 with 400FSB, etc) Which one is actually correct? I thought matching would've yielded best results, but I'm not as knowledgeable in this area. The timings are another mess, but I don't think I'll want to touch those, I'll just try to buy ones with the lowest timings available.

Sorry for the long post, but I'm quite confused about how everything works together during OCing, and I don't want to fry anything without knowing what they do first.

Thanks in advance

 

[MadScientist]

Read Graysky's thread Memory Bandwidth Tests and then you decide whether you want to run 1:1 or higher.

 

[DarkRogue]

I read that thread actually, but to be honest, I'm not quite sure what was going on lol.

From what I've read, the recommended divider is 1:1 I think. Then I also read elsewhere that Intel (processor or chipset, this I don't know) stuff works best if the FSB speeds of the CPU and RAM were both matched up. I'm not going to try to overclock the RAM, I'm simply trying to figure out what type of RAM I need to buy with the upcoming Wolfdales for best performance. The CPU is what I'm going to try to overclock a bit, maybe bring FSB to 400 and lower the multiplier a little bit so that it corresponds with DDR800 but then other people say if I were to have 400FSB I should go with 1066 DDR so I'm confused.

Then there's the other problem that I think something about the PCI-E or PCI speeds going up with you raise the CPU FSB.. No idea what that means. Just would like some clarification or advice about it.

 

[BonzaiDuck]

To some extent, choosing a ratio is subjective. We all agree that 1:1 is better, because there are "no dead clock-cycles" -- the overall memory latency is reduced.

On the other hand, some other ratios allow clocking of memory at a higher speed that the CPU, using a lower voltage than might be necessary for the tightest latency settings allowable at 1:1.

The various bandwidth benchmarks are a basis of comparison. But they're synthetic benchmarks.

According to Anandtech's article on OC'ing the QX9650, 4:5 CPU-to-RAM ratio is the next best thing. For myself, I confirm this through benchmarking. I break even or actually score an increment higher in the benchies if I can tighten the latencies at the higher RAM speed, and I can drop the voltages a notch. I'm still testing, but I think I can drop my VCORE one notch, and it is confirmed that I can drop the VDIMM a notch, while I can keep the same CPU speed, and get a slight edge in the memory-cache performance benchies.

But it also may depend on the choice of processors. Running a dual-core, I might be more inclined to stick with 1:1 at memory speeds between 700 and 800, with the lowest possible latency settings.

If you use DDR2-1000+ modules, you might find a different regime of settings, but I imagine the same principle would apply.

 

[DarkRogue]

Okay, the voltages thing is another thing that I'm unsure about lol. What's the general rule when it comes to that, lower is better?

From what I can tell, it seems increasing the FSB of the E8400 to 400 and couple that with DDR2 800 with a divider of 1:1 seems to be best. Some say using DDR2 1000 will give slightly better performance, but it seems to me there will be "dead clock-cycles" as you call them. If I recall correctly, increasing FSB also increases the bus speed of the PCI slots which are supposed to run at 33MHz.. I'm not sure how to handle that.

Also, most processors' multipliers are "locked." Does this mean it is stuck at 9 for the e8400 or does it just mean I can't raise it above 9 (meaning I can lower it to 8 or 8.5?)

That's as far as I can understand, after that comes all the voltage adjustments which I'm unsure of how to look at the results and numbers.

Thanks again

 

[BonzaiDuck]

The problem of over-clocking PCI-E is handled similarly to the problem with the older AGP technology -- you fix the PCI-E bus-speed to its default value, or choose a safe value for over-clocking it. To my knowledge, though, I haven't heard anyone say that doing the latter offers much in the way of improvement.

You are absolutely correct about using a 1:1 ratio. In my case, I've got settings for 1:1 and 4:5. Again, it's a balancing act between L2-cache and memory speed, and the latter is a balancing act between latency tightening and memory-bus speed.

The high-end memory modules would help either way, except that at the lower end of the spectrum, I don't think latencies can be tightened very much more than the lesser DDR2-800 modules in the same model-line and manufacture.

Having burned out a Crucial Ballistix DDR2-1000 module in my system, and after reading other comments of the same experience when pushing the modules to their recommended maximum voltage, I'm more inclined to set a lower limit on my choices of memory voltage and watch the monitored value at those settings carefully.

 

[The-Noid]

Originally posted by: BonzaiDuck
To some extent, choosing a ratio is subjective. We all agree that 1:1 is better, because there are "no dead clock-cycles" -- the overall memory latency is reduced.

On the other hand, some other ratios allow clocking of memory at a higher speed that the CPU, using a lower voltage than might be necessary for the tightest latency settings allowable at 1:1.

The various bandwidth benchmarks are a basis of comparison. But they're synthetic benchmarks.

According to Anandtech's article on OC'ing the QX9650, 4:5 CPU-to-RAM ratio is the next best thing. For myself, I confirm this through benchmarking. I break even or actually score an increment higher in the benchies if I can tighten the latencies at the higher RAM speed, and I can drop the voltages a notch. I'm still testing, but I think I can drop my VCORE one notch, and it is confirmed that I can drop the VDIMM a notch, while I can keep the same CPU speed, and get a slight edge in the memory-cache performance benchies.

But it also may depend on the choice of processors. Running a dual-core, I might be more inclined to stick with 1:1 at memory speeds between 700 and 800, with the lowest possible latency settings.

If you use DDR2-1000+ modules, you might find a different regime of settings, but I imagine the same principle would apply.

Just looked at Gravsky's thread hard to really agree with a lot of his findings.

Just to point out some problems.
A.) For memory to shine at high speeds like that you need to start out with a higher default fsb. As I increase fsb I can get lower and lower latency settings, and higher bandwidth results without changing the mhz.
B.) The motherboard he uses is a P5B which does not allow you to change nb strab in bios manually. As you go up in mhz you also increase your strap. With a p35/x38 that allows you change strap and performance number you are going to be able to set a tighter latency to the nb, at higher mhz.

All in all best case scenario you are talking about 3-5% going from DDR2-800 to DDR3-2000 though. Overall I have always been one to run my memory as fast as it can though.

 

[DarkRogue]

Wait, what is this about a northbridge strap?

lol I thought there was enough stuff to worry about, but I've no idea what that is.
I'm trying to decide between a 780i or an X38/48 mobo atm, so I'm unsure if those mobos have these things, or what I need to do to get to them. That of course comes after I figure out a list of things to do to accomplish this overclocking business.

I still haven't seen much in the way of voltage recommendations, do these adjustments happen automatically when I change the FSB or is it something I must manually do along with changing the FSB? Also, what happens if you do not change the voltage after increasing the FSB?

Sorry for all the questions, getting into overclocking is a lot tougher than I expected.