Thinking of Overclocking? Read this!

[Mr Techguy]

Many problems have been created by novice overclockers. I hope this little tid bit of advice can help. Most of the problems stem from a mild overclock causing system instability to a lack of good POST. Now why would a mild overclock cause problems with the same chip that has been oveclocked aggressivily by others? Since the invention of the "locked CPU multiplier", overclocking a chip is accomplished by increasing the system bus or front side bus. But when this FSB is kicked up a knotch, the user will effectively increase the FSB for the entire system. Every component that is attached to that bus is overclocked. For example, I have a sound card that is plugged into a PCI slot. PCI runs at 33mhz. I increase the FSB by 20mhz, now the sound card is running at 53mhz. RAM. I have DDR2 ram running at 800mhz, or 400mhz each. I increase the FSB 20mhz, now each stick runs at 420mhz or DDR2 840mhz. I have an E6750 running at 2.66 ghz, 333 FSB, 8x mult. I want to run my chip at 3.4ghz. I need to increase my FSB from 333 to 425. That is a medium jump for an E6750 but think of your DDR2 800 RAM. 92mhz jump brings each stick to 492 or DDR2 984mhz. Now if you had 1066 RAM, this would be under. Some motherboards can stop this from happening, cheaper boads may not stop this. My mother board will allow you to unlink RAM and Proc FSB. Which is what needs to happen for crazy overclocks.

 

[vonwombat]

Most motherboards let you set the FSB:RAM ratio and limits for the PCI and PCI-E. But some people don't know that, first time I tried to overclock my CPU by setting thr FSB to 400 I forgot to set the FSB:RAM ratio to 1:1, and let the setting on default so the memory was running at 1200Mhz, and my system wouldn't even POST.

 

[Mugz]

Two points I'd like to raise.

One: So what else is new? This has been known since the days when people were replacing 16MHz clock crystals in their 80286- and 80386-based machines with 25MHz crystals, and that was about twenty-five years ago.

Two: Like vonwombat said - some newer boards let you lock the different bus speeds.

Have fun!

 

[skyguy]

If you plan on overclocking, then spend an extra $30 and get a decent mobo that will let you lock the PCI bus, etc. Otherwise, don't bother OC'ing. End of story.

Any reasonably decent mobo will let you lock it. So lock it. I could push my 6300 to a 498 FSB no problem. Why? Locked the bus and had good RAM and cooling. Easy. No crashes, no problems, no fuss or muss.

And this is NOT to be confused with unlinking the FSB:RAM ratio. These are 2 completely separate things. Unlinked ratios are not needed for great overclocking.....just overclockable RAM.

 

[SpeedyVV]

And this is NOT to be confused with unlinking the FSB:RAM ratio. These are 2 completely separate things. Unlinked ratios are not needed for great overclocking.....just overclockable RAM.

skyguy,
As a n00b OCer, I wonder about this. I use my evga 680i to OC my e4300 to 3.0GHz, but I unlinked the FSB:RAM, and left the RAM at 800 as per corsair spec.

Any advantages to OCing the RAM? If so, should I link it, unlink it and manual up the RAM speed, or use some sort of ratio??

 

[MikePHD]

You can get better bandwith on your FSB with OCed ram. I have my 800mhz gskill running at 1066 no problems. If I go much higher than this then I have to loosen timings, but Ive been up to 1200+.

 

[SpeedyVV]

You can get better bandwith on your FSB with OCed ram. I have my 800mhz gskill running at 1066 no problems. If I go much higher than this then I have to loosen timings, but Ive been up to 1200+.

Mike

I have done that before, but had no impact on 3dmark06 or pcmark so I left it alone at 800. Is there any type of app that will benifit from OCed RAM?

 

[skyguy]

There are apps that will benefit, but the question is Will you notice a real world difference? 😉

You will get faster benchmark times and higher scores, but will those translate into meaningful real world improvements? And if so, how much? 3-5% AFAIK. So when you extract a .rar file, for example, or apply a Photoshop filter, instead of it taking 1:30 seconds, it might go 2 seconds faster.

So, does the hype match the return?

Now therein lies the real question.......... 😉

CPU speed and cores are KING. Apps want to suck up the speed and the horsepower. Any performance increase in bus speeds, etc will far outweigh any tweaking to timings and latency, this is fact. If someone is arguing nuances about these, then they have no life, or are in the wrong field because it's minutae......it's like arguing how much faster a 480 hp sportscar will turn in a 0-60 sec time than a car with 475 hp. Foolishness......

 

[Hatman]

It does make a difference, tehres a reason why we went from ddr to ddr2 to get extra speed, and are now going to ddr3.

It does help but only as far as your bandwidth goes. Faster ram should help get slightly faster loud times on programs, more ram will let you run lots more things in the background with your dual/quad cores.

In the future we're going to be running ram at 1:2 anyway, since DDR3 clocks are so high.

 

[MikePHD]

Tighter Timings actually help a bit more than you give credit skyguy. A super fast CPU will get you nowhere if it sits around waiting for memory all day long.

I never understand the people who say, no thanks, I don't want any more power, I'm good right here.

The thing is, it takes minimal time to figure out what stable settings/timings your memory can handle, even if you don't want to overvolt/overclock your modules any more. In return, you can shave a few seconds off applications here and there.

 

[skyguy]

Yes Mike, good point. Tighter timings do help, that is definitely true, but higher bandwidth and latency (specifically as a result from a higher FSB) will trump the performance gains from timings.

I should have clarified that a bit better: timings do help, but not as much as the higher clocks (generally speaking).

Thanks for pointing that out Mike.