- Joined
- Dec 4, 2002
- Location
- Hell's Kitchen
I wanna try to do my part to help out the overclocker community so here's some advice I have garnered over the years of overclocking. Anyone with more/other knowledge than me, please feel free to add to this thread.
1. Where to start?
RESIST THE URGE TO OVERCLOCK EVERYTHING IN YOUR SYSTEM AT ONCE. When you're dealing with a fresh system, you don't want to overclock everything; have your system crash; then fish around your hardware to try to figure out what the hell went wrong. Since the CPU is more or less the heart of a system, starting there is as good as place as any. Set your memory timings to the most relaxed as possible and set your memory dividers to the lowest your motherboard can go. Then, start cranking up the FSB (or, in conjunction, changing multipliers for you AMDers).
A lot of magazines/hardware sites recommend overclocking 5 MHz at a time. Personally, I think this is a waste of time and makes overclocking a complete hassle. A good way of going about an FSB to start with is to do a search on these boards and try to figure out what a reasonable expectation for your particular chip is. For most people, reasonable will NOT include basing your CPU on a guy who is running a Prometeia with a Northwood running on a vmodded motherboard @ 2.05 v. Do a search, compare your cooling with other people's and come up with a modest estimate of where to start.
2. How far should I overvolt?
Well, how much money you got? This is what it really boils down to. If money/time is no object to you, then by all means, keep pumping juice to your system until it begs for mercy. But for most sane people out there, money IS an issue and should not be taken lightly. Most of what people around these forums consider safe voltage comes from anecdotal evidence. In other words, person A will hear that x voltage has caused problems for person B, C, and D. Again, doing a search on these boards will give you a pretty good idea as to what voltage is considered safe, and which voltages are considered more risky. In general, most people consider a 5-10% hike fairly harmless whereas 15-25% is considered risky. Once you hit a wall with your FSB and the voltage you feel comfortable with, try turning the voltage down until your system becomes unstable again, at this point, turn it back up to where your system was crash free.
3. What's stable?
It depends on the sub-system. Since we started with the CPU, let's start there. The most generally accepted form of testing for CPU stability is Prime 95 which can be downloaded here. Prime95 is probably one of the more stressful/robust tests you can submit your CPU to. I'm not much of a pro on chip engineering but I believe Prime95 stresses the FPU unit on your processor. Overclocked and unstable chips will produce errors while running Prime95 - usually in the form of round off errors. 24 hours of error free Prime95 running is sufficient enough to determine if a CPU is stable or not. As of yet, there have been NO reported errors in the software with Prime95 so if your system spits an error out, it's your hardware and not the code.
Now, here's a sticky point with Prime95. If your system produces, say, 3-4 errors during a 24 hour Prime session, does that mean your system is prone to blue screens? Probably not. Prime is a lot more stressful to a CPU than about 99.5% of the programs you'll ever use. Theoretically, you could run a system that isn't 100% Prime stable and not notice a difference while gaming, browsing, or doing the millions of other things people do with their computers. Whether or not you want a computer Prime stable or just stable enough is really your call. Personally, I wouldn't run a system that wasn't 100% Prime stable and I certainly would not try running a CPU intensive progrqam such as Mathematica that wasn't 100% Prime stable. That being said, if you're comfortable running a system that isn't totally stable, you'll probably only gain a handful of more MHz before your system will be so unstable, it'll start to detrimentally effect your computing experience as a whole.
4. What's next?
Now that you've isolated your CPU and have determined a stable clock rate for it. You can move on to another subsystem. From here, I usually start fiddling with my memory system. I usually start setting my memory timings to the most relaxed (Usually 3-3-3-7) and then start messing with the memory dividers. Once you hit a wall with the memory clock speed, start setting your memory timings more aggressively. Since all motherboards have slightly different memory options, I won't go into all the available tweaks that are possible, but generally speaking, you're gonna have a tradeoff between timing and clockspeed - ie, the higher the clockspeed, the less aggressive you'll be able to be with your memory. Usually, memory that's overclocked too aggresively will cause immediate problems. Either your system won't pass the self test upon boot-up or it'll BSOD or crash in Windows. I usually test my memory with the memory bandwidth module in SANDRA. You can use the Burn In module there and set it to loop infinitely. Again, a 24 hour run without crashes is typically sufficient enough to determine stability.
5. Video
Since running video benchmarks rely on other subsystems in your computer as well, I usually save overclocking my videocard for last. 3DMark2001 and 3DMark2003 is good for testing for stability here. If you haven't tested sufficiently for your memory or CPU at this point, errors may turn up and you'll have to backtrack to figure out what went wrong. But, if you've done a thorough testing of your other systems, 99.9% of the time, you can isolate the problem to an over-overclocked videocard. Again, loop both benchmarks for a 24 hour time period to determine your card's stability.
6. Other stuff
Now, this guide was written with the assumption that you have a system that has locks for your PCI/AGP slots. If you motherboard doesn't have this feature, then you may want to start overclocking your system with everything except your videocard ripped out. From there, you can start adding cards to see if any of your hardware is having problems with high FSB speeds.
Hope this helps!
1. Where to start?
RESIST THE URGE TO OVERCLOCK EVERYTHING IN YOUR SYSTEM AT ONCE. When you're dealing with a fresh system, you don't want to overclock everything; have your system crash; then fish around your hardware to try to figure out what the hell went wrong. Since the CPU is more or less the heart of a system, starting there is as good as place as any. Set your memory timings to the most relaxed as possible and set your memory dividers to the lowest your motherboard can go. Then, start cranking up the FSB (or, in conjunction, changing multipliers for you AMDers).
A lot of magazines/hardware sites recommend overclocking 5 MHz at a time. Personally, I think this is a waste of time and makes overclocking a complete hassle. A good way of going about an FSB to start with is to do a search on these boards and try to figure out what a reasonable expectation for your particular chip is. For most people, reasonable will NOT include basing your CPU on a guy who is running a Prometeia with a Northwood running on a vmodded motherboard @ 2.05 v. Do a search, compare your cooling with other people's and come up with a modest estimate of where to start.
2. How far should I overvolt?
Well, how much money you got? This is what it really boils down to. If money/time is no object to you, then by all means, keep pumping juice to your system until it begs for mercy. But for most sane people out there, money IS an issue and should not be taken lightly. Most of what people around these forums consider safe voltage comes from anecdotal evidence. In other words, person A will hear that x voltage has caused problems for person B, C, and D. Again, doing a search on these boards will give you a pretty good idea as to what voltage is considered safe, and which voltages are considered more risky. In general, most people consider a 5-10% hike fairly harmless whereas 15-25% is considered risky. Once you hit a wall with your FSB and the voltage you feel comfortable with, try turning the voltage down until your system becomes unstable again, at this point, turn it back up to where your system was crash free.
3. What's stable?
It depends on the sub-system. Since we started with the CPU, let's start there. The most generally accepted form of testing for CPU stability is Prime 95 which can be downloaded here. Prime95 is probably one of the more stressful/robust tests you can submit your CPU to. I'm not much of a pro on chip engineering but I believe Prime95 stresses the FPU unit on your processor. Overclocked and unstable chips will produce errors while running Prime95 - usually in the form of round off errors. 24 hours of error free Prime95 running is sufficient enough to determine if a CPU is stable or not. As of yet, there have been NO reported errors in the software with Prime95 so if your system spits an error out, it's your hardware and not the code.
Now, here's a sticky point with Prime95. If your system produces, say, 3-4 errors during a 24 hour Prime session, does that mean your system is prone to blue screens? Probably not. Prime is a lot more stressful to a CPU than about 99.5% of the programs you'll ever use. Theoretically, you could run a system that isn't 100% Prime stable and not notice a difference while gaming, browsing, or doing the millions of other things people do with their computers. Whether or not you want a computer Prime stable or just stable enough is really your call. Personally, I wouldn't run a system that wasn't 100% Prime stable and I certainly would not try running a CPU intensive progrqam such as Mathematica that wasn't 100% Prime stable. That being said, if you're comfortable running a system that isn't totally stable, you'll probably only gain a handful of more MHz before your system will be so unstable, it'll start to detrimentally effect your computing experience as a whole.
4. What's next?
Now that you've isolated your CPU and have determined a stable clock rate for it. You can move on to another subsystem. From here, I usually start fiddling with my memory system. I usually start setting my memory timings to the most relaxed (Usually 3-3-3-7) and then start messing with the memory dividers. Once you hit a wall with the memory clock speed, start setting your memory timings more aggressively. Since all motherboards have slightly different memory options, I won't go into all the available tweaks that are possible, but generally speaking, you're gonna have a tradeoff between timing and clockspeed - ie, the higher the clockspeed, the less aggressive you'll be able to be with your memory. Usually, memory that's overclocked too aggresively will cause immediate problems. Either your system won't pass the self test upon boot-up or it'll BSOD or crash in Windows. I usually test my memory with the memory bandwidth module in SANDRA. You can use the Burn In module there and set it to loop infinitely. Again, a 24 hour run without crashes is typically sufficient enough to determine stability.
5. Video
Since running video benchmarks rely on other subsystems in your computer as well, I usually save overclocking my videocard for last. 3DMark2001 and 3DMark2003 is good for testing for stability here. If you haven't tested sufficiently for your memory or CPU at this point, errors may turn up and you'll have to backtrack to figure out what went wrong. But, if you've done a thorough testing of your other systems, 99.9% of the time, you can isolate the problem to an over-overclocked videocard. Again, loop both benchmarks for a 24 hour time period to determine your card's stability.
6. Other stuff
Now, this guide was written with the assumption that you have a system that has locks for your PCI/AGP slots. If you motherboard doesn't have this feature, then you may want to start overclocking your system with everything except your videocard ripped out. From there, you can start adding cards to see if any of your hardware is having problems with high FSB speeds.
Hope this helps!
