I have a motherboard that does OC for me, should I use that or do it manually? Also, when I increase the frequency do I lower the voltage?
If you are overclocking, you will eventually want to increase the voltage, rather than decrease it.
You can always try the motherboard's auto OC if you know nothing about OCing. However, you will almost always get better results with a manual OC (lower temps and higher OC frequency).
You should start with a guide specific to your processor and, if possible, your motherboard/motherboard chipset. If you tell us what you have, someone here with similar or identical hardware may be able to offer you specific guidance. If you are new to this and want to do things manually, be careful with the voltage. Any voltage.
It would help if you told us which processor you're using - namely, the brand and part number. There's a lot of CPUs out there running at 3 ghz stock. It would also help to know which motherboard (again, brand and part number).
For example, my machine is an old AMD Sempron 140 OCed to 3.7 ghz (2.7 ghz stock). It's in an MSI 790FX-GD70 motherboard.
Tell us something like that, and we can go from there. It will be hard to recommend a cooler if we don't know the socket in your motherboard at a minimum.
You can always try the motherboard's auto OC if you know nothing about OCing. However, you will always get better results with a manual OC (lower temps and higher OC frequency).
ASRock n68c-s ucc AM3
Phenom II x4 945
Hey! Great. You have a CPU old enough for me to know something about it from first-hand experience. I've never used that board, but you'll figure it out soon enough.
Here is your CPU in good detail. Some important details from that page:
V core 0.875V - 1.425V
That tells you voltage range for the chip during normal operation (the "default" value is 1.35v for your chip, but it jumps around thanks to power saving features and other stuff. Interestingly enough, I don't think your processor will ever reach 1.425v during normal operation. It doesn't have turbo). Vcore is the voltage setting that is most important. We'll go over that below.
Bus speed 667 MHz Memory controller
One 2000 MHz 16-bit HyperTransport link
That tells you two things in an awkward sort of way: Your memory is supposed to run at DDR3-1333 speeds (default), and your Hypertransport link is supposed to run at 2000 mhz.
Incidentally, your Northbridge (NB) - an important part of the processor that has the memory controller in it, among other things - should also be running at 2000 mhz @ stock.
One last, and very important, speed: HTT, also sometimes called FSB or other things. Your default HTT speed is 200 mhz. All the speeds of your other components (CPU, RAM, NB, and Hypertransport) are based off the HTT speed by way of multipliers.
CPU: 3000 mhz (15 x 200 mhz)
RAM: 1333 mhz* (8 x .83375 x 200 mhz)
NB: 2000 mhz (10 x 200 mhz)
Hypertransport: 2000 mhz (10 x 200 mhz)
*Your RAM isn't actually running at 1333 mhz, it's really 167 mhz. Sometimes memory multipliers appear as "ratios" which we'll discuss later.
If you mess up anything, make sure your speeds in your BIOS are reset to those values. The motherboard should have a way for you to "restore defaults" that will do just that, but it pays to know them by heart anyway.
Now that we've covered some basics for your processor, you need to get your feet wet with a manual overclock. The easiest way to deal with manual overclocking is to disable all power saving features on your CPU: namely, Cool n' Quiet (CnQ). CnQ is hard to balance with overclocking. It can be done, but for many overclockers, it is not worth the trouble.
Once you've done that, we'll try a stock voltage overclock. This means we are not going to raise the voltage on anything to make sure you cause minimal damage to your system. It's theoretically possible that you could cause some data corruption on your harddrive(s) with an unstable overclock, but that is about the only kind of damage you would cause. Well, I've seen boards complain about bad BIOS checksums also, but you can usually get around that.
So, now you must familiarize yourself with resetting your motherboard's BIOS settings in case you try to boot with unstable settings and your motherboard can not recover. Sometimes this happens, and you just have to dust yourself off and try again. There is an important thing called the Clear CMOS Jumper which is shown on page 11 of your motherboard's manual. Learn how to use this. Practice if you must. My board has a handy little button I can push to do the same thing as using the clear jumper, but it's all the same thing really. Page 20 of the same manual will tell you which pins need to be covered by the plastic doohicky to reset all your BIOS settings. If you do not know how to use this jumper, here are some basic instructions:
1). Power your system down. Unplug it and press the power button for a few seconds to discharge capacitors and what not.
2). Make sure you don't zap anything with static by remaining in contact with your case (ESD Strap, touch the case with your other hand, yada yada ya).
3). Move the plastic thingy onto pins 2 and 3 and wait a few seconds (your manual says 15).
4). Move the plastic thingy back where it was before. Note: You must do this before turning your system back on!
5). Plug in your system, power it up, and get back to overclocking.
Now that you know about that thing, it's time for you to download some stability testing tools to let you stress-test your overclocked system and check temperatures during testing. I use some tools that are probably considered out-of-date, but since we have about the same kind of CPU, they should work just as well for you:
Prime95: Run any of the three stress tests. They're all pretty good. I recommend Large FFTs to check to see how hot you can get your processor, and Blend to test for stability. More on this later.
LinPack (and the front-end, LinX): I don't know where to download this anymore, and you may need some help learning to use it. Make a separate thread here if you must, or look at this thread.
Memtest86+: There are 2495902348509840 different versions of this tool floating around, but they should all work for you pretty well. You need to run it from some kind of a boot disk, which is easy to do if you have a bootable Linux install on a thumb drive or something or other. Usually bootable Linux distros have MemTest86+ as a boot option. Well, lots of them seem to anyway. You can make your own boot disk to run just MemTest86+ if you must.
. . . and that's about all I use for CPU/RAM testing. Others may recommend different tools. We'll go through how to use Prime95 and MemTest86 later.
edit: finishing post
To monitor your CPU/system temperatures during testing in Windows:
CoreTemp: Lets you know temperatures for your CPU on a per-core basis, and some other stuff. Pretty neat. It's a bit out of date (last update: oct. 2013), but then so is your processor. And my processor. Heh.
SpeedFan: This program is so well-liked that it gets mentioned in some books (really). It can do lots of nifty stuff, but most importantly, it will display temperatures from lots of different temp sensors in your system.
These programs are pretty easy to use for temp monitoring. Just start them up and they'll tell you what the sensors are reporting. Very useful while stress-testing your CPU.
Okay! Now down to the nitty gritty. We're going to use an old-but-good technique called Isolate and Consolidate. What you are going to do is try to figure out how fast your CPU, RAM, and NB alone can go without making anything else go faster at the same time. Then you try to get all three as close to those speeds as possible at the same time. We'll start with your RAM, then go to your NB, and finally your CPU.
FIRST STEP: We are NOT overclocking your Hypertransport link. This provides no real benefit. Set your Hypertransport multiplier to something lower than 10x. Many AMD machines have Hypertransport multipliers that bring about instability, but it's hard to know which ones will be bad for yours. Still, 7x/8x/9x/10x should all be safe. I would say try 7x for this stage. Set it there for a Hypertransport speed of 1400 mhz @ 200 mhz HTT, and leave it there during the RAM, NB, and CPU steps.
RAM: Since you are a noob it seems, leave the timings alone. Set your CPU multiplier to 10x, and your NB multiplier to 7x. Then, start raising your HTT speed 5 mhz at a time and try to boot into Windows. Keep doing this until Windows refuses to boot or until your machine refuses to boot, period. Once you find out which HTT speed is the one that makes Windows (or your computer) not want to start, back it off to 5 mhz below that speed, boot into Windows, and try running Prime95 Blend for about 10 minutes. If it doesn't crash, reboot into MemTest86+ and run all nine tests about 5 times each (it should start automatically, and you shouldn't have to configure anything). If you are confused about how to make a bootable disc/device that can load MemTest86+, try reading this thread, especially near the end, as one example of a howto.
Anyway, if it clears MemTest86+ 5 times, congrats, you have a semi-stable RAM overclock. Note the DDR3 speed you achieved somewhere and move on. If Prime95 crashes or MemTest86+ reports errors, back your HTT off another 5 mhz and repeat both tests again until they both succeed as listed above.
NB: Keep your CPU multiplier set to 10x, and set your RAM to the DDR3-1066 setting, which may appear as a ratio like 3:8. Restore your NB multiplier to 10x, and then start raising your HTT in increments of 5 mhz like with the RAM. Run the same tests you ran on the RAM (Prime95 Blend, then 5 loops of all 9 tests in MemTest86+). You're following the same procedure here as you did when testing your RAM. If you get crashes/errors during testing, back your HTT off 5 mhz and try again.
CPU: Keep your RAM set to 3:8/DDR3-1066, and put your NB multiplier back to 7x. Set your CPU multiplier to 15x. Raise HTT in 5 mhz increments until Windows or your computer won't start anymore, and then go down 5 mhz, boot windows, and run Prime95 Small FFTs for 10 minutes (or Blend, or run LinPack for 10 minutes, your choice). If it passes 10 minutes of stress testing, congrats, you have a semi-stable CPU overclock.
Now the tricky part: You want to get to your semi-stable CPU speed while keeping your RAM and NB speeds as close to the semi-stable speeds you achieved in the above testing. Usually the emphasis is hitting the CPU target, though if you're like me, you'll try to hit the NB target instead (but I'm weird . . . or am I???!?). Let's do an example:
Pretend that your RAM crapped out at HTT speeds above 215 mhz (yuck), giving you DDR3-1432. Then, your NB managed an HTT speed of 240 mhz, giving you an NB speed of 2400 mhz. Finally, your CPU managed an HTT speed of 240 mhz, giving you a CPU speed of 3.6 ghz (btw, this would be a fairly nice OC without raising voltage). So getting the CPU and NB to sync up would be easy, since all you'd have to do is set HTT to 240 mhz and you'd be done . . . right? Well, not so fast. The RAM won't play ball at HTT speeds above 215 mhz if you use the default DDR3-1333/ 5:6 ratio. So set the RAM to DDR3-1066/3:8 ratio; then you can push your HTT up to 240 mhz, and your RAM will only be running at DDR3-1280 (8 x 240 x .667).
In real life, you will probably see things being a little more complicated than that. The easiest thing to do is use the 3:8 ratio if your RAM is being a pain-in-the-butt, because your CPU and NB have so many multipliers available that you can probably dial those in together pretty well.
Once you have your target HTT speed, CPU multi, NB multi, and RAM ratio (which will probably be 3:8 unless your RAM kicks ass), boot with these settings. If it won't boot windows or boot at all, back the HTT speed off by 5 mhz and try again. Repeat this until it works. Go into Windows, and run:
Prime95 Blend: Run this for at least 4 hours. I recommend 24 hours. Be advised that you can do casual web browsing and other stuff while Prime95 is running, but don't do it all the time - loading other programs will actually relax the stress from the CPU a bit. Check your temperatures. If you see your CPU going over 50C, then you've got trouble, and you'll need a better heatsink/fan (HSF) or at least a better mount on your current HSF. Stop all overclocking because you're overheating at stock voltage, which is bad, and check your mount. Is it loose? Too much thermal paste (TIM)? If it passes, move on. If it crashes, back off your HTT by 5 mhz and repeat.
Prime95 Large FFT: Run this for about 30 minutes to an hour and watch temps as above. If you go over 50C, there's a problem you need to address, stop all overclocking etc.
LinPack: Optional, but run this for 4+ hours (24 hours preferably), and set it to use all your available memory. Watch your temps! They may go higher here than in Prime95 Large FFTs.
Then, reboot with your MemTest86+ disc and loop all nine tests for 4+ hours (24 hours ideally). If it produces errors, back your HTT speed off by 5 mhz and repeat.
Once you have achieved these milestones:
1). Running under 50C under LinPack or anything else
2). Not crashing under Prime95 Blend or LinPack
3). Not giving any errors in MemTest86+
. . . then you have a pretty stable overclock. Of course, this takes time, so prepare accordingly. Save a BIOS profile with stock settings if you need to use your computer for something in the meantime but you don't want to clear EVERY setting, and another profile with your testing target. That is, if your BIOS supports profiles.
Once you have done all that, come back to us and we'll talk about RAM timings, voltage tweaks, and aftermarket HSFs. Whoo nelly! Ain't that a long post.
Good deal. Yeah it's a lot to take on at first, but if you can get all that stuff down pat, then you can OC just about any AMD processor from the first Athlon 64s all the way up to the Thuban hexcores. Newer AMD chips are a little different.
Overclocked it to 3.3GHz, didn't touch the voltage and it was unstable only when it ran at 62C, but is it because of the temperature or the voltage? I could only OC it by using the motherboard feature, though all it does is raise the clock limit.
If I get a new heatsink, should I get some TP for it too?
First off, if you didn't touch voltage, then there's no way your CPU should have been running at 62C. The stock cooler isn't THAT bad. Clock your CPU back to stock and try running LinPack or Prime95 Large FFTs (for the heat) and check the temps. Still 62C or so? That's wayyyy too hot, and something is wrong with your mount, or your case airflow situation, or something.
Those coolers won't be any improvement on the stock one, I'm afraid. (In fact the MASSCOOL one looks slightly worse than the stock cooler.) I would recommend waiting and saving up until you can afford something like the Cooler Master 212, though I'm not an expert on CPU coolers- there may be a better value offering worth your time.
The heatsink from an FX chip should actually be better (or at least as good as) the stock heatsink for your x4 945, and the mounting system should be the same. Did you remember to use thermal paste (TIM), and did you apply it correctly?
Using a stock HSF properly mounted with generic TIM shouldn't result in temps over 45-50C at full load, and that would still be a bit much. Make sure the application of the TIM is thin, and . . . well hold on:
Here is a video tutorial on removing and applying TIM. I can't vouch for it being the best method ever (you can use coffee filters instead of those cloths from Staples), but if it helps you get a very thin application of paste without air bubbles, then that would be a good thing. Note that the guy uses wayyyy too much TIM there, I use less.
Your CPU should never hit 62C under load. No fan in your case? You may be having a bad (as in no) airflow situation. So you are saying, your case has no fans, period?
If you have a desk fan handy, try this experiment: pull the side panel off your case (assuming small/mid-sized tower) and have your desk fan blow into the case at an angle across the major motherboard components/RAM/heatsink. See if that brings down your temps at load. Or, heck, even without a fan, just pull off the side of your case and see if that improves things. If it does, that definitely signals poor case airflow.