PC Building and Upgrading 101

So:

Recently I've noticed a lot of people posting some truly excellent info, and I'm thinking it would be a horrible shame if all of it went to waste, so with that in mind, maybe we should have it all in one place?

My thinking is that we can have (hopefully) a sticky, detailing everything about everything to do with PC building and upgrading. nesf did a great piece on computer audio, deconduo did an excellent bit on PSUs, and partyatmygaff did one on motherboards just yesturday.

I'm thinking that we'll use this thread as a sort of assembly. People can re-post guides here (and new ones), and they can be refined and critiqued by the rest. When we've got everything, we can do a nice contents-post in a new thread, with every post after that being a separate section, to make things as easy-to-find as possible for newcomers.

The little bit I've just finished is on overclocking. Yes, I realise it might be more applicable to overclocking and modding, but there have been quite a few questions here about it too, and if it's going to be a guide about everything computer hardware, I think OCing deserves a place in there too.

Anyway, to start what I sincerely hope will be one of the most informative threads to grace the Internet: (suggestions/corrections/etc. welcome)



Overclocking
__________________________________________________


Introductory note

This is a section perhaps more suited to the enthusiasts, but it's something almost every new computer builder looks into at one point or another: Overclocking. This guide aims to give you the basics in CPU and GPU overclocking. This guide assumes that you're familiar with the components of a computer, their functions, and can find your way around a BIOS. My recommendation is that you read the whole section relevant to whatever component you’re going to overclock, in its entirety first, then start, using the guide as a reference.

Unfortunately, as regards CPU and RAM overclocking, I am only going to focus on socket 1155 systems. I may add 1366 (my own) at a later date, or if another users wants to write up a 775/AM2/AM3 version, they’re more-than-welcome, but the simple fact is it would just take me too long. Also, in particular with AMD systems, I wouldn’t be doing users any justice, as I haven’t used an AMD system in a few years now, and am not very familiar with them as of this point.


What is overclocking?

Overclocking (OCing) is the process by which computer components - be it the CPU, GPU, whatever - are made to run past their out-of-the-box speeds, thus yielding (among other things) an increase in performance. This is done by changing certain BIOS settings. More on this later.


Should I overclock?

This is something that you need to consider carefully before you go fiddling around in the BIOS too much. Is your computer slower at certain tasks than you'd like? Would you like your games to run a little smoother, or be able to enable a couple of extra graphics settings? If yes to any of these, then odds are OCing might be something to think about. If you're a gamer, and play lots of RTS or MMO games, then you'll stand to benefit quite a bit, as RTS and MMO games are among the most CPU-intensive games around. If you're more into FPS titles, you might be more suited to a GPU overclock (though a CPU OC couldn't hurt either). What you have to figure out is where you're being limited. In a gaming context, this will usually be a toss-up between the CPU and the GPU. The easiest way to do this is to leave a monitoring program open while you're gaming. For your CPU, the Windows Task Manager will display CPU usage over a small period of time. If you see your CPU usage rising to above 95% or so, then an overclock could help. A good GPU monitoring program is MSI's Afterburner. It's also the program I recommend for GPU overclocking.

Now you should think about what sort of overclock you want. Many people only overclock their processors enough so that they're sure that their graphics card(s) won't be bottlenecked. This is often the most cost-effective route, as it means you get maximum performance, at minimal cost in terms of an aftermarket CPU cooler, and your ESB bill. If you're going to be overclocking your graphics card as well, then to avoid bottlenecking, you'll want to scale your CPU overclocking with your GPU speeds. If you want see how far you can push your components, then that's fine too! Just make sure you have what you need to do this effectively and with minimal risk (following sections).


What are the risks?

There is really only one risk (that can't be rectified by a simple reset) when overclocking, and that is the scourge of all computer components: Heat. Because components are being made to run faster than they usually would, more heat is going to be generated, and you'll need to have a way to dissipate this excess. As long as you follow a good overclocking guide (hopefully this one), and have adequate cooling, the risk of anything catastrophic happening will be minimal. Having said that:

I take no responsibility for damages, etc. All users undertake overclocking at their own risk, blah blah, you get the point.

Now, on to the good stuff.


CPU Overclocking
________________________________________

What do I need?

The most important things you need are a motherboard and processor that will allow you to overclock. With Intel's newer processors, overclocking is all-but impossible on all but the K and X variant CPUs, and P67 and Z68 motherboards.

• Core i3 & i5 2100 series - Locked

• Core i5 & i7 non-K series - Partially unlocked

• Core i5 2500K & i7 2600K - Fully unlocked

If you're not sure, or have an older processor, your best bet is to Google "[processor model] overclock" and see what speeds other people are getting. If they're only slightly above the stock speed, then it's almost-certainly not overclockable. You also need to double-check your motherboard chipset, though on anything earlier than a second-generation i3/i5/i7 board, you'll most-likely be fine.

The next thing you need is a power supply that can handle the extra power draw created from the speed increase. If you've had your build specced out on Boards by someone here, then you most-likely have a high quality power supply and are fine. It's only those people who bought the €20 XtremeCoolGamerPro 800W UltraRaid XXX Edition PSUs that need to be concerned. For reference, most PSUs from manufacturers like Seasonic, Corsair, Cooler Master, Enermax, and Antec are of high quality. Check your PSU from comparisons on other forums to be sure.

Last, but certainly not least: cooling. This comes in two forms; your case, and your CPU cooler. As long as your case has at least two decently quick fans (~1000RPM+) you should have enough ventilation to cope with the extra heat. Just be sure that you have at least one fan pulling cool air in, and at least one pushing air out, either at the back, or top, of the case. That brings us to the CPU cooler. There are two main types of coolers available today. Air coolers and water coolers. An example of an affordable air cooler would be the Freezer 7 Pro Rev. 2 from Arctic Cooling. For about €20 (Q1 2012) you can obtain moderate to high overclocks, though with potentially higher-than-normal noise and/or temperature levels. Examples of high-end air coolers would be the NH-U9B SE2 and NH-D14, both from Noctua. This brings us to water coolers. These must not be confused with a watercooling setup. This usually implies a custom-made loop, comprising separate pump, CPU block, radiator and tubing. The water coolers I'll be talking about for this piece are closed-system, single piece cooling solutions, such as the Corsair H70. These typically offer quieter operation and lower temperatures than all but the best air coolers though can be more expensive.


How is it done?

The CPU’s speed is calculated by multiplying the base clock (BCLK) by the CPU multiplier. For example, Intel’s i5 2500K comes by default with a multiplier of 33. Multiply this by the BCLK (100MHz) and you get 3300MHz, or 3.3GHz. The RAM frequency is derived by multiplying the BCLK by the memory multiplier in much the same way.

The basic methodology for overclocking is to increase either the BCLK or CPU multiplier until either the system becomes unstable, or too hot. So the first thing you should try to increase is the CPU multiplier. Raise it by one, keeping all other settings the same, and then boot back into Windows. If you can successfully do this (if you can’t, you have a problem) then we can try and test the overclock for stability.

A note on BCLK overclocking though: While you can increase the BCLK, it’s problematic. Because the base clock controls everything in your system – from the memory frequency to your PCI lanes and I/O – when you increase it, everything runs faster, not just the CPU. This can lead to a lot of instability problems, and to be quite honest, is better left alone. At most, you might be able to increase the BCLK to 106MHz without affecting instability, but in the interests of simplicity (and well-rounded numbers) my advice would be to ignore it as an option entirely.

A second note on power-saving features. While in previous generations of processors power-saving features such as C1E and SpeedStep sometimes adversely affected overclock stability, this does not appear to be the case with 1155 CPUs, therefore, I’d recommend that unless you’re going for very high overclocks (5.0GHz or more) that you leave them enabled to save power.


Testing for stability

Well what everyone has to remember here is that there is a difference between "benchmark" and "stress test". A benchmark is used solely to gauge performance. You run a pre-defined test, and you get a number back. A stress test on the other hand is used for stability. You don't get any sort of rating, you just get a pass, or failure.

So, to start off: I don't bother with CPU benchmarks at all, really. I don't do any video transcoding, or anything like that, so I don't need to know exactly how fast my CPU is. As long as it doesn't bottleneck my games or applications, I don't much care, so I'm only going to talk about CPU stress tests here. One of the best - and an old favourite with many, is Prime95. This has three different tests within it, and you'll be asked to choose one upon running.
Small FFTs is usually the best to give the quickest indicator of stability. Usually, errors will show up here within an hour.
Large FFTs, tends to stress the parts of the CPU that aren't the core - cache, controller, etc. - and to be honest, I don't use it a whole lot.
The last Prime95 test is Blend. This tests everything in your CPU, as well as your memory.

The next CPU test that I use is called Intel Burn Test (IBT). This uses the Linpack algorithm, and is the same test used by Intel to bin CPUs. You want to run it at its maximum settings, for a minimum of 5 passes. If you're going for long(er)-term stability, try 20-25, but I wouldn't use more than that - Linpack isn't really something you want to leave running for 24 hours like Prime. For complete stability in an overclock, I usually use IBT for 5-10 passes each increment, as well as about 15 minutes of Small FFTs. When I've found an overclock I'm happy with, I use 24 hours of Blend, 24 hours of S-FFTs, and about two hours of IBT. If it passes all that - along with the usual games / whatever else I normally do, of course - then I deem a CPU OC stable.

Now: Temperatures. While running these stress tests, you’ll want to keep an eye on your temperatures, making sure they don’t exceed the specified maximum temperature set down by Intel (much). In the case of the 2500K for example, Intel specifies that for long-term usage, the processor should not exceed 72.6°C. What must be made clear here though, is that the temperature specified by Intel is tCase. This means it’s the average temperatures across the CPU heatspreader, not the individual cores (tJunction), which usually run hotter. Because of this, if you see cores running at 75-80°C on your 2500K (or 5-8°C above tCase for your particular processor) during a test, this is ok, though it is at the upper-limit. Expect also to see quite a difference between core temperatures, sometimes as much as ten degrees. This is normal, and is usually down to imperfect contact between the CPU, and the heatsink. This is not to say it’s incorrect; we’re talking about tiny abnormalities here. It should also be noted that IBT will usually exhibit higher peak temperatures than Prime, unless Prime is left running for 12-24 hours. HWMonitor is an excellent program for monitoring (among other things) CPU temperatures.


Tell me how to overclock already!

Fine, fine! To start off, boot into your BIOS. If you’ve tried overclocking before, it would be best to load defaults, and restore any hard drive or boot settings from there, to ensure all overclocking-related settings are as they should be.

Start by increasing your CPU multiplier by one. Save this and boot back into Windows. Run whatever stress tests you’d like, though I’d recommend a minimum of 15 minutes of IBT.

If this is stable, boot back into your BIOS and up the multiplier another notch. If it’s not, and your temperatures are still in the green, increase the CPU voltage (vCore) one notch, and test again. If at any point your temperatures have gone above the numbers outlined earlier, then you’ll need to drop the multiplier down one. Depending on what you’ve changed, you might also be able to drop the voltage down a notch.

The most important thing to remember here is that you must only change one thing at a time! If you start making a bunch of changes and something goes wrong, you’re going to have a hell of a time figuring out what caused the problem. Hopefully, if you follow the above, you’ll be able to achieve your desired overclock. It’s best to start from the beginning with a goal, rather than with a “we’ll take it as we go” attitude.


GPU Overclocking
________________________________________


What do I need?

As with CPU overclocking, you’ll need a compatible graphics card, a case capable of ventilating any excess heat, a power supply capable of supplying the necessary power, and as always, the appropriate cooling.

As far as a case goes, it requires similar traits to that of a good case for CPU overclocking. As long as you have at least one fan pulling, and one fan pushing air, you should be fine, though a side intake fan can also be helpful, particularly with multi-GPU setups. In fact, I’d go so far as to say it’s a requirement if running SLI or Crossfire.

A good PSU is a particular must for GPU overclocking, as high +12V amperage is one of the main characteristics of a good quality power supply. Most modern high power graphics cards such as the HD7970 or GTX 580 (Q1 2012) would require about 40A to leave enough room to have enough comfortable headroom to cope.

As for the GPU cooler, on most cards, a fairly high overclock can be achieved with the stock cooler (though it will be loud as hell). Many companies ship cards with custom coolers, that are often more powerful and quieter than their reference counterparts. Though these keep the cards cooler – and are therefore better for our purposes – they have one big disadvantage: They dump virtually all of that heat back into the case. As you can imagine, having a good case becomes even more helpful here.


How is it done?

For our overclocking, we’ll be using MSI Afterburner. Though released by MSI, this application will work with all cards from AMD and NVIDIA, regardless of brand. With this program, you can monitor almost everything about your card - GPU temperature, memory usage, framerate, core and shader clocks, core voltage, and so on.

Overclocking your graphics card is somewhat simpler than the CPU in certain respects. When you open Afterburner, you’ll see a series of sliders on the left. From top to bottom, these control: Core voltage, core clock, shader clock, memory clock, and fan speed. We will changed some of these to overclock the card.


Testing for stability

While overclocking, you’ll want to manually set your card’s fan speed to something higher than it would otherwise be. You can either set it to 100%, or whatever the highest fan speed you can tolerate (in terms of noise) is. Every time you increase clocks or voltage, you’ll want to run appropriate tests. I don't know about AMD GPUs, but NVIDIA recently changed the type of memory is uses in its GPUs, with the result that the 4xx/5xx series GPUs almost never show artifacts now, so tests like Furmark don't really work as well as you'd think. The best test, is to play the most intensive game you can - jack up the AA if it's not stressful enough - and just keep playing. If you see drops in performance, and your temps are still ok, then it's the card throttling for stability reasons, and you know your OC isn't stable. For testing, benchmark loops or heavy gameplay will almost always be better than synthetic stress tests. Don't ask me why, they just are. As for good benchmarks and/or stress tests to use, 3DMark 11, Heaven 2.5, GTA IV, Just Cause 2, and pretty much any other modern games with a built-in benchmark is a good bet.

A word on temperatures. Generally speaking, graphics cards from both the red and the green time are fine with anything up to around 100°C as an absolute maximum. In-game, I would call 95°C your maximum acceptable-but-highly-uncomfortable temperature. Try to keep things in the 80s if possible.


Overclocking

On newer AMD and NVIDIA cards the core and shaders are directly linked, so changing one proportionally changes the other. So, we’ll start by upping the core clock 10MHz. Simply move the slider to the right, and hit apply down the bottom right. Then it’s time to open up 3DMark, or whatever other tests you're using, and check for artifacts and/or drops in performance that would indicate instability. If all’s well, and your temperatures are ok, keep upping the core clock, testing as you go.

Once you reach a point of instability, and assuming your temperatures are ok and you want to keep going, you’ll need to increase the voltage. This is where problems can arise. It is entirely possible to ruin your card when overvolting, so be sure to monitor your temperatures. In this regard, graphics cards are more fragile than CPUs. Some people will say that anything up to 1.15V is fine for NVIDIA cards, and the voltage can be upped even higher with a custom vBIOS, but I’d stick with 1.15, personally. For 6000 series AMD cards, 1.15V is also the highest possible, and what I would stick with.

Ideally, you want to be about 10-20MHz (at minimum) below the lowest instability point. Riding the very edge of the line the whole time you're gaming isn't a good idea for the card's lifespan, so you're better off giving it a bit of breathing room.


Useful Links

Intel ARK – CPU Specifications: Thermal and voltage limits
HWMonitor – PC component voltage and temperature monitoring
MSI Afterburner - GPU monitoring and overclocking

Prime95 - CPU stability test
Intel Burn Test - CPU stability test
3DMark 11 - PC benchmark / GPU stability test

3 Step Overclocking – Overclocking guides from TechREACTION
Overclocking & Modding - Boards.ie Overclocking forum.