Overclocking is futile without good cooling
Oct 02 '00
Successful overclocking is like anything else in life; the more involved the process, the more the process seems like it involves some sort of black magic and a great deal of luck to the uninitiated. While in the past it required a bit of luck finding the right CPU for overclocking, more and more we are seeing not only higher quality yields from Intel and AMD, but vendors that are handpicking and even pre-testing parts to sell specifically to overclockers. Unfortunately, buying handpicked parts carries a premium... Overclocking isn't mysterious, nor does it require any luck--it relys on attention to detail more than anything else.
The first thing we need to do is to calculate how much cooling we are going to need. The ultimate speed at which you'll be able to overclock is heavily dependant on the cooling that you provide to not only the CPU itself, but the entire system.
I am going to split this into 3 areas; The CPU, the case and the hard drives. I've seperated it this way because expansion cards rely heavily on effective case cooling and HDD's are frequently out of the case airflow circuit and sometimes require additional measures. I could argue with myself and say that in any good overclocker's system, the vid card is overclocked as well and also requires specific cooling. *But* anything that you do to cool a video card affects case cooling as well (with the possible exception of liquid cooling the vid card as well), so....
CPU Cooling
A good place to start when figuring out how much heat a CPU will give out is this article by Joe Citarella (http://www.overclockers.com/tips40/). It's got most of the popular Intel and AMD chips as well as an Excell spreadsheet to help figure out further calculations of processors that aren't discreetly listed in the list.
Now that you know how much heat that your CPU will pump out at a given voltage and frequency, you need to decide how you are going to cool it. You can go two ways; remove the amount of heat that the CPU will pump out or remove more heat than the CPU will put out. The former is usually fairly straightforward and can be accomplished with a high-quality heatsink and fan while the latter will require much more effort and expense but will most likely result in a much higher overclock.
It's hard to estimate what the ultimate speed will be with either method, as overclocking yields vary wildly from CPU to CPU (and not just parts of different Mhz numbers either, but lot by lot as evidenced by the classic celeron300a's and 333 parts that were popular a few years ago). It's probably safe to say that 20-30% overclock is a reasonable target for aircooling. My preferred approach to overclocking is to look at the various CPU databases around the web and see which CPUs are making it to my target Mhz number--I started working alot with the Intel P3 750e as a result of this type of research. This will give you not only realistic expectations, but a fairly good idea of what type of coolin will be required to get there as well.
From my point of view, there are distinct types of overclockers, based on their CPU cooling method;
The Entry-Level Geek who uses the stock HS&F and pushes his/her CPU as fast as it'll go before lockup--This was me 6 years ago. Constantly plagued by random lockups and crashes, the system never really felt rock solid.
The Novice Overclocker--someone who gets a vastly superior HS&F combo to either make their overclock stable or eeek a little more outta the CPU.
Then comes the Veteran overclocker--someone who's tried 3 or 4 different HS&F's, reads the overclocking websites frequently, isn't able to get anywhere near the Mhz that are posted in the various CPU databases, so they take the big step into the realm of peltiers.....only to find a whole new world of problems spring up. They may or may not overcome those problems, but if the do, they step up to becomming...
The Professional Overclocker--The Guy/Gal that's been around the hardware block a few times and knows what it takes to super-cool a CPU. Preferred methods are mixed between massive air cooling with a peltier or liquid cooling. Regularly achieves +70% overclocks with most CPU's they try with. And finally, what we all strive to become...
The EXTREME Overclocker. These guys perform everything from motherboard surgery to replace the clock crystals, to totally immersing their systems in coolant--I've seen one guy even use Liquid Nitrogen to cool the CPU...*none* of the methods of the EXTREME Overclocker are for the faint of heart, nor are the methods employed very practical for use on an everyday system, but are built mainly "just because" and they are looking for the ultimate bragging rights. I am going to hopefully help you into the realm of Professional Overclocking, because frankly, Extreme Overclocking isn't that practical--while I'd love to see the looks on peeps' faces at a LAN party when I fill my CPU with LN2 and boot to a 200Mhz FSB, I just can't see it happening, ya know?
Just so you know, my preferred method of cooling a CPU is by the use of liquid cooled peltiers. There is a lot of work involved in building a system like this, but the benefit is super-cooling from a *very* quiet system. While it's been proven that massive aircooling can be just about as effective as liquid cooling if done properly, I've yet to meet a spouse that will tolerate an 80+db computer.
While a good heatsink and fan can go a long way towards getting extra performance out of your CPU, peltiers can help you go much farther. Think of Peltiers (also referred to as Thermal Electric Couples or TEC's) as nitrous for your cooling system. Peltiers are to the overclocker what lead is to the Alchemist--both are used as foundations to try and make gold. While Alchemy failed to produce gold from lead, it's quite possible to use a pelt (or several!) to get insane speeds out of your CPU. But just running out and buying a pelt and wedging it under your HS&F isn't the path to glory, and could lead you to the path of component replacement very quickly if you aren't careful. Overclocking can lead to free performance gains, but it still takes some expenditure of effort on your part to realize these gains.
Two things you need to know about using peltiers;
1)They produce alot of heat. If you don't have a hunk of aluminum big enough to dissipate not only the heat from the CPU, but the heat from the pelt, you won't be able to clock as high as you could and you could quite possibly damage more than just the CPU. A great reference for deciding what your final overclock target should be in a non-supercooled system is posted by Swiftech here(http://www.swiftnets.com/Store/cputempdata.htm). If you plan on supercooling, you need to know how much heat your CPU is going to produce and add that to how much heat the peltier produces--that's how much heat you need to get rid of at the minimum. If you are running an aircooled setup, realize also that all of that heat is getting dumped into the case and must be dealt with, or you'll get lockups from other components. Good HS&F combos come in all shapes and sizes, and some are made specifically for particular CPU's. Swiftech, Alpha and GoldenOrb are all very good, with alot of people getting great results. Again, read as many reviews of these products that you can get your eyeballs around.
2)Peltiers produce condensation if colder than ambient. While there are charts that tell you at which temp and humidity condensation will form, a good rule of thumb that will keep you outta trouble is: If it's colder than ambient, it'll form condensation. I'm positive that you know what will happen if condensation forms anywhere inside you computer, let alone near your CPU. For this reason, steps must be taken to insulate the cold part of the CPU to prevent condensation from forming. One popular way of diong this is to make gaskets from a material like closed cell foam or neoprene. This method has proven to be very effective at controlling condensation and is very inexpensive. Swiftech sells *very* good premade gaskets to fit a variety of applications and are all well under $20. My preferred method is to pot the whole peltier, coldplate, CPU and bottom of the heatsink in expanding spray foam. Pain in the ass to make a mold for and messy as hell, but I end up with what I feel is superior insulation with a guaranteed airtight seal for less than $10 on any CPU/heatsink combination.
When using peltiers, it's also highly adviseable to either make or buy a cold plate to put between the peltier and the CPU. The cold plate act as sort of a buffer from heat spikes simular to the way a capacitor deals with short term drops in current. Downside is that this adds to the thickness of the cooling device being attatched to the CPU and may make mounting quite a chore, but several companies are making products that are effective workarounds. Browse around and find ideas to make your own mounting hardware, or splurge and buy one ready made. The performance increase more than justifies the extra effort in mounting IMHO.
I feel I must insert a warning here about active cooling with peltiers and passive cooling with programs like Waterfall and CPUIdle--these programs work by effectively putting the CPU in "suspend" when the cpu isn't doing anything--since the CPU is consuming less power and producing more heat, it is effectively cooling the processor. If you are supercooling your CPU, chances are the CPU slug will freeze when sitting at idle. I've cracked the slug of a lapped c366, and I can't say for sure whether or not this was related to heat cycling, lapping or just a manufacturing defect made worse. I do believe however, that just like overclocking reduces the lifespan of a CPU, so does radical heat cycling. I now use MotherboardMonitor to heat the CPU when the on-die thermistor drops below -20f. Since my PC is, and has been on 24/7 since February, it is critical that there be some form of automated monitoring. The most spectacular burnouts and system failures come from systems that have had something go wrong while the PC was unattended. Hopefully, I won't crack my fc566... =P
Peltiers work by making a temperature differential from one side to the other. You can expect peltiers in the $30 range get about a 20degree difference on 12v. What this means is that if the hot side gets up to about 100f, the cold side will only get down to 80f. Not that great. Now if we take that same pelt and keep the hot side to about 50f, than our cold side will be around 30f, and if you can keep that temp or colder against the CPU, then you'll be able to really wring that puppy out. It's not nearly as easy as it sounds, and 20-30f is reallistically about as cold as you are going to get under load with aircooling.
Things to look for when buying a pelt. Thermal differential is obviously the most important traight in a pelt, but you also have to determine what the pelt needs as far as power and cooling to acheive the desired differential. An excellent guide for the selection of peltiers has been published here (http://www.ferrotec-america.com/3ref9.htm) and goes into much more technical detail than I wish to. Pay attention to the power requirements--while RadioShack sells pelts online, they are rated at 15v and a variety of amperages (more juice=more temp differential generally), and your power suppply might not be up to the task of powering one, let alone multple peltiers. Your PC also only provides 12v, which if you were to buy one from RadShack, you would be losing a good portion of the temperature differential.
Multiple peltiers you ask? Why yes! Where one is good, is not more better?? Well yes and no. There are several ways to to use multiple pelts to cool a cpu. Probably the most popular method is to place 2 pelts side by side, both cooling a single cold plate. The advantage to this is that the cold plate will get down to temperature much more quickly than with a single pelt, as you are pumping twice the amount of heat out of the cold plate. The downside is that you will not get any colder, you will just get to temperature quicker. Another way would be to stack the peltiers so that the cold side of one pelt is pulling heat off the hot side of the other. This creates and even greater temperature differential from cold to hot, and gets you down into the negative numbers. Downside is that the hot side is very, very hot and we now have more heat to get rid of than realistically air cooling can handle. Also, I've found that you can't quite get 2x the differential either, but that may be due to what I was using to power the pelts--I might not have been giving them 2x the juice.
Ah, so you've decided that you really *need* to run your CPU at -20, but what do you need to get there? 2 realistic options here; Vapor-phase refridgeration a'la VapoChill and Krytech or liquid cooling a couple of peltiers. I hardly recommend vapor-phase, as there are alot of problems with condensation and a VapoChill or Krytech are so expensive that they almost defeat the purpose of overclocking. Building your own vapor-phase setup isn't that hard, nor is it that expensive (less than $50 if you cannibalize a mini-fridge), but I wouldn't recommend this to someone who has any doubts of their hardware and fabrication skill. I've built a couple, and frankly I don't have the tools to make either setup into anything more than an experiment (I'm cooling dual 80w pelts with the freon and getting close to -90f!). Liquid cooling has evolved out of the garage of hardcore tweekers and somewhat into the mainstream, as waterblocks of various shapes, sizes and materials are now offered as retail items from at least 15 different companies and has been around long enough that there is *alot* of support and articles around the web on the subject. I'll focus on liquid cooling for the remainder of the article, but feel free to contact me for info or advice on other methods of cooling.
There are several ways to make a liquid heat exchanger for your CPU (called waterjackets). If you have access to a mill, then you are golden--take a hunk of metal, drill a few holes in it, attatch some taps for tubing and you're off. While designs can be as simple as a single hole through a chunk of aluminum, they can be as intricate as a serpentine channel with ridges in a peice of copper. Either way, you are looking at a difference of 15degrees in performance, maybe more depending on either the crudeness or the intricateness of the approach. A 15f difference might not seem like alot, but this difference gets compounded when peltiers are used, so it could be a signifigant difference after all. Just like aircooled heatsinks are more effective with more surface area, so is a liquid cooled one. One way to make an outstanding performing waterblock is by taking an old 486 or OEM Celeron heatsink and enclosing it ina plastic box. seal it with RTV or someother type of silicon and you're there. The advantage of using a 486 heatsink is that it will easily fit into the end of one of those PVC caps and is very easy to seal. There's also enough room left to affix the inlet and outlet for the hoses. For less than $10 worth of materials, you've got a watersink that'll perform as well, if not better than an all-metal one that you can buy. The internal volume is really tight on this type of setup, and would really be struggling to get rid of more than 180w worth of heat.
The big things that determines waterjacket performance is the flow rate through the jacket, material the jacket is made of and turbulence of the flow through the jacket. Remember that when you are making (or buying)a waterjacket, you've got alot area to work with compared to the space that a good HS&F takes up, so you usually have very little to no clearance problems. Remember also that if the waterjacket ends up being the size of an Alpha cooler or something, water is pretty heavy and could put alot of strain on your socket ears or possibly pull a slot1 card out completely. Definitely make sure you've got enough support. This should be obvious once the unit is assembled. If building your own waterblock seems like way too much work, you tend to lose appendages when working with power tools, or just don't have access to any tools, you're in luck and there are several companies making quality waterblocks for under $30. That's really not that bad a price IMHO for what you are getting and what you will be getting out of it. My favorite is TidalPool (www.tidal-pool-cooling.com) If you can get past the aweful webpage design, you'll find one of the best waterblocks that you can buy. Certainly better than anything I've built.
Now that you've got something to move heat off of your CPU or CPU/pelt combo, you need a way to get that life giving water through the waterblock and a place to put that something. Garden pumps like those manufactured by Becket are probably your best bet. They are the most widely used and can be found at just about any major retailer that has a garden section for about $40. Not exactly cheap, but if you shop around for pumps of this type, you'll find that a pump that flows with at least 90gph and 2 feet of head, you'll see that they cost around $40 from any manufacturer. Now, I've grown to almost *hate* Beckett pumps 'cause they are noisy and expensive for what you get--but they are available everywhere. A better choice I've found are inline aquarium pumps *or* bilge pumps. Why bilge pumps? Well, they are 12v and usually inline and move alot of water. I've found them for as cheap as $25 for an inline model that moved 200gph at 4feet of head--WAY better than any Beckett model.
It's also adviseable to use as large a volume of water as practical. Reason being is that the more water you have in the total system, the longer it will take to raise the ambient temperatire of the water. PVC electrical junction boxes are all the rage for this now as the plastic is easy to work with when cutting holes for the inlet/outlet, they have enough room to hold a Becket-size submersable pump, is easy to seal up and holds a fair amount of liquid in reserve. It's also small enough to fit inside all but the smallest ATX cases. Bonus! In this case too, bigger is better. The more GPH you pump through your waterjacket, the more you'll be able to remove heat. The more water you have in reserve, the longer it'll take to raise the ambient temperature of the water. You'll have to decide what the practical limits are for this yourself. An 8000gph swimming pool pump cycling the pool water through your PC would kick ass, but you'd probably end up with 8000gph streaming out of your case from the pressure. Besides, as great as it would work, a 2 inch line running in and out of a PC just isn't pracical (but definitely in the realm of EXTREME!!(If anyone manages to do this, send me pics!!)). Now, if you decide that a JBox is sufficient, you'll quickly find that it only takes about 20 minutes or so for the water in your system to raise up by 15degrees or so. As you know, the peltiers work on a temperature differential, so if ambient goes up, so does the pelts and consequently the CPU. Not good. We need a way to cool our coolant!
What we need now is a radiator. While a large volume of water might keep things cool for a while, it's not going to stay cool forever. If you live in a warm climate like I do, you could benefit from lowering the temperature below abmbient. This can be achieved by dumping ice in a 5gallon bucket and circulating that, but eventually your ambient temps will rise again. Good for a frag every now and then, but not a permanent or very convenient fix. What you need is a nice small radiator. Something that's about 5x5 is about right to smack a 100mm fan accross. It's also big enough to not only dissipate the heat coming off the CPU, but possibly cool the air a few degrees below ambient. Bigger isn't always better with a radiators. There's definitely a point of diminishing returns, e.g. throwing 2 or 3 radiators into the system won't cool the water much more than one would. Overclocking specialty retailers are already selling radiators for around $50. Ones that work just as well, and possibly better can be had at a junkyard in the form of an oil cooler. You don't necessarily need a radiator that can be bought, nor one in the formal sense either. Thinwalled copper tubing wrapped in a coil serves just as well, and for alot less money.
There are so many articles around the web, and even more websites devoted to everything from overclocking in general to case cooling mods specifically, that if you are thinking of doing a project like this you are silly not to research it further. When I started doing this, there was really nothing on the subject. Anadtech had just started and Dr.Tom was running essentially a driverguide and advice site--there really was nothing else. We kinda guessed at how to do things and started liquid cooling more because of the "What the H...." reaction I'd get from friends. As I got better and time went on, my liquid cooling expertise grew to where all of my PC's are now liquid cooled. My wife's PC which isn't overclocked that much (750@900) is liquid cooled simply because it's quieter. The power supply fan is the noisiest thing in the system-and it's damm near silent. This is a good thing after having everything from Compaq ProLiants with the extra fan kits installed to a frankenstein of a case with 10 80mm high speed Sunon fans in it in my Bedroom--ug. Never again. So. Should you liquid cool? Depends on how much effort you wnat to put into your overclock.
Case cooling
Unfortunately, not everything in a PC can be liquid cooled. I guess you could conceivably put a waterblock on every IC that gets warm in your PC, but that's not entirely practical, nor would it yield better performance the well-planned aircooling (but it would be damm cool to see!). The most important thing to think about is airflow. Look at the inside of your case. Is it cluttered with wires and ribbon cables? Are expansion cards placed close together? Is the power supply pulling air into the case instead of blowing out? Most case designs are setup so that air comes in through the front at the bottom and exits and the top of the rear. I've always had a problem with that. Draw a diaginal line from the fron bottom to the top back. The only thing that's really getting cooling from direct airflow is the CPU--the expansion card and drive bays are completely out of the circuit. Since *ALL* computers manufactured use actuve cooling on the CPU (a fan on the heatsink), there is really no need to direct airflow to the CPU. The components that now need cooling the most, hard drives and expansion cards, ar not really getting any.
There's three ways of attacking this; positive pressure airflow, negative pressure airflow and directed airflow. Positive means you are pumping air into the case, negative you are pumping out and directed is, well, directing airflow to specific spots in the case. Any of these methods can be combined (before you split hairs on me). OK. How about an example? Take your average mid-tower atx case. We've got a spot and vent for a fan low in the front and a power supply that exits the rear. For all intents and purposes, the airflow goes from the hole in the front of the case to the vents on the powersupply inside the case. I can think of 3 items that need attention in a majority of overclocked computers in theis scenario; the vid card, the hard drive, the clockgen and possibly the northbridge. Zip-tieing the power cables, rounding of the ribbon cables and keeping everything as tighly bundled and as neat as possible goes a *long* way towards keeping the inside of your case happy.
The easy two to fix are the chips on the motherboard--the northbridge already has a 'greenie' on it with most likely a thermal pad underneath. Replace the pad with 'real' thermal grease and that alone helps out alot. The clockgen in a little tougher. My favorite fix is to JBWeld and decent heatsink on it. Why JBWeld? It's got alot of steel in it and frankly conducts heat better than most thermal epoxies which usually have a lower mineral content. It's also really strong. If the fins are tall, it's that much easier to get cooling as both chips are in the direct path of the normal cooling circuit.
A little tougher is the hard drive. My recommendation would be to place a small fan in front of the drive bays. All you really need is a 40-60mm fan and *maybe* 20cfm of air. It won't take much at all to keep even a 10k rpm drive happy. More *is* better, but the more air a fan moves, the noisier it is.
Even tougher still is video card cooling, either to overclock or not. Until very recently, the only concession made to cooling the GPU was a passable heatsink. For some reason, the heatsink is always on the bottom of the card in a standard tower case totally defeating any natural convection that would happen. The most efficient fix I've found for this is to place a small blower like a NidecGamma28 near or on the heatsink blowing air AWAY from the heatsink. But wouldn't blowing air direclty on it make more of a difference? Possibly, but think of the direction the airflow goes. By pulling hot air away from the heatsink and directing it to the main airflow of the case, you are helping to get it out of the case as fast as possible. This goes for any other hotspot in the case that's outside the airflow. Will it help you overclock your vid card? Not at all, but the rest of the case will be better for it.
You could also go the overkill method and use a really huge fan, like a NidecBetaV that's 6" and moves 300cfm of air. Make a blowhole for something like this and simply poke vents where the hotspots are--you've got 300cfm of negative pressure to play with.
Ok, the more I write and think about this, the more I realize that there really is no hard and fast set of rules for this. Take the whole case cooling thing as merely suggestions. Most of this should be common sense. I've made cases that had a fan blowing on every heatsink that I could find. I've also made cases that had a single fan that stayed pretty cool. I personally prefer to use the least amount of fans as possible for noise reasons. Another thing to keep in mind is the bigger the fan, the quieter the fan. The above NidecBetaV can move 300cfm, but what if I put a rheostat in it's circuit and slowed it down to say 150cfm? It would be almost silent. That is good.
It's the old addage "How fast do you wanna go....". Time, effort, research and a little help can get you a PC for well under $2000 that will perform like a retail PC costing over $4000. It never seems to stop there though. Every overclocker I know has their current setup pushed to the max just waiting for the next upgrade part to come. It's almost like tattoos, not *if* you are going to overclock again, but what and when...
 |
|
|
|
|
|
|
Epinions.com ID:
psykosis_fc
|
|
Member: Jeff
Location: Pasadena, CA
Reviews written: 42
Trusted by: 22 members
|
|
|
© 1999-2009 Shopping.com, Inc.