The new computer that I’ve been talking about lately was built for one reason: overclocking. Why else would I use a processor like the Pentium D 805 that, at stock speeds, is inferior to my old computer’s processor? And why else would I get water cooling?
Overclocking is the process of setting components inside your computer to perform higher than their stock (default) speeds. The main reason why overclocking works is because processor manufacturers want to be sure that their processors will be 100% stable even with the measly fan that is usually included with it. However, if you have superior cooling methods, like the water cooling in my case, you can set the clock speed of the internal components much higher than their desired speeds without any consequence in stability.
A variety of motherboard manufacturers create boards that are aimed at overclockers. DFI, ASUS, and Abit are just a few well-respected manufacturers in the field, but there are truly a lot of boards great for us overclockers. I personally am using a board from Abit called the AW8D, which is most definitely meant for overclocking. Abit even includes a special sub-menu in the BIOS that has all the overclocking options in one convenient location. Even without that, though, overclocking is a rather easy process with modern motherboards. The trick, however, is not setting your components to a faster speed, but rather balancing speed, temperature, power consumption (which is linked directly to voltage), with the desired result of stability. Most people would like 100% stability, so really the only things you then become limited by are your temperatures and power supply’s ability. And, of course, the components themselves. It is really a mix between a science (actually knowing what settings will help achieve stability in certain conditions) and an art (achieving the perfect balance with the hardware and BIOS available to you).
Not every processor or RAM module can be overclocked well. There are certain processors, especially from AMD (AMD actually overclocks their processors a bit before selling them, which makes them less overclockable by consumers), that can barely increase by 100mhz, while there are others that can be pushed to over 150% of their stock speed. Probably the two best modern chips for overclocking (that are dual core) are the Pentium D 805 and the D 930. I chose the 805 because it was $60 cheaper, but the 930 starts at a higher clock speed (3ghz versus 2.6ghz) and so can end up at a higher speed. The 805’s maximum ever achieved is around the 4.25ghz range, with normal cooling (by normal I mean either fan or water… not liquid nitrogen or anything insane like that). The 930 has been known to reach 5ghz in certain conditions.
Regardless, every single processor is manufactured differently, so even if you were to purchase the exact same models of every component that someone else used, your overclocking results would vary. Some processors, just by the nature of their production, have lower voltage requirements. This means that they are better to overclock. It seems my 805 is somewhere in the middle in terms of performance. There are those, like Tom’s Hardware Guide, that can reach over 4ghz. However, there are also people who can’t break 3.6ghz. I made it to 3.8ghz in the end, which I think is definitely respectable (and a definite improvement boost over my previous system).
Anyway, what I would now like to do is go through the steps that I took in overclocking, step by step. Any true overclocker pays attention to one key virtue: patience. As intriguing and tempting as it is to just clock your 805 to 4ghz and hope it boots, it can seriously damage your processor. However, if you take small steps (like 100mhz at a time), then there is almost zero chance you could ever do harm to your system. The worse thing that can possible happen is that your computer fails POST, in which case you reset CMOS and try either upping the voltage, relaxing the memory timings, or admitting your defeat.
Before I begin, however, I wanted to cover the software utilities that I used to determine my computer was stable. First of all, I used the wonderfully feature-filled program called CPU-Z in order to see the actual clock frequencies of my components. If you set the FSB to 180mhz in the BIOS, it will likely only be 179mhz when you boot up. It’s just the way it is for some reason. CPU-Z will show you exactly what the current bus speeds are, despite what the BIOS may want you to think. Also, Windows XP updated the System Properties dialog box to show the amount of RAM, as well as processor speed. Previous versions of Windows just showed the description hard-coded into the processor itself. (Most of the time this says GenuineIntel… that doesn’t tell you much about the processor, now does it?) Also note that CPU-Z can be useful if you want to upgrade your RAM. It tells you the speed of your RAM, which is what you need to know when purchasing an upgrade.
Aside from CPU-Z, the single most useful application is something called Prime95. At first, it seems to be a program only for mathematicians. It does something regarding the Mersenne primes. Don’t ask me what that means, because I have no idea. The general idea is that whatever this prime number calculation is, it taxes your computer to insane proportions. The Prime95 application comes with a Torture Test option that will do the calculations over and over and over again, until you tell it to stop. This is used to recognize stability in computer systems. It is recommended that you run Prime95 between 6 and 24 hours straight to get an accurate idea of your system’s stability. I didn’t feel that was necessary, so I used 4 hours as my magic number. However, I will say that when it did fail, it was always within 1.5 hours (the longest one I had go that failed was 1 hours and 13 minutes). I just did the four hours to be safe. Do realize that if it fails within two minutes, you are going to have major system instability. If it fails after an hour, you may be able to safely ignore it, as Prime95 taxes your computer in ways that even modern games don’t. But I refuse to run my computer if it is less than 100% stable, and I hope you do the same. Anyway, because I have a dual-core processor, I ran two instances of the Prime95 program at once (this is referred to as Dual Prime in the overclocking world). I should probably also note that while running Dual Prime, I was also downloading with Azureus, surfing the net with Firefox, and using GAIM for IMing capabilities. I figured that if my CPU withstood that, it was stable. And I would like to point out that the system was quite responsive at this time, despite Dual Prime doing 100% CPU usage.
Also before I begin (you’re probably thinking I’m never going to say anything, aren’t you?), I wanted to make sure you know where I am getting my temperature units. I am giving two temperatures per overclocking step: idle and load. Idle is the temperature of your computer when it isn’t doing anything. I got this value by booting my computer up and letting it sit for 30 seconds after I was fully logged in and booted. I then took whatever value I saw pop up the most (the temperature changes ever second, usually within a 5 degree range). Load is the temperature your computer reaches when at 100% load. The way I determined this value was to run Dual Prime for 4 hours. During those four hours, I would occasionally (once every twenty minutes), look at the temperatures (which were taken with the utility ABIT includes with their motherboards, by the way). The load temperature I recorded was the HIGHEST value that I saw. Most of the time, the temperature was 2-3 degrees lower than what I said is the load, but I figured I would come out and say the highest value my CPU reaches.
OK, so let’s begin, shall we? My CPU has a stock speed of 2660mhz (we will be dealing in megahertz from now on). It has a multiplier of 20, which makes the stock front-side-bus (FSB) speed 133mhz. If you take 133 * 20, it would equal the 2660mhz of the processor. Thus, to find a processor’s speed, you take the multiplier (this cannot be changed on most Intel processors) and multiply it by the FSB. Also, to find the memory bus frequency (this will be hereby called memory speed, but it is not to be confused with the actual memory speed, which is only two times the FSB… this is what the speed is actually at, but the bus speed is the thing that is often referred to when you buy RAM), you multiple the FSB by 4. So, at 133, my memory speed would be 533mhz, which is good considering I bought memory rated at DDR2-533 (simply put, DDR2 memory meant to run at 533mhz). Simple enough, right?
However, when I first booted up my computer it wasn’t at 2660 for some reason, but rather at 2720, slightly higher than it is supposed to be. I’m not sure what caused this, but the increase is so little that no major increase in speed over the stock would be recognizable. I ran Dual Prime for four hours on this original setup just to make sure my computer was stable before overclocking (and what exactly would I have done if it failed the test? LOL). The temperatures were as follows: Idle 22, Load 35. (all temperatures are in Celsius, by the way. Keep in mind throughout this whole thing that 60 C is considered the safe temperature for Intel processors, while 45 C and below is considered very good.) Now it is time to actually begin overclocking. Nothing really interesting happens for awhile, so I’ll just give you the speed it was clocked at and the temperatures. My first step was 2880 mhz, which yielded 26 idle and 39 load. 3100 mhz clocked in at 25 idle and 40 load. Finally, 3350 mhz was 25 idle, 42 load.
I am going to take this time to introduce some other settings that I have no yet had to change but will shortly. Almost every component you connect to the motherboard has a certain voltage pumped out to it from the power supply, by way of the motherboard. The two main voltages that matter during overclocking are the VCore (voltage of the CPU), and the DDRV (voltage of the RAM). Another important setting are the memory timings. There are four main memory timing settings you can change, and what they mean is way beyond the scope of this document. Just remember that these settings are measuring latency (delays), and so lower values are better. Just know that the lower your memory timings, the overall speed increase of your computer. However, in some cases it is better to relax (increase) the memory timings and get higher speeds. The VCore for my CPU was set to 1.350 V by default, while the DDRV was at 1.80 V. The memory timings were 4-4-4-12 (this is how memory timings are expressed, each number between the hyphens a separate setting).
So, I upped the CPU clock speed to 3520 mhz. To my surprise, the computer would not pass POST. I had to reset CMOS to get back into the BIOS. I figured this would be where our good friend Mr. VCore would come in. I then upped the VCore to 1.400 (from 1.350). The computer now passed POST, but when it started to boot into Windows I got an error about a missing System32 file or something. I knew my Windows install wasn’t corrupt, so I figured it was a problem with my BIOS settings. I then realized that since I was at 3520 mhz for the CPU, that meant my memory was running at 705 mhz. Considering it is only rated to run at 533 mhz, that is a substantial increase. I then visited the memory timings, and relaxed them to 5-5-5-15, which I read in a review about my memory is a good place for them to be during overclocking. With the adjusted memory timings and CPU voltage, the computer could finally boot into Windows XP, and seemed pretty stable. I went ahead and started up my Dual Prime, started an Azureus download, and started to watch some anime in VLC Player. I kept the windows positions so I could see the temperature at all times. Also, you can monitor Prime95’s progress just by lookign at the system tray. The icon will be red if still running, or yellow if idle (if yellow, when you were previously running a test, it means it errored out). I went through about three episodes (50 minutes total) of the show, with the two Prime95 icons still red. Then, somewhere in the middle of the fourth episode I saw that one had turned yellow. The test had failed, meaning my computer was instable! At this clock speed, even though it was instable, I still recorded the temperatures. They were 28 idle, 42 load.
Now, the thing you need to realize about Prime95 is the errors will mean nothing to you. For example, one error I got was “Rounding error. Expected less than 0.4, got 0.49728193.” Thanks for the info, Prime95! But in reality, it actually is very helpful. You see, when I do my Dual Prime test, the first instance is a Large block test, and the second is a small block test. All math aside, the large block test is primarily to generate heat and RAM usage, while the small block test (which can fit inside the L2 cache and doesn’t use RAM), tests the CPU primarily. That being said, I had failed the small block test, so that told me I had a CPU-related problem. This is where Prime95 really shines, because now I knew to up the VCore but not worry about the memory (the timings I set would last me all the way to the end). Still, though, I chose to up the DDRV as well, just to be safe.
So, the next step was for me to up those voltages. I brought the VCore to 1.425 and the DDRV to 1.90 (my RAM is actually meant to run at 1.9, so I figured it couldn’t hurt at this point). I also upped my clock speed to 3580 mhz (up by 50mhz), just because I figured it would be able to work with the new voltages. This is where the art part of overclocking comes in, because the science aspect would tell you that there is no logical reason to up clock speed when trying to fix instability. But, I ran Dual Prime for 5 hours (wanted to be sure this time), with no problem whatsoever. I was officially stable again. This was the first real hurdle during the overclock, and it had been overcome. Temperatures were at 28 idle, 44 load.
However, as stable as the system was, my system started beeping after about one hour of Dual Prime. I checked in the ABIT utility, because I knew the beep was being caused by one of the Abit EQ settings. Sure enough, there was a Beep EQ set for temperature monitoring (it would beep if something overheats, a nice feature). I turned off the beeping so it wouldn’t annoy me, but I had to address the problem. The culrpit was a little sensor called PWM1. That meant absolutely nothing to me, until I went over to the Abit forum (a very helpful place, by the way. MUCH better than ASUS’s site). It seems the PWM overheating is a common problem with motherboards, the Abit one’s especially (because the way they are designed), if you don’t use standard CPU cooling (air). The PWMs on the motherboard are some type of voltage regulators. As it was explained to me, they make sure the components get the right voltage (that you set in the BIOS), so as to not overvolt (i.e. fry) the components. Regardless, when you overclock, your voltages are going to be higher, so these PWMs need to work extra hard. The end result is that they overheat. The way the motherboard is designed, the PWMs are placed right above where the CPU socket is. Therefore, if you use a standard CPU fan, the air it blows onto the CPU will also be blown onto these PWMs. In short, a CPU fan will also cool the PWMs.
However, if you use heat pipe (like most members on the Abit forum) or water cooling like me, the CPU fan doesn’t exist and so no air is blown over the PWM region. This means they will overheat a lot quicker than usual. So, I was forced to mount a fan with L-brackets purchased from Home Depot above the PWM region. You can see a picture of that here. The sky-blue metal things are heatsinks for the PWMs themselves, and so that is where I was aiming (no pun intended) to have my air directed towards. The L-bracket mounted fan, while it increased the noise level of my computer substantially (the case fans in my system are all ultra-quiet, and the water system is also fairly quiet. This fan is the main noise maker in my case, unfortunately) capped the PWM1 temperature to 55 C, while it was previously reached 83 C (80 C is considered the limit). I think that is a substantial benefit: it is an eighty-two degree difference in Fahrenheit!
After the PWM1 overheating was dealt with, I continued the overclocking process. The next step was 3680 mhz. Rather than test it first, I decided to just up the voltages right away. I knew this would require some substantial boosts, so I went ahead and brought the VCore to 1.475 and the DDRV to 1.95. I was now at 3.7ghz, which I considered to be the home stretch. Only 300mhz away from the desired 4ghz. That is why I didn’t hesitate to raise the voltages more than I normally would have. I might as well get them near what I wanted them to be in the end, since it is indeed coming to the end. This clock speed was Dual Prime stable for four hours, so it looks like my voltage increases paid off. Unfortunately, the CPU temperatures took a nice jump, now at 27 idle, 50 load. Pass the 45 C mark, my CPU is no longer in the “very good” range for temperature, but it is still quite far from the 60 C safe range. I just wanted to quickly note that my AMD Athlon XP 1600+, running at stock speeds with the included CPU fan, ran at 48 C idle, 65 C load, with absolutely no problem. A lot of people get really into following that 60 C mark, but you can realistically go up to 75 C without risking damage to the CPU (though you will have to change the thermal paste often), and one guy on the Extreme Overclocking forums has his system running at 102 C load, fully stable! But if you are really willing to go to 102 C , then you would never want water cooling. Most people get water so they never have to see a temperature above 60 C again, and I tend to agree.
The next step was 3780 mhz. I had a couple hiccups in Dual Prime using the voltages from before, so I used a VCore of 1.550 and a DDRV of 2.05. Both voltage values were substantially increased in order to achieve complete stability. I am actually sitting directly on the “safe” recommendation for VCore, which for my processor is exactly 1.550 like what I am now at. The DDRV could probably go up to 2.1 without any problems, so I’ve got a little bit of room with that. However, I doubt it will be necessary because in the review that mentioned the 5-5-5-15 memory timings, it also had my exact memory running at DDR2-884 at only 2.00 V, which is much better than I even thought possible (if my memory reached 884, then it would mean a CPU clock of 4.4 ghz… dream on). This seemed to be about the voltage limit, then, so I was hoping for stability. This was Dual Prime stable, but the temperatures showed the voltage increases. 28 Idle, 56 Load.
My next step was to try 3880 mhz. At first I kept the voltages where they were, but Dual Prime failed extremely fast. Previously, the failures happened after a good 45 minutes of testing, if not more. However, this time it literally failed within 5 – 15 seconds. That was definitely a bad sign. I upped my voltages as far as I was willing to go, with a VCore of 1.600 and a DDRV of 2.10. Still no luck, as the Dual Prime was failing after about 15 minutes now. So, I bought a little bit of time, but it wasn’t stable by any stretch of the imagination. I figured testing a higher CPU voltage couldn’t kill anything, so I tried 1.675 (the max my motherboard allows, and the voltage that Tom’s Hardware Guide used to get 4.1 ghz). However, my motherboard refused to boot at that voltage. I also tried boosting the DDRV to 2.15, but it wouldn’t boot there either.
So, it seems that I have crossed beyond the limit of my hardware. 3780 mhz, or 3.78 ghz, is as far as my system is going to go. The temperatures are still way lower than they could be (I was willing to hit 65 C), so the water cooling is doing its job. I can’t imagine how hot the CPU would be with air cooling… it would probably catch fire! Needless to say, I am disappointed that I didn’t make 4ghz. In all reality, the extra 220mhz probably won’t even make a two frames per second difference in a game, but the 4ghz mark is just such a braggable place 🙂 But hey, this is my first time doing a serious overclock, and I think 1.11 ghz above the stock speed is some great stuff.
Maybe in a year or so, if I really see the need to get a faster CPU (somehow I fail to see a 3.8ghz CPU as a bottleneck, but who knows), I’ll spring for the D 930. I should be able to overclock that to around 4.7 ghz with my cooling setup, a full 1 ghz above what I’m at now. And with the Conroe chips from Intel being released the end of July, the 930 price is set to decrease substantially.
If anyone out there is in the market for a PC, the D 805 is a great option for cheap. With some good air cooling (like the $55 Zalman monster), you could probably make it to 3.4ghz without trouble, but remember that high-end air cooling does not take silence into consideration. I heard the Zalman CNPS9500 is akin to a jet engine :p In reality, though, anyone truly looking for a high-performing PC should just wait for the release of Conroe. Early benchmarks of the stock processors will destroy an overclocked 930 any day. And given the lower voltage and temperatures of the Conroe at stock speeds, it can be overclocked to insane proportions. One fan overclocked an early Conroe version to 4.8ghz, and it’s performance is off the charts. It absolutely blows away Quad-Core computers and the $1000+ offerings from both the Intel EE line and the AMD FX line. It is definitely something to look out for in the coming weeks, and very possibly the thing that secures Intel’s position as market share leader for microprocessors for many years to come.
I will post Excel-generated graphs of benchmarks comparing my old computer, and then the new computer at stock and overclocked speeds. Expect them in a week or so.