How to build your own super-computer

by ChessBase
6/22/2003 – Just in case you decide to do this, here are instructions from a well-known grandmaster. John Nunn says that it is time to get a new computer when you get twice the performance of your current system. But it may not be advisable to simply order an off-the-shelf product. Instead he shows you how he went about building his own state-of-the-art machine.

ChessBase 17 - Mega package - Edition 2024 ChessBase 17 - Mega package - Edition 2024

It is the program of choice for anyone who loves the game and wants to know more about it. Start your personal success story with ChessBase and enjoy the game even more.

More...

My new computer

John Nunn


John Nunn, Director of the chess publication company Gambit Publications

I like to keep my computer hardware up to date, and my basic rule of thumb is that when it is possible to have a system which is twice as fast as my current system, then it is time for an upgrade. Normally, I do this by replacing the motherboard and processor within an existing computer, which is a relatively cheap method of improving performance. It also has the advantage that the data on the computer does not need to be transferred to a new system.

However, this time I decided to build a new computer completely from scratch. Why not just buy a computer, you might ask? There are two reasons why I preferred the do-it-yourself route.

The first is that the system I intended to build would be a bit special, and not an off-the-shelf product. Of course, if money is not taken into consideration then you can buy virtually anything ready-made, but a quick check showed that my intended system would cost too much if bought this way. The second reason is that I have invariably been disappointed with complete systems I have bought. My worst experience was relatively recent. I bought a complete system from a well-known UK company, and to begin with everything seemed OK. Indeed, if all you did was run Microsoft Office then you would probably never realise that there was any problem with the machine. But if you set Fritz to analyse a position for, say, 12 hours, then there was about a 50-50 chance that it would crash during this period. Obviously this is unacceptable. It would also sometimes crash when running leading-edge video games – perhaps about once every two hours. The highly intermittent nature of the problem made it hard to prove that there was anything wrong with the system, and in fact I never did manage to get the company involved (which had appalling customer support) to accept that the machine was faulty. My suspicion is that such companies, which often operate on very low profit margins, scour the world for the cheapest possible components in order to save a few pounds. The fact that these components might be slightly sub-standard and become unreliable under stress is probably of little concern to them. If you are building your own machine, you can of course choose to pay a small premium in order to obtain higher quality components.

In my opinion, reliability is the number one criterion for a computer. All else, including to some extent performance, is secondary. This applies particularly if you are using the computer for business purposes. What level of reliability is acceptable? Here you have to distinguish between an individual application crashing and the whole computer crashing, so that a reboot is required. There is little you can do about the first of these. A poorly written application will crash whatever hardware and operating system is being used, but the key point here is that the operating system should close the dead application without any fuss, and then continue running as normal. However, the second type of crash is much more serious as a reboot will lose you any unsaved work in other applications which are running, and may cause corruption to files on the hard disc.

I would say that for a computer running eight hours a day, one system crash a month is probably the limit of acceptability and there is no reason why this should not be achieved (both my current main systems have at least this level of reliability). It's probably not possible to achieve such a level using Windows 98, because of the tendency for one crashing application to bring the whole system down; this implies that Windows 2000 and Windows XP are the only viable versions of Windows. Personally, I prefer the former but I accept that this is to some extent a matter of taste.

After reviewing the various possibilities, I decided on the following basic components:

  • Supermicro X5DAL-G dual Xeon motherboard
  • Two 2.8 GHz Xeon processors
  • 200 GB hard disc
  • 1 GB ECC Registered RAM
  • Supermicro SC762-420 case

Of course, a few other miscellaneous components are necessary, but I could save on most of these by recycling them from earlier machines. As this computer was to be focussed on chess, video performance was not important.

The above picture shows all the components ready for assembly. I would like to thank my brother David for his assistance in assembling the computer (in truth, he did almost all the work...). The tricky bit is putting the processors on the motherboard:

This picture shows the motherboard ready for the processors. The sockets for the Xeons are on the top right; the memory slots are to bottom right. The heatsinks for the Xeons are so massive that they have to be attached to reinforcing plates which are mounted under the motherboard. Presumably the motherboard would simply bend if this were not done.

Both Xeons are in their sockets and the heatsink has been attached to the one on the right. Before attaching the heatsink you should squirt some thermal compound on top of the Xeon. Intel thoughtfully provides a hypodermic with the Xeon for precisely this purpose.

Here both heatsinks have been attached and the cooling fans mounted on top.

The next step is to mount the motherboard in the case. For a dual Xeon system, it is a good idea to use a case specifically designed for the purpose, as the power and cooling demands of the Xeons are considerable.

The motherboard has now been mounted in the case. You can see the 1 GB of RAM on the right side of the motherboard.

There isn't much more to say as all the other components are standard. The above photograph shows the inside of the case with everything installed and ready to switch on.

With home-built computers, there is usually some small problem to be resolved at switch-on time, but on this occasion everything worked immediately and it was only necessary to put the operating system CD-ROM in the drive and wait for it to install itself.

Of course, you're probably all waiting to know how the chess is. One of the problems with currently available processors is that they are not particularly well suited to the integer calculations used for chess. A Pentium 4 will be slower at chess than a Pentium 3 of an equivalent clock speed. The upshot is that although I was switching from a dual-processor 1.2 GHz Pentium 3 machine to a dual-processor 2.8 GHz Xeon machine, the new computer wasn't 2.2 times as fast, which is what you would expect from the clock speeds. Instead, it turned out to be about 1.7 times as fast at chess. FritzMark, which only uses one processor, gives a value of 1276 on the new machine. Since the second processor generally adds a factor of 1.7 to the speed, you can say that the estimated FritzMark of the machine is about 2170.

A typical middlegame position with Deep Fritz 7 gave about 2.15 million nodes/sec (Mn/S) while Deep Junior 7 gave about 2.55 Mn/S. Here is some analysis:

It didn’t take long to find a missed win by Adams against Spraggett at the Elista Olympiad in 1998. The key point here is finding the move 27 h3 which is the move that makes the combination work. Once the machine displays this move, you can say that it has ‘found’ the combination. As you can see from the screenshot, it took Deep Fritz 7 running at 2.608 Mn/S just 2 seconds to find the win. It makes a big difference to the time how many lines of analysis you show; here it is two.

Here is a famous combination (Zuckertort-Blackburn, London 1883). The position is normally given after 26 gxh7+ Kh8 27 d5+ e5, but the computer found 28 Qb4!! so quickly that I couldn’t accurately measure the time. Therefore I set it back two moves and started it running. The evaluation suddenly jumps from about equal to +5 when it sees that 28 Qb4!! is winning. Time taken: 6 seconds (you can’t see the time in this screenshot, but it really was six seconds!) at 2.313 Mn/S.

All in all, the new computer seems quite satisfactory, although changes in processor design mean that chess comes off second best relative to other types of application, so perhaps the performance isn't quite what I might have hoped for.

I should add a couple of words of warning. Firstly, if you like quiet computers, this sort of machine is not for you. The massive fans needed to take the heat out of the case are quite noisy. Secondly, you shouldn't attempt to build you own computer unless you are confident of your ability to do so. Obviously, if you do it, then it is entirely at your own risk.


Reports about chess: tournaments, championships, portraits, interviews, World Championships, product launches and more.

Discuss

Rules for reader comments

 
 

Not registered yet? Register