How to Build Your Own PC - the Smart Way (part 2)

In the last article we looked in depth at the heart of the PC build:

• The CPU (processor)
• The Memory (RAM)
• The Motherboard (main board)

Now we will look at the remainder of the PC:

• The Storage subsystem (hard disk or HDD)
• The Graphics Processor (GPU)
• The Case
• The Cooling (HSF or heatsink & fan)

When you save your work, run an application or game, stream or encode to disk or in any other way require to access to or from the permanent storage your disk subsystem offers you are constrained by its performance.  The storage system is almost the only mechanical system within the computer (although this is changing now with the advent of Solid State Drives, SSD's).  Data is stored on a hard drive magnetically and arranged in concentric circles around the disk from the centre to the edge.  Therefore if your computer needs access to data at one end of the disk and then the other the drive has to physically move the head backwards and forwards, and wait for the disk to spin around to the right location to begin the read or write process.  The time taken to do this is known as latency, and the average time to move the head to any given point and read data is given as Average Seek Time (usually in milliseconds, ms).

Manufacturers provide spin speed and average seek time data with their drives.  In addition all drives have an onboard memory cache that the drive software uses to intelligently store previous (or read ahead) read requests so that it can almost intelligently pre-empt fetching of the data you want read.  Data can then be quickly retrieved without requiring the disk to move the head.  The other significant factor in drive performance is having as big a possible pipe (i.e. bandwidth) between the drive and the CPU to make sure that as soon as the drive has data available it can be transferred.

Our design imperatives for good performance storage are, in roughly this order:

• high bandwidth - achieved with;
• high areal density (more data packed into a smaller space can be read faster, target density around 300 gigabits per square inch)
• a high drive speed of at least 7500rpm or better still 10000 or more,
• high data transfer speeds of 3Gb/s,
• and supporting the latest interface standards of SATA-II or SAS (SCSI)
• a large onboard cache of at least 16MB and better 32MB or more
• best possible Average Seek time (should be under 10ms, under 5ms for extreme random access performance)
• RAID (Redundant Array of Inexpensive Disks) - using arrays of disks working in parallel to achieve higher speeds, higher transfer rates and redundancy for resilience in the unlikely event there is a drive failure

Most manufacturers of mainstream PC's will only quote you disk capacity in GB and this is used as a selling point, they are often unable to give you the data above and are unlikely to have considered it in the design.  There is a balance to be struck between performance and value, but also a delicate balance between bandwidth and seek times.  If the work you do is predominantly sequentially accessing the disk it is probably more important to have high bandwidth than fast seek times (i.e. photography, multimedia).  You also consider this in the design of RAID storage arrays as they impact on average seek times, write speeds and have their own critical configuration parameters that affect how effectively they will function.

The rest of the computer is solid state it has no moving parts and can therefore operate at high speed with no considerations for the physical limitations of movement or wear and tear.  With memory prices getting ever cheaper and cheaper the logical next step is to remove the last mechanical device from the heart of the computer system and replace it with solid state non-volatile memory or flash memory (NV-RAM).  This is where the current leading edge of drive development is and solid state drives are now available in smaller sizes (up to 250GB on a single drive) but blazingly fast and very resilient.  SSD's have some considerable benefits and a few technical obstacles that make it still imperfect as a hard drive replacement worth consideration:

Pros

• Random access is blazingly fast - as there is no mechanical head to move to drive to spin
• Resilient - very long life without moving parts to wear and deteriorate, or suffer from stress or shock damage (an SSD can typically tolerate a shock load of over 1000G!)
• Lightweight - without motors or magnets, cylinders and disks, or a stiff chassis the SSD is very lightweight
• Lower power consumption - mechanically moving parts use generally more power
• Future roadmap - ultimately mechanical drives will probably die out and be replaced by the SSD technology, currently it is at its very early stages

Cons

• Its still very expensive - typically an SSD is many times more expensive than the equivalent mechanical hard disk
• Write speed is generally lower than a mechanical hard drive - due to the way in which data is retrieved from NV-RAM its transfer speeds currently lag behind mechanical drives.  Random writes on a fragmented drive can be a lot slower.
• Capacity is currently limited - the biggest drives are only around 250GB
• The technology is still maturing with different designs for cell access (MLC/SLC) and organisation being worked through the market trading off performance for size
• Marginal improvement over the best mechanical RAID arrays - for the same investment you could buy and configure a SAS RAID5 array of 15K rpm drives that would outperform current SSD technology

Graphics processing is now so demanding that it needs at least one dedicated processor of its own, a GPU (usually resident on an onboard card). The GPU is a real workhorse when 3D graphics modelling or games are being processed. If the CPU had to perform all 3D graphics work it would quickly get bogged down as it is not optimised for it. However the CPU has an important roll in determining the workload that gets fed to the GPU(s) and if they are not balanced performance will suffer to one or other not being up to the job.  The CPU must be powerful enough to keep the GPU’s busy with instructions and the GPU must be capable of processing it fast enough to keep instructions moving and not queuing up and truncating frames (i.e. low fps).  The main manufacturers of graphics processors are ATI (AMD) and nVidia, both are excellent graphics processors and at any time one or other leads the maximum horsepower race in terms of graphics processing capability.

Modern graphics processors are full length cards, requiring their own power supply (sometimes two) and their own dedicated cooling. They use PCI express slots for maximum speed and can be installed in multiples to work in parallel together usually in pairs (known as either nVidia SLI or ATI Crossfire configuration).  It is now also possible to run triple or quad graphics processing systems (the latter using 2x dual processor cards).

Our own benchmarks and tests have shown that by-and-large the best value is obtained with single cards as there are still some applications and games that are unable to make effective use of multiple GPU’s. The benefits of multiple cards are non-linear against the significant linear cost increase. Of course multiple cards remain an available upgrade route in the future. However if you want the ultimate in speed and performance then Quad GPU's is certainly the way to achieve it. My advice would be build your system around the best single card you can afford and then you will have an upgrade path available for multiple cards should you wish to go this route in the future.  Like CPU's the graphics processor works best if it’s kept cool. So if you want the ultimate out of it go for enhanced air or water cooling that will further enhance their performance significantly.

The standard heatsinks provided with graphics cards, motherboards and processors are usually of relatively low quality alloy and provide inadequate cooling for over performance.  Invariably barely cooling for even standard performance, remember these things are made to an aggressive cost limit.  The heatsink is important as if it doesn't move enough heat away from the component quickly enough the heat builds up and up, until either shutdown or meltdown!

At standard factory settings there’s a lot of margin built in to allow for quality variability so you shouldn’t have a problem, but if you want to push your system hard then you are going to need to think more about improving cooling.  To move heat away quickly you need a good contact with the processor with high pressure and specialist thermal compounds that conduct heat very well and very fast.  To move enough heat away so that it doesn't build up you need a large enough heatsink with enough cooling surface area (usually assisted with a fan).

The qualities to look for in high quality heatsinks are:

• a high quality pure metal, copper or aluminium construction with good heat conductive properties
• a perfectly machined flat surface for good contact with the component
• a very large surface area over which heat can be exchanged with the air (or water in the case of a water block)
• a rapid transfer of heat from the component to the cooling surface area fins usually through the use of specialised heat conduction pipes (or in the case of water a high flow, high head water system and half inch diameter tubing)
• specialised pure silver based thermal compounds that enhance the thermal interface between the component and the heatsink
• specially designed fans with fins that are aerodynamically designed to move a lot of air at low rpm, for the most cooling with the least noise
• remember a very good air cooling system is so good now that they can outperform many of the poorer water cooling systems.  Don’t always assume water is better…

A good case is often seen as a nice to have but the case is another vitally important component in the holistic design of a high performance system.  You should look to ensure:

• there is appropriate circulation of air through the system (usually front to back, bottom to top) and that the volume of air flow is adequate.  Well designed cases now zone thermal regions so that hot air from one component does not ‘pollute’ another with already heated air

• there must be the right number of fans in the right places moving the correct volume of air.  Often 'flashy' cases have powerful fans lined up with LED's lit up making them look impressive but actually they are moving very little air, and not over the hot spots in the PC like the graphics card and CPU

• did you know that as air heats up inside the PC case its volume increase by as much as 3:1, in other words, for optimal cooling you need to shift three times as much volume of air out of the case as into it, few designs take account of that.  Ducting into racing engines and exhausts is designed specifically around this principle hence why you see a small hole in the front to let air in and a much larger hole at the back

• A case made out of solid aluminium not pressed steel, and anodised rather than painted will be far more durable to upgrades, maintenance and component changes and inspections.  Solid well built cases make less noise, how many times have you had a PC develop an annoying rattle...

• A well designed case should allow easy access to all components so they can be installed and removed without placing stress or risking damage to the system, or the engineer!

• Few cases are designed to easily accommodate large heatsinks, full length graphics cards, large power supplies, disk arrays, radiators, pumps, reservoirs and the plumbing of water cooling systems, ensure yours is fit for purpose

• Last but not least a case needs to be pleasing to your eye.  There are a lot of garish looking cases out there I wouldn’t want in my house, and some nice unobtrusive classic designs that I would, but then that’s your personal choice…