01 Aug 2017

How to build your next workstation from scratch

  • Written by 
  • submit to reddit  
Build your own machine and you end up with exactly the specification that's required Build your own machine and you end up with exactly the specification that's required Shutterstock.com

Index

In a couple of recent articles, we've discussed ways to get the most computing horsepower for the money, comparing the specifications of various options and considering the expansion and upgrade options. To take that to its natural conclusion, it's not that difficult to build a workstation completely from scratch. Really.

Building your own workstation is not only generally the most price-competitive option, but it also lets you make the right choices about things which nobody usually thinks about. For instance, probably the most common failure in off-the-shelf PCs is the power supply, which can easily destroy every other component in the machine, and all the data, if it fails. The power supply, however, is rarely mentioned on the shelf-edge sticker at a showroom.

The absolute best thing about a custom build, though, is the configurability; the ability to build a machine to exactly the specification that's required. Doing that requires we understand a bit about the specifications of each component, but we also need to know how to find parts that'll fit together. This is less scary than it seems, since generally if things fit in a slot they'll work.

CPU

The CPU (shown upside down, revealing the pins) must fit the socket on the motherboard.JPG

The CPU (shown upside down, revealing the pins) must fit the socket on the motherboard

CPU specification is the go-to measure of a workstation's performance. Even in modern workflows, where a lot of the heaviest lifting is offloaded to a graphics card, CPUs are often kept quite busy decoding video and running the rest of the system. Nobody ever complained that the CPU was too good.

Principal measures of performance include obvious things like clock speed and the number of cores, but figuring out what each core can do per clock cycle requires a bit more information. The i-series numbering on Intel's CPUs serves to collect them into broad capability groups. Things beginning i7, for instance, typically have hyper-threading, which is how Intel describe the ability to handle more than one stream of instructions per core to ensure that the core is kept busy even when it's waiting for information from other parts of the system. CPUs also have a couple of levels of cache memory, which is used as a sort of quick reference area to store data that's likely to be quickly needed again. Video workflows, which pump lots of picture data through the CPU, can often make good use of this.

Intel's top-of-the-range Xeon line tends to include all of the available bells and whistles, and can often work together in groups for more total cores, though they may be clocked slightly slower to keep overall power consumption down. If in doubt, you can compare individual CPUs on a website such as CPUBoss, to figure out the best compromise.

CPU compatibility is expressed in terms of the type of socket on the motherboard; current Core i7 7000-series chips use Socket 1151, for instance (the number refers to the pin count). In rare circumstances, particularly old motherboards may not be able to run particularly recent CPUs of the same socket, but Google will generally answer these questions.

Motherboard

Motherboards break down broadly into smaller types for domestic computers and larger ones for big workstations. Both are useful.JPG

Motherboards break down broadly into smaller types for domestic computers and larger ones for big workstations. Both are useful

The motherboard, as the name suggests, is the basis of the system into which more or less everything else plugs. It needs to have a socket compatible with the desired CPU, and effectively all of them have additional features such as hard disk controllers.

Motherboards intended for consumer PCs tend to have built-in sound which is probably good enough for basic stereo monitoring, USB and other ports, and a controller for SATA disks. The RAID controllers are often very basic and should really only be used to control single disks, for RAID0, 1 or 10; they may be able to create RAID5 arrays, but the performance may be poor. Motherboards intended for Xeon CPUs may have much better RAID controllers, though they may require expensive SAS disks. They will typically have at least enough SATA ports to support a conventional startup disk (of which more under storage) but may entirely lack things like sound and USB-3, which you can add as a plug-in PCI Express (PCIe) card. Check that the motherboard supports a revision of PCIe that will allow any expansion cards (a disk controller, perhaps, or SDI board) to work at full potential.

Some motherboards have lots and lots of things built in, up to and including graphics capability (which may itself be part of some CPUs.) This is often absolutely fine for office work, and performance of onboard graphics has improved a lot recently, though it might not be suitable for particularly advanced video editing.

Crucially, the motherboard also has room to take...

RAM

Memory comes in modules. Often, they're used in pairs or triples.JPG

Memory comes in modules. Often, they're used in pairs or triples 

Memory has long been the second item listed in a computer's specification, and it's fairly straightforward: get as much as the motherboard and operating system will support, in a type the motherboard supports. It isn't outrageously expensive and more is more. RAM comes in speed grades, but beyond a certain point, faster RAM tends not to produce easily-detectable performance gains, so don't pay a huge premium for the very fastest stuff.

At the time of writing, there were a few current types: various revisions of DDR RAM, used in consumer machines, and the “buffered,” or ECC, type used in more industrial installations with Xeon processors. It's usually simple to discover and purchase the right stuff. Sometimes ECC and non-ECC RAM will fit the same slots, which is a rare example of things fitting but not working; usually, things that don't work won't slot in (and nothing will be damaged even in this case.)

Storage

These are SATA ports, of which modern motherboards tend to have at least six.JPG

These are SATA ports, of which modern motherboards tend to have at least six

Storage for video workstations breaks down into two areas. First, there's the boot disk, so called because it contains the software (windows, linux) which the computer uses to start up, and the applications. This is usually a solid-state disk, which is fast but expensive, and generally too small to contain bulk video data. The counterpart is usually a big disk array made out of conventional spinning disks.

Boot disks are usually SATA or M.2 format; SATA connects up like a conventional hard disk, whereas M.2 is a plug-in module that sits on the motherboard. Disks in the array – at least two, as many as 16 or more – are simpler. They'll be conventional 3.5” hard disks, though look for types which run at 7200rpm or faster, for better performance. 5200rpm options are cheaper and run cooler but suffer reduced performance. Size is a cost issue: at the time of writing, disks of two terabyte capacity were fairly normal. Read up on disk arrays here before making any decisions, and check the revisions of storage standards supported by disks and controllers, to be sure everything is running as fast as it can.



« Prev |


Phil Rhodes

Phil Rhodes is a Cinematographer, Technologist, Writer and above all Communicator. Never afraid to speak his mind, and always worth listening to, he's a frequent contributor to RedShark.

Twitter Feed