To save some money I'm sure we would all prefer to build our own workstations. But this takes a lot more work than simply buying one off the shelf. With any luck, you'll end up with a better machine for a lower price.
What methodologies do you use to decide on which parts to buy? Do you look at benchmarks? Do you find things that have compatible bus speeds? How do you know where to look to make sure everything will be compatible? Do you buy a bare bones machine and extend it?
The first question to ask is what you're using the computer for? Will this be a gaming machine or just a workstation for mom to check her e-mail? It also depends on how much you value your time and support. If you spend 12 hours researching parts and variations on hardware, and another hour or two for assembly, you're putting a lot of time into the system. Once you've built the system, you'll need to keep track of warranties for individual parts and troubleshoot and replace yourself if something breaks. If this is for a business, just buy something from Dell or HP with a service warranty and be done with it. Any money you save on parts will be quickly absorbed by the time you'll put into researching and building the system, as well as internal support after something breaks. If it's a personal computer and you have time to spare, go for it. I don't have a particular methodology for selecting parts, just buy whatever will meet your needs within your price range.
I usually look for raw speed of the memory bus. This gives me a list of the chipsets to investigate. Once I have that list, I look at the review sites to see which motherboards win in the realm of raw speed. I tend to ignore compatibility as I usually buy specific hardware that is known to work well with that motherboard.
With the ridiculously low price of memory these days, I think I will start looking for the motherboards that allow the largest amounts of RAM. Today, this means looking at server-class motherboards. Unfortunately, those boards are usually much slower than their consumer-grade cousins.
Fast memory is good whether you're a programmer or a gamer.
What methodologies do you use to decide on which parts to buy?
I first determine the machine's purpose. A development machine has different usage than a gaming machine, which has different usage than a server machine, which is different from a home user.
Starting from the usage scenario, and a budget, I look at what's available at what cost.
In selecting parts I generally look for the point where you're not getting the cheapest stuff (which is only a little cheaper than midrange) but is below the most expensive stuff (which has only a little more performance than midrange. It's not exactly in the middle - for some parts it's worth making a chart of cost/performance (or other metric which is important for that part) and seeing the part of the curve where you get the most bang for your buck.
Then I make compromises based on usage and cost.
Do you look at benchmarks?
Yes, but only peripherally. I use them primarily to make sure there isn't' a huge disparity between price/performance for a given component, but beyond that in this industry you generally get what you pay for in the mid-range.
For some usage scenarios it's important to take this into account, as certain usage may have critical requirements, such as when selecting between a hard drive and an SSD. SSD is better in most cases, but for a few usage scenarios a few cheap hard drives in raid is better because raw sequential access is more important than seek time.
Do you find things that have compatible bus speeds?
As long as the buses are compatible, then I'll optimize to requirements and cost. It may be vastly cheaper to go for the slower RAM now, and upgrade later if/when needed, even though the ram is slower than what is possible.
How do you know where to look to make sure everything will be compatible?
Only at the bleeding edge (very high end/high cost/newest) is this actually important.
In the mid range (best bang for the buck) the hardware has been out for months, has had at least 2-3 revisions, and has stable drivers. This isn't an issue unless you're using the latest and greatest MB, video card, OS, etc.
Do you buy a bare bones machine and extend it?
I buy my components individually and assemble it. Barebones machines, when I've done price comparison, are often not as good a deal as they appear to be at first - they usually make some small tradeoff. An equally good deal can be found with the parts you actually want with careful shopping.
-Adam
"To save some money I'm sure we would all prefer to build our own workstations"
Sorry, that's not why I build custom PCs. I've actually found it can cost more because resellers buy parts in bulk, while you will likely be buying individual OEM/retail parts. Also, cost is also increased if your time is valuable to you.
Good question though, I always focus on the Motherboard first and then everything else usually follows from that.
I narrow down motherboard options by looking at the primary use I expect the machine will have.
For example:
If you want a media centre PC you'll most likely want a board with HDMI and optical outputs, e-Sata, Firewire, stable ACPI (power managment) and good on board audio. Also, support for low power CPUs.
If you want a gamer PC, you'll likely want dual PCI-E 16 slot, Crossfire/SLI, fast RAM / FSB.
A developer PC will definitely want support for more then 2 monitors (eg have 2 PCI-E 16 slots) and as much RAM as possible.
For a basic email/web browsing and word processing PC, a lower cost board that will run quietly and provide good USB conectivity to the case, on-board video.
This helps to greatly reduce narrow down the range of boards that would meet your needs. Then you can start looking at CPU, RAM, Video Cards (if needed).
Just remember, if you don't get the foundation (motherboard) right, you'll never be 100% happy, no matter how fast your video card is.
Finally, because PC technologies are constantly changing, you realy need to do your research, every time you build a new PC. Don't rely on your knowledge from a year ago, much of it is probably no longer very useful.