Hi Don Green; Re: "In most cases buying the fastest hard drive is the best way to improve performance, not the fastest CPU or ram.."
While this is possibly true, it really depends on what you need the computer to do. My suspicion is that workstations are dominated by (1) memory size, and (2) CPU performance. At least that is what I have found with the applications I use all the time.
There is no doubt that the machine responds faster to the hand and eye with a fast hard drive, but that isn't why I am buying a high end machine. I really don't care whether "Notebook" loads in 1.3 or 2.0 seconds. This kind of speed is what the average user notices in a typically underused machine, and you are right on that the disk speed is the thing to max out for such a user.
But with a workstation, I don't care about how long programs take to load from disk (more or less). What I care about is whether synthesis, place and route takes 40 minutes or 20 minutes. Chip design code doesn't use the hard disk at all, for the vast majority of those 40 minutes, unless there is so little memory that the application (or its data) doesn't all fit in the memory at once. In that case, the program will thrash the life out of a hard disk and slow down incredibly, or even crash.
You get an enormous performance improvement by adding enough memory to eliminate disk thrashing. Something like 20 to 1 speed improvement. From then, increasing memory size doesn't do help any further at all, but increases the amount of time required to boot...
Once you have enough memory, then you get almost a 1 to 1 improvement by increasing CPU speed.
Because of the incredible importance of having enough DRAM to eliminate disk thrashing, I always put as much as possible on my workstation. I've got 768MB on my latest, but I only seem to need around 384MB so far. But with newer stuff coming out later this year, I will probably get into the 512 to 768MB area. With these memory sizes, RDRAM is not a viable choice, given the other areas where the same money could be put. PC800 may provide a 1% performance improvement over PC133 CL3, but PC133 CL2 is faster than PC800, costs under half, and is available with cheaper motherboards/machines as well.
These 1% performance differences aren't worth paying big bucks for when I can take the same money and add 200MHz to the processor, which will improve performance by 10% or more.
These basic facts are why memory has to be cheap to sell to (rational) customers. The best workstation is one with lots of cheap memory, and the money saved is then put into a faster processor.
-- Carl |
