I tend not to agree with many of the posts here. RAM speed on a desktop can make a big difference but it's typically used rather differently on a server. Programs on a server tend to be smaller but more instances of them. On a desktop they are few of them but they're bigger. On a server RAM's biggest function on a server by far is caching. So quantity matters much more than speed. I'm not trying tosay that you won't get a few more #s on UnixBench or other benchmarking, but in real world use the amount of RAM is going to have a bigger effect (by how much is in cache) than the speed. A real world example is a recent upgrade I did on a 7 yr old mid-range HP-UX box - close to 1/4 mil when bought. Loads were running 4 to 5 with frequent "naps". So I quadrupled the RAM in the system. Loads now run 0.1 to 0.2 with 200 or so active users (interactive real-time transaction processing system with heavy database work). sar -b (system activity reporter) now shows the system is running 97%+ disk read cache hits. And the "naps" are gone. The RAM used ... Reg. ECC PC-133!
And while a company may use desktops, I wouldn't recommend it. Server components including RAM is designed differently (and is actually slower) than the same bus speed desktop RAM in the case of registered memory which holds the data for 1 write cycle in registers. Plus server memory should be Error Checking and Correcting or ECC... How bad would it be to take down the whole server in a kernel panic because of a single bit error? I won't use anything but Registered ECC memory in our systems. It's just not worth it.
More important to me for most servers is the disk subsystem. All our new servers are being configured with hardware-based RAID10 controllers with "RE" or RAID Edition drives. Again server components designed for a different use - WD's RE drives are rated for twice the MTBF than their regular drives.
And while a company may use desktops, I wouldn't recommend it. Server components including RAM is designed differently (and is actually slower) than the same bus speed desktop RAM in the case of registered memory which holds the data for 1 write cycle in registers. Plus server memory should be Error Checking and Correcting or ECC... How bad would it be to take down the whole server in a kernel panic because of a single bit error? I won't use anything but Registered ECC memory in our systems. It's just not worth it.
More important to me for most servers is the disk subsystem. All our new servers are being configured with hardware-based RAID10 controllers with "RE" or RAID Edition drives. Again server components designed for a different use - WD's RE drives are rated for twice the MTBF than their regular drives.