I got a lot of flack from leadership at the lab where I worked in '97 and '98 when I bought Dual Pentium Pro 200MHz and later a Dual Pentium II 400MHz that ran Linux. They asked why I didn't buy HP, or SGI desktops. I pointed out that the grant i had- for $5K - wasn't really enough to buy a single machine (our lab had the cheapest SGI Indy, which IIRC was $10K) from HP or SGI, while for that amount of money, I could get dual processor experience, and more RAM/disk than the lowest-end HP or SGI. I ported many codes over from SGI/Tru64 to Linux on those machines.
Later I got a $15K budget to build a cluster, and again was asked why I didn't just buy a single SGI machine, and I pointed out I was able to get six PCs, each of which was 50% the speed of a single SGI machine, and I could cluster them to get experience running MPI jobs. 10 years later, all the labs had retired their non-Linux machines aand were running various white box PCs or HP servers running Linux, and I had 10+years of experience running HPC jobs on Linux clusters.
If you had a very large budget you could get high-end SGIs that had custom capabilities that no PC at the time could match but even then I specifically chose to not need those capabilities until they were commodity.
Comments
I got a lot of flack from leadership at the lab where I worked in '97 and '98 when I bought Dual Pentium Pro 200MHz and later a Dual Pentium II 400MHz that ran Linux. They asked why I didn't buy HP, or SGI desktops. I pointed out that the grant i had- for $5K - wasn't really enough to buy a single machine (our lab had the cheapest SGI Indy, which IIRC was $10K) from HP or SGI, while for that amount of money, I could get dual processor experience, and more RAM/disk than the lowest-end HP or SGI. I ported many codes over from SGI/Tru64 to Linux on those machines.
Later I got a $15K budget to build a cluster, and again was asked why I didn't just buy a single SGI machine, and I pointed out I was able to get six PCs, each of which was 50% the speed of a single SGI machine, and I could cluster them to get experience running MPI jobs. 10 years later, all the labs had retired their non-Linux machines aand were running various white box PCs or HP servers running Linux, and I had 10+years of experience running HPC jobs on Linux clusters.
If you had a very large budget you could get high-end SGIs that had custom capabilities that no PC at the time could match but even then I specifically chose to not need those capabilities until they were commodity.