4 ms·
I'm far from an expert here, but I guess it really depends on how well you can scale performance out of a single machine. For "embarrassingly parallel" problems
by dietr1ch 6y ago
I'm far from an expert here, but I guess it really depends on how well you can scale performance out of a single machine. For "embarrassingly parallel" problems that's "trivial", but if you constantly need data that would end up on a different machine then it might start making sense to have specialized hardware that greatly reduced those costs.
That being said, I'm also on your side in guessing that the problems that are money-wise worth running on a supercomputer are getting pretty limited as commodity hardware improves.
- semi-extrinsic 6y agoHPC is almost entirely "commodity hardware" on the CPU and GPU side these days. Meaning there is no problem for someone with a bit of money and time to get one node (~ one motherboard with all its associated parts) equivalent to a top supercomputer. As others have said, it is the interconnect that makes HPC special. The ability to run a job across tens, hundreds, thousands of nodes, where all nodes need to exchange large amounts of data with all other nodes every millisecond, without network being a huge bottleneck that leaves your CPUs idling. If your job is never big enough to outgrow the biggest and baddest single server money can buy, then indeed you don't need HPC. But if you are running e.g. a full fluid dynamics simulation of an F1 racecar, a wind turbine, or a jet airplane, you sure do.