(v13) Hardware selection

This page applies to Harlequin v13.1r0 and later; and to Harlequin Core but not Harlequin MultiRIP

The discussion in (v13) Initial performance tuning recommendations implies that you have already made a decision on the server hardware on which you will be running the RIPs. If that decision has not been set in stone, we recommend that you repeat the optimization tests on alternative hardware as well, in case they perform better than expected.

In particular Harlequin is sensitive to the size of L2 and L3 caches in the CPU and to bandwidth to main memory, which are often not highlighted in technical specifications.

For some classes of relatively simple jobs fewer cores with higher clock speed is likely to deliver the required data rate at a lower cost. For a very high data rate requirement, especially combined with greater graphical complexity, a larger number of RIPs using more cores (which inevitably tends to mean lower clock speeds) will provide a better cost/performance result.