Benchmarks using single problems are often of limited value or even totally useless. E.g. the Spec benchmark measures mainly memory throughput and the size of the largest cache. I would like to see more Hint benchmark results because these are are much more meaningful than single-number benchmarks. Hint is designed to compare different machines in different memory regimes (like cache, RAM, swap) and not different combinations of machines, problems and compilers (like Spec).

Source Code

Source code ( Tar GZip 17 KB ) of the Hint benchmark I have used here. The Hint benchmark is Open Source (GPL).
Data ( Tar GZip 76 KB ) from which the results have been generated.

Results

The results below are roughly sorted by peak performance with data type "double".

Thanks to Oliver Wenisch (LRZ Munich) who has contributed much of the data.

If you can contribute more benchmark results, please let me ( mail@Franosch.org ) know and I will collect the data here.

Intel Xeon Dual Core E3110 3.0 GHz, 4 GB RAM

AMD Athlon Dual Core XP2 6000+, 3.1 GHz, 8 GB RAM

AMD Opteron Dual Core 2218, 2.6 GHz, 16 GB RAM

AMD Opteron Quad Core 8380, 2.5 GHz, 64 GB RAM

AMD Opteron Dual Core 275, 2.2 GHz, 16 GB RAM

The fall off of the "double" performance after about 1 GB and end of the data at about 5 GB does not occur because the machine becomes slow, but because of the limited accuracy of the double data type. Only the "long long" data type sufficiently exercises the machine.

Hint Benchmarks - T35 TUM

Contents