I am also wondering where the tool reads the Active Cores (my guess it's derived from per-core C-state?) and this pesky Current Throttling values. Could it be that the reported throttling indication is incorrect? And the reported power dissipation with linpack is around 250W, much higher than that under the XTU benchmark, ~175W (and the temperature at 250W did not exceed 71☌, so it's a damn good cooler!). HPC workstation builder Puget Systems, so I do not believe the CPU could be in fact throttling. This score matches (actually, even exceeds by about 5%) that reported by the. The machine is constructed quite well, with decent cooling (the Noctua largest NH-14 dual-fan heatsink), and scores a bit above 1000 GFLOPs in the MKL linpack test (the quite recent one that came with the Parallel Studio XE 2018.2, and using AVX-512 instructions). HWINFO does not have an indication for it, either, but my guess it reflects cores C-states. There is a correlation in the graphs with the CPU clock (x36 vs x33), and another possibly interesting one with another parameter, Active Core Count. I am attaching a graph screenshot from the second, the 7960X machine, if it can help. In fact, none of the throttling or other flag type indications in HWINFO change during the event. The former may be mildly overclocked, the latter is not at all, just the default configuration from the motherboard. I also always see the "Current limit throttling" on two workstations with Skylake-X CPUs, one 7820X, another 7960X.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |