On Tue, 09 Feb 2016 10:36:02 +0100, Peter Graf wrote: > Am 08.02.2016 21:19, schrieb Thierry Godefroy: > > > > Q60 @ 66MHz (overclocked 68060RC50): > > 127.82 BogoMips > > Writethrough cache mode: 24.937 VAX Mips/43813.5 Dhrystones/s (5M runs) > > Copyback cache mode: 47.812 VAX Mips/84005.4 Dhrystones/s (5M runs) > > A 68060 with fully activated caches should deliver (almost) exactly > twice the clock frequency in BogoMIPS. And for my machines, it did, at > least under Linux.
The memory cache mode (write back or write through) is irrelevant for BogoMips since no data is written to the memory during the timing loop: only the pipeline and execution units number count, and the 68060 indeed can execute two simple instructions per clock. > I can't explain myself why it is 127.82 and not 132.0 BogoMIPS. The maximum accuracy of the measure is determined by the granularity of the timer (for BogoMips, it's the 50Hz frame timer, so the maximum error is 2* 1/50th of a second on the timing, and since the timing loop lasts for 10 seconds, the accurracy of the final measure is of 0.5% or so. There's also the problem of the linked frame timer routines: at each tick of the 50Hz timer, even in supervisor mode, QDOS/SMS executes those routines. Depending on how long they are and how many have to run, they induce a (small, but measurable) overhead on any running job. I said I made the measures with no other job running, but I got many extensions loaded by default on my QDOS/SMS systems, some with scheduler loop routines (mostly used by device drivers), so there's a slight overhead (but it's the same overhead for all systems, since I booted with the same extensions for all of them). In fact, what surprises me, is that you got exactly twice the clock speed in BogoMips... You should get slightly below that value, because even with the default device drivers, there's a slight OS overhead... Are you running BogoMips v1.5 ? Thierry. _______________________________________________ QL-Users Mailing List
