??? 17/11/2002 1:24:05 ??, ?/? Marcel Kilgus <[EMAIL PROTECTED]> ??????:
> >Bill Waugh wrote: >> 1. speed of Q60/60 in bogomips >> 2. speed of QPC2 v3 on a 1.8 gh celeron based PC in bogomips > >I probably would have stayed quiet on the subject (because of the >famous wars), but as you asked for BogoMIPS values I need to write >some lines: > >I don't have exact figure for either of the two, but I can tell you >nonetheless that QPC will easily lose this contest (in a devastating >way) because BogoMips has no real life meaning at all (and I don't >mean the usual "benchmarks don't tell you everything", I really mean >it has NO meaning at all). The benchmark basically consist of 1 line: > With a JIT compiler, QPC would probably win there easily for the reason Marcel said). It's best I think to get ahold of someone with a Q60, test it and then run the same (Dhrystone) tests on your QPC demo (You cannot write on the disk but you can run programs ;-) I don't think that there is any performance loss between the regular and the demo versions... Only YOU can test how QPC will fare on your machine :-) >Just to show what this means: the Q40 has a BogoMIPS value of about >27. This means that the 68060/80 has an impressive 600% increase in >speed while only having 100% more clock rate. If you however compare >the Dhrystone (which is far from being perfect, but at least much >better than BogoMIPS) results you get 36443 to 100603, i.e. 276% more, >which is much more realistic (tasks involving hardware like graphics >will BTW usually gain less). Don't forget that the 68060 is a superscalar processor, so processing percentages are quite higher... >From the m68k FAQ: "MC68060: This is the latest and most powerful member of the 68K family. The '060 is designed as an upgrade from a '040 with 2.5 to 3.5 times the performance of the 25 mhz '040. It uses Superscalar pipelined architecture which means it can perform more than one instruction at a time. The 68060 allows simultaneous execution of two integer instructions (or 1 integer and 1 float instruction) and one branch during each clock cycle. A branch cache allows most branches to execute in zero cycles. The '060 offers 100 MIPS @ 66mhz and 250 million operations per second @ 50 mhz. SPECint = 50 @ 50Mhz. " >OK, one might ask, "then how do they compete in real life". In that >case I must say this depends on how you define real life. I have no >doubt that the 68060/60 will beat the mentioned QPC/Celeron >combination in pure integer number crunching any time, any place (only >the most recent PCs come within the range of close competition). On >the other hand I'm sure it will lose when there's a lot of QDOS >floating point arithmetic to do (SBASIC programs), in matters of I/O >(WIN, PAR) or certain basic high colour graphics tasks (e.g. draw a >block or move a rectangle, very important for the PE). With FPUFNs performance increases dramatically. Even on the Q40 it's there to be seen... In complex math-oriented operations (that I am doing at least;-) like absolute pixel based circles, fills and screen effects (all of the above use floating point arithmetic) and school work (Time Series analysis, ANOVA tables, coefficient correlations on data sets where again floating point arithmetic is needed 99,99% of the time) the performance increase is dramatic with FPUFNs... The FPUFNs117.zip readme file isn't joking... in some categories performance enhancement of 800% over the regular 68EC040 (measurable on the QXL too as I do have a full 68040 there as well) is realistic (SOME categories not all.. overall average performance boost from regular QDOS functions to FPU enhanced functions is about 300% (and that's only on the regular Q40). Now on the Q60 I can only guess... Marcel is indeed right about the cache and BogoMips and he is also right to tell you that the best benchmark should be Dhrystone21 (From Thierry, get the gcc Compiled version). I really cannot tell how it "feels" on disk access, but given the fact that Marcel has heavily optimised the code and that PC drive access is extremely fast nowadays... QPC must easily outperform the Qx0 there. A great bonus is the DOS device that the Qx0 doesn't have... -However DOS formatted drives can be read by Qx0 Linux- The PAR device is a lot better on the QPC as well, as for the graphics I really cannot tell. I am very satisfied with the Q40 (once the FPU is turned on) but under normal operation QPC is a lot faster on the screen (the 128bit graphics engines on modern cards are very helpful you see :-) > >Of course. I'd just not rely on BogoMIPS. There's a reason some people >advertise with it ;-) I'd suggest Dhrystone, QSBB (on identical SMSQ/E >versions) and Test909 (the latter two being especially good if the >main use for the machine is SBASIC). > Marcel is absolutely right here as well (see also above) >> I used to be indecisive but now I'm not so sure, > >Even more confused now? In that case it's the best to get both ;-) > Yes :-) Sell some old QL stuff to some poor collector on eBay and there you are :-) (I've seen old non working QLs go for 60 pounds EASY) >And now finally off to bed, sun's raising soon ;) Hehe my turn as well :-) Phoebus
