On 9/10/2012 1:13 PM, Michael Haberler wrote: > Am 10.09.2012 um 18:48 schrieb Jon Elson: > >> Michael Haberler wrote: >>> here's linpack figures for the Rpi and an Intel D525. >>> >>> the D5252 is almost a factor of 6 faster than the Rpi for this benchmark >>> >> Thanks much for doing this! Not a great result, though. My feeling is >> that the >> Atom processors are marginal for LinuxCNC, especially in cases with long >> programs or contouring, where the slowness would be more obvious. >> So, if the Pi is 1/6th that speed, that will not be good at all. >> Exactly how this compares on LinuxCNC, where there is also a lot of >> non-FP code, isn't clear. If LinuxCNC was run with the GUI on another >> system, it might be OK, and for special projects which are not typical >> machines, it might be fine. > Well, I'm not envisaging the whole LinuxCNC thing on the Rpi, but maybe just > the HAL/RT backend > > if so, an interesting question would be: where is the FP performance actually > needed most? Interpreter: I dont think so, probably motion. UI a bit for > drawing? But that is just a feeling. > > I'm unsure to qualify "FP-bound" and am not aware of any measurement methods > for that > > maybe a way to approach this is to measure the servo function chain times and > see how much that eats into reserves, I guess some timestamps are there > already but I've never looked into them > > -m >
That's the trouble with synthetic benchmarks. They remind me of the old joke about the drunk looking for his lost car keys out in the street under the bright streetlamp instead of in the dark alley where he lost them. Keeping in mind the hardware that was available when EMC was retooled as EMC2, the RPi (and its ARM relatives; I have no brand loyalty) doesn't look so bad. (Just noodling out loud here) I haven't focused my attention directly on floating-point operations usage since my minicomputer days. Could running one of the available profiling tools on the LinuxCNC simulator give any insight into the practical consequences of FP as used in LinuxCNC operation? Preferably, it would be something that could be done on several different platforms to get some comparable information. I think we all have different ideas of how we would apply ARM-based boards. Whether or not we find a single solution for LinuxCNC in the near-term, I believe the effort is worthwhile. Let a thousand flowers bloom, and all that. It's also true that the marketplace is moving faster than we are. For several years now, the Cortex A8 has been getting dinged for its floating-point performance. Stung by criticism, ARM Holdings has been assuring influential analysts that newer architectures will have better floating-point performance. Seeing how well they responded to previous performance criticisms (a matter of tracing through a decades worth of blogs---ugh), I suspect they will succeed, which means a single solution becomes more likely over time. As for running an RPi headless, the darn things are so cheap one could waste one as the X-server for the other and still pay less than the price of an Atom MB. (I seem to remember saying the same thing regarding marginally useful VIA MBs years ago. The words may change but the tune remains the same.) Now if only the RPi brought out as many pins as the BeagleBone does. Regards, Kent ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Emc-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/emc-developers
