Reply to thread

Message: <blockquote data-quote="Herr R aus B" data-source="post: 847539" data-attributes="member: 62807">Hi! If the answer to your question would be so easy, then one would maybe simply say: Just do the math! <img src="data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///yH5BAEAAAAALAAAAAABAAEAAAIBRAA7" class="smilie smilie--sprite smilie--sprite1" alt=":)" title="Smile    :)" loading="lazy" data-shortname=":)" /> But well... I try some brief explainations eventhough not knowing anything about your personal computer skill level - so no offence meant and surely no discrimination <img src="data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///yH5BAEAAAAALAAAAAABAAEAAAIBRAA7" class="smilie smilie--sprite smilie--sprite8" alt=":D" title="Big Grin    :D" loading="lazy" data-shortname=":D" /> this K computer has more than 80000 computer nodes - says Wiki. Without knowing much more, this might equal 80000 CPUs, maybe multicore CPUs (would be time for at least some little math then). One approach to do the conversion on such a system achieving maximum performance might be, to use some multithreading - means, simply spread as small as possible computing portions over a possibly high number of CPUs and thus get into real parallel computing. but then - into how many parallel equal tasks can one split the conversion? there very likely would be a reasonable limit of seperately runnable threads as it surely wouldn't make sense to try a framewise conversion or so. anyway, this might really reduce the conversion time. but then you have to take into account the overhead of merging all seperately converted parts of the original file also including the amount of workload for all the synchronization stuff to be done when splitting the main task into little tasks, waiting for them to end, collecting their results and joining them together again. Maybe you can break it down to 100 threads or so - but thats nothing compared to the number of computing nodes, means: this K thing will be the absolute overkill <img src="data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///yH5BAEAAAAALAAAAAABAAEAAAIBRAA7" class="smilie smilie--sprite smilie--sprite1" alt=":)" title="Smile    :)" loading="lazy" data-shortname=":)" /> Anyways... One could also very roughly judge from the performance. VERY ROUGHLY AND VERY UNPROFRESSIONAL just to display the relations... This K computer is supposed to execute a maximum of 10 petaflops, meaning 10 * 10^15 (or 10^16) floating point operations per second - thats 10.000.000.000.000.000 floating point operations per second. Now randomly asking google I found a <a href="http://www.notebookjournal.de/storage/show/image/image4c52e39fabad6" target="_blank">report</a>, stating, that the ATOM D510 (1,6GHz) executes some 0,735 gigaflops - being generous, let's estimate the performance of your ATOM with 1,8GHz to be 1 gigaflop meaning 1 * 10^9 floating point operations - thats 1.000.000.000 floating point operations per second. That means, the K computer could (VERY!!!) theoretically be 10^7 (10 millions) times faster, than the ATOM. Now the final math is on you: 25,2 * 10^3 seconds (7 hours) divided by 10^7. If MY math didn't give up on me, that's supposed to be 25,2 * 10^-4 seconds. Thats 0,25 * 10^-3 and that would be a quarter of a millisecond. ROUGHLY, THEORETICALLY and absolutely NOT PRACTICALLY - but it might tell you something about the relations. The resulting fact is, that you by no means would even realize, that some time elapsed. Your brain's latency regarding the notification of say different audio events is somewhere between 4 and 10 milliseconds. Your eye is even slower - you can distinguish roughly 25-30 changes per second before your eye simply senses smooth and continuing motion. Surely there will now pop up some math guys, physics guys and computer science guys (I myself am one of the latter) and point out, that my calculations are to rough, don't take into account this and that, blablabla... But even if you move the scale up or down by some factor 10^3 , the entiry comparison is senseless, because anyway: coding the conversion of an HD .ts file on this K thing might take much longer than those 7 hours conversion time using the ATOM and if thats no point (no isnt, you didn't ask for that, I know), the named overhead for controlling the multithreading stands in no reasonable relation to the parallelifying (sorry - my english obviously goes berserk on me) and thus the effect of accelerating the conversion process... btw - i missed one scenario - its the scenario where we could assume, that one computing node of the K computer resembles one core of the ATOM. then maybe one could assume, that if the 80000 nodes provide for a performance of 10petaflops, 2 nodes of the K computer might perform at... You know what? That particular math is definately on you now <img src="data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///yH5BAEAAAAALAAAAAABAAEAAAIBRAA7" class="smilie smilie--sprite smilie--sprite1" alt=":)" title="Smile    :)" loading="lazy" data-shortname=":)" /> Question answered? No? Well - it's Off Topic anyways. <img src="data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///yH5BAEAAAAALAAAAAABAAEAAAIBRAA7" class="smilie smilie--sprite smilie--sprite2" alt=";)" title="Wink    ;)" loading="lazy" data-shortname=";)" /> Best regards form Berlin/Germany Axel PS: Scientists - don't kill me verbally, please!!! I just tried to make the overall relations understandable and imaginable <img src="data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///yH5BAEAAAAALAAAAAABAAEAAAIBRAA7" class="smilie smilie--sprite smilie--sprite8" alt=":D" title="Big Grin    :D" loading="lazy" data-shortname=":D" /></blockquote>

Verification