Posts by Tex1954
|
21)
Message boards :
Number crunching :
Optimized app ?
(Message 380)
Posted 7 Aug 2015 by Tex1954 Post: Sesef is a great programmer with strong interrests in BOINC. He optimized a lot of apps for many projects. For many years. IF the app is unstable, work in progress, etc., then Sesef should be working with the project developers to test it and NOT put it into production for ONLY one team. I went to Sesef's website and saw only two optimized apps. Didn't see any mention of this project. If the app is legitimate, then the source and compile options and platform revisions and changes should be publicly documented and/or at least submitted to the project developers for evaluation. I don't think using a BOINC project as a "test bed" for personal testing is a safe idea... in fact, I think the developers should BLOCK all data from a suspect app until verified. Having said that, I WANT A COPY! LOL! 8-) |
|
22)
Message boards :
Number crunching :
Optimized app ?
(Message 379)
Posted 6 Aug 2015 by Tex1954 Post: Is this perhaps the results of one of those OPTIMIZED apps? http://denis.usj.es/denisathome/workunit.php?wuid=3714787 I want it! It goes many times (almost 10 times!) faster!!!! 8-) PS: My 2p 3.33GHz X5680 setup is Comp# 4552... |
|
23)
Message boards :
Number crunching :
OSX App
(Message 371)
Posted 21 Jul 2015 by Tex1954 Post: Thanks for your patience and understanding and help! In the (further) future, I hope to do that as a retirement exercise...and for fun. Thanks again! 8-) |
|
24)
Message boards :
Number crunching :
OSX App
(Message 369)
Posted 20 Jul 2015 by Tex1954 Post: WOW! That data processing stuff is actually an order of magnitude (at least) more complicated than the machine control I spent most of my life developing. Seems my basic idea's on the subject in terms of optimizations will have to be in the structure and not content unless "I" dive deep into your project. I assure you, I am not qualified to do that. But, for the sake of mental exercise, how would I test things if I decided to try some optimizations or different compilers myself? 8-) PS: Keep up the great work! As it turns out, I may put together a "Hackintosh" later to play with because of this issue... |
|
25)
Message boards :
Number crunching :
OSX App
(Message 366)
Posted 19 Jul 2015 by Tex1954 Post: I find it completely weird that a i7-3770 Macbook (or whatever) can go faster and take 1/3 the time as a Xeon X5680... http://denis.usj.es/denisathome/workunit.php?wuid=3008079 This really makes me think some special optimization or special intrinsic instruction is being used on the 3770 that isn't available or not coded for on the Xeon X5680 I would REALLY like to understand this. I can guess at a lot of things, like SSE2 being used on Mac and not on X5680 or some such thing... Maybe the AVX or F16C instruction that the X5680 doesn't have? Anyway, that wouldn't explain the huge difference compared to my 4.5GHz i5-3570K setup since it has the same instructions (except SSE4) as 3770... Anyways, still scratching head... 8-) |
|
26)
Message boards :
Number crunching :
OSX App
(Message 363)
Posted 18 Jul 2015 by Tex1954 Post: In my day, I wrote code for DSPs and CPUs in assembly language, long before the C language was invented. I used PASCAL and FORTRAN and others and at times I looked into the code generated with a debugger and discovered a lot of major differences on how functions were treated. Are you using GCC 5.2 for Linux? Would another work better? Have you looked into code optimizations or looked at what is being generated at the assembly level? Simple things such as calling the OS too often can slow things down tremendously... To be fair, I notice this on many BOINC projects and it suggests two things: 1) YOU the BOINC project would do better to make a Linux or Windows Wrapper and run Apple compiled tasks in it... 2) A detailed analysis of what the real differences could be might improve your overall project speed up to 50%. Another problem I heard discovered on a new project was the code was checking if it was time to checkpoint far too often causing a similar slowdown... A simple thing like using Macros or Inline Code could adjust for the compiler idiosyncrasies... one never knows without deep checking... 8-) PS: If DENIS could figure this out, I am sure other projects would benefit... There is absolutely NO WAY an old iMac or Macbook should beat my 4.5GHz 3570K setup... or many other setups people run for that matter!!! |
|
27)
Message boards :
Number crunching :
OSX App
(Message 362)
Posted 18 Jul 2015 by Tex1954 Post:
Well, seems there could be something to improve somewhere... For instance, check this out... http://denis.usj.es/denisathome/workunit.php?wuid=2953065 That supposed 3770 is in a laptop (computer #3270) and no way is it faster than my 4.5GHz 1866MHz Mem setup (computer #4416). 3270: Measured floating point speed 4583.16 million ops/sec Measured integer speed 12064.15 million ops/sec 4416: Measured floating point speed 5726.15 million ops/sec Measured integer speed 30458.46 million ops/sec That task took my setup 22.3 minutes and the Darwin setup took 17.9 minutes. THAT is a HUGE 25% difference, yet, my setup is 25% faster!! There MUST be a better compiler out there to use... One point I would like to make is I have seen this before on another project and it turned out a loop count in the software was wrong causing a huge speed difference... the output seemed okay, but close examination determined that the output wasn't 100% the same. Anyways, I think this needs to be looked into a bit closer... 8-) |
|
28)
Message boards :
Number crunching :
OSX App
(Message 358)
Posted 17 Jul 2015 by Tex1954 Post: I am wondering WHY the DARWIN OS (mac OSX?) does tasks so much faster than Linux or Winderz... Seems to me, everywhere I look in statistics, Darwin does things faster on lesser CPU setups... Is it real or a time recording error or what? The points these machines make seem to support faster processing speed... (scratching head) 8-) |
Previous 20
