<html><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8"></head><body ><div>I think im coming from a different aspects not from academia but more from an enterprise aspect in terms of clustering. Basically a data center clustered into one big one so to speak </div><div><br></div><div><br></div><div><div style="font-size:75%;color:#575757">Sent from Samsung Mobile</div></div><br><br><br>-------- Original message --------<br>From: Prentice Bisbal <prentice.bisbal@rutgers.edu> <br>Date: 01/07/2014 18:51 (GMT+01:00) <br>To: jaquilina@eagleeyet.net <br>Cc: beowulf@beowulf.org <br>Subject: Re: [Beowulf] Gentoo in the HPC environment <br> <br><br>On 07/01/2014 12:35 PM, Jonathan Aquilina wrote:<br><br>> I think my question though is this. can one see negative impacts if the<br>> compiler gets upgraded regardless of if its gcc or intel.<br>If you're talking about the distro-provided compiler, no. They are <br>usually tested well by the distro maintainer, and any updates within a <br>distro's release version will only be maintenance security and bug fixes.<br><br>Now if you're talking about installing GCC 4.7.2 on your own and then <br>updating to 4.8 or doing something similar with the Intel Compilers or <br>Portland Compilers, it's always a good idea to install the different <br>versions in their own install tree and keep the old versions around. <br>It's quite common for new compilers to have new bugs. I never delete old <br>versions of compilers. In academia, researcher like to be able to <br>reproduce their work, or compare their results using a new compiler to <br>an old compiler. Removing the old version breaks that. If a compiler <br>version is really, really old, I'll ask my users if it's safe to remove it.<br><br>--<br>Prentice<br>><br>>> On 07/01/2014 12:17 AM, Matt Wallis wrote:<br>>>> On 01/07/14 13:45, Jonathan Aquilina wrote:<br>>>>> This question probably sounds like a stupid one, but what difference<br>>>>> in an<br>>>>> HPC environment and to parallel written code does compiler version<br>>>>> make?<br>>>> Depends on the day of the week, the processor, the code, and your<br>>>> cluster really.<br>>>><br>>>> Sometimes code from GCC is faster, sometimes code from the Intel<br>>>> compiler is faster.<br>>>> Some people write code that only compiles with the Intel compiler, or<br>>>> link against libraries distributed with the compiler.<br>>>><br>>>> Then of course you might have a BlueGene with PPC cores rather than<br>>>> x86, and while the difference in performance between GCC for PPC and<br>>>> the IBM XL compilers is slight these days, it wasn't always the case.<br>>> One of my researchers has found that his code that is integer-heavy runs<br>>> up to 5x faster on our BG/P when compiled with GCC over the same code<br>>> compiled with the IBM XL compilers. For floating point heavy codes, the<br>>> IBM XL compiles produce much faster code.<br>>><br>>>> Intel also distributes highly tuned versions of some maths libraries<br>>>> with their compilers, and libraries designed for offloading work to<br>>>> the the Xeon Phi, without recompiling the application.<br>>>><br>>>> And then when it comes time to debug why the application doesn't work,<br>>>> commercial compiler suites often come with very good debugging tools.<br>>>><br>>>> Matt.<br>>>><br>>> _______________________________________________<br>>> Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing<br>>> To change your subscription (digest mode or unsubscribe) visit<br>>> http://www.beowulf.org/mailman/listinfo/beowulf<br>>><br>><br><br></body>