Intel C++ reported 37% faster than VC++
-
Open Magazine has a story at http://www.open-mag.com/754088105111.htm reporting that "Intel's C++ Compilers for Linux and Windows smoke GNU C and MS Visual C++ in number crunching benchmarks." For real-world numerically intensive computations, they find that Intel's C++ compiler is 37% faster than VC++ 6 and 25% faster than VC++ 7 beta. Linux bashers should note that Intel is over 47% faster than Gnu C++ and that code compiled with the Intel compiler runs about 10% faster under Windows XP than under Linux. They report that they have noted for several years that VC++ under Windows is significantly better than GCC under Linux but had not had a way to apportion blame between the compiler and the operating system. Now they find that even using the same compiler, Linux just can't keep up with Windows for high-performance scientific computing! This is particularly interesting given that Linux is particularly favored over Microsoft in the science and engineering communities, where it is widely believed that it is faster than Windows for numerically intensive computations. To me (a true VC++ zealot, if a cranky one), the fact that the Intel compiler blows the doors off VC++ is great news. Microsoft needs serious competition to keep it from getting fat and lazy, and it sounds as though Intel is going to provide a much-needed thorn in the side of MS's VC++ developers to get them off their rear ends as far as attention to numeric performance goes. (How lazy are they? I reported bugs with their standard C++ library several years ago (e.g.,
std::numeric_limits<double>::signalling_NaN()returns infinity, not a NaN) and these bugs are not even addressed in the Knowledge Base, much less fixed in any of the service packs). -
Open Magazine has a story at http://www.open-mag.com/754088105111.htm reporting that "Intel's C++ Compilers for Linux and Windows smoke GNU C and MS Visual C++ in number crunching benchmarks." For real-world numerically intensive computations, they find that Intel's C++ compiler is 37% faster than VC++ 6 and 25% faster than VC++ 7 beta. Linux bashers should note that Intel is over 47% faster than Gnu C++ and that code compiled with the Intel compiler runs about 10% faster under Windows XP than under Linux. They report that they have noted for several years that VC++ under Windows is significantly better than GCC under Linux but had not had a way to apportion blame between the compiler and the operating system. Now they find that even using the same compiler, Linux just can't keep up with Windows for high-performance scientific computing! This is particularly interesting given that Linux is particularly favored over Microsoft in the science and engineering communities, where it is widely believed that it is faster than Windows for numerically intensive computations. To me (a true VC++ zealot, if a cranky one), the fact that the Intel compiler blows the doors off VC++ is great news. Microsoft needs serious competition to keep it from getting fat and lazy, and it sounds as though Intel is going to provide a much-needed thorn in the side of MS's VC++ developers to get them off their rear ends as far as attention to numeric performance goes. (How lazy are they? I reported bugs with their standard C++ library several years ago (e.g.,
std::numeric_limits<double>::signalling_NaN()returns infinity, not a NaN) and these bugs are not even addressed in the Knowledge Base, much less fixed in any of the service packs). -
Open Magazine has a story at http://www.open-mag.com/754088105111.htm reporting that "Intel's C++ Compilers for Linux and Windows smoke GNU C and MS Visual C++ in number crunching benchmarks." For real-world numerically intensive computations, they find that Intel's C++ compiler is 37% faster than VC++ 6 and 25% faster than VC++ 7 beta. Linux bashers should note that Intel is over 47% faster than Gnu C++ and that code compiled with the Intel compiler runs about 10% faster under Windows XP than under Linux. They report that they have noted for several years that VC++ under Windows is significantly better than GCC under Linux but had not had a way to apportion blame between the compiler and the operating system. Now they find that even using the same compiler, Linux just can't keep up with Windows for high-performance scientific computing! This is particularly interesting given that Linux is particularly favored over Microsoft in the science and engineering communities, where it is widely believed that it is faster than Windows for numerically intensive computations. To me (a true VC++ zealot, if a cranky one), the fact that the Intel compiler blows the doors off VC++ is great news. Microsoft needs serious competition to keep it from getting fat and lazy, and it sounds as though Intel is going to provide a much-needed thorn in the side of MS's VC++ developers to get them off their rear ends as far as attention to numeric performance goes. (How lazy are they? I reported bugs with their standard C++ library several years ago (e.g.,
std::numeric_limits<double>::signalling_NaN()returns infinity, not a NaN) and these bugs are not even addressed in the Knowledge Base, much less fixed in any of the service packs). -
Open Magazine has a story at http://www.open-mag.com/754088105111.htm reporting that "Intel's C++ Compilers for Linux and Windows smoke GNU C and MS Visual C++ in number crunching benchmarks." For real-world numerically intensive computations, they find that Intel's C++ compiler is 37% faster than VC++ 6 and 25% faster than VC++ 7 beta. Linux bashers should note that Intel is over 47% faster than Gnu C++ and that code compiled with the Intel compiler runs about 10% faster under Windows XP than under Linux. They report that they have noted for several years that VC++ under Windows is significantly better than GCC under Linux but had not had a way to apportion blame between the compiler and the operating system. Now they find that even using the same compiler, Linux just can't keep up with Windows for high-performance scientific computing! This is particularly interesting given that Linux is particularly favored over Microsoft in the science and engineering communities, where it is widely believed that it is faster than Windows for numerically intensive computations. To me (a true VC++ zealot, if a cranky one), the fact that the Intel compiler blows the doors off VC++ is great news. Microsoft needs serious competition to keep it from getting fat and lazy, and it sounds as though Intel is going to provide a much-needed thorn in the side of MS's VC++ developers to get them off their rear ends as far as attention to numeric performance goes. (How lazy are they? I reported bugs with their standard C++ library several years ago (e.g.,
std::numeric_limits<double>::signalling_NaN()returns infinity, not a NaN) and these bugs are not even addressed in the Knowledge Base, much less fixed in any of the service packs).Micrsoft is getting very sloppy. I reported for instance that the Pocket PC 2002 SDK they have made available is corrupt on the webserver. I tried to download it serveral times, over different connections, to no avail. doh! Regards Ray "Je Suis Mort De Rire"
-
It is probably designed to make specific usage of the P3 and P4 instruction set... The VC++ 6 compiler is designed to work the same on ALL processors... There is a processor pack available for VC++ 6, anyone tried this one ??? -= Forgive my English =-
-
-
Thanks! :) Simon "Does the room have a CView?" Sonork ID 100.10024
-
Thanks! :) Simon "Does the room have a CView?" Sonork ID 100.10024
Stay away form it. It can't be deinstalled, the code generation is not selectable to P3 or P4 (only in ASM statements). The code generator has been rebuild but generates bugs in exception handling in debug builds (Not the C++ exception is the bug, but the exception handling code itself generates another OS exception) Im shure its not a bug in my exception handling code (does nothing at the moment only beeing catched to not bubble up in callers). This has been working OK without PP, so I blame the PP for it. Compiling on a PC without PP results in a clean run. I am a signature virus! Help me spread and copy me to your sig! Ooops I am infected
-
Open Magazine has a story at http://www.open-mag.com/754088105111.htm reporting that "Intel's C++ Compilers for Linux and Windows smoke GNU C and MS Visual C++ in number crunching benchmarks." For real-world numerically intensive computations, they find that Intel's C++ compiler is 37% faster than VC++ 6 and 25% faster than VC++ 7 beta. Linux bashers should note that Intel is over 47% faster than Gnu C++ and that code compiled with the Intel compiler runs about 10% faster under Windows XP than under Linux. They report that they have noted for several years that VC++ under Windows is significantly better than GCC under Linux but had not had a way to apportion blame between the compiler and the operating system. Now they find that even using the same compiler, Linux just can't keep up with Windows for high-performance scientific computing! This is particularly interesting given that Linux is particularly favored over Microsoft in the science and engineering communities, where it is widely believed that it is faster than Windows for numerically intensive computations. To me (a true VC++ zealot, if a cranky one), the fact that the Intel compiler blows the doors off VC++ is great news. Microsoft needs serious competition to keep it from getting fat and lazy, and it sounds as though Intel is going to provide a much-needed thorn in the side of MS's VC++ developers to get them off their rear ends as far as attention to numeric performance goes. (How lazy are they? I reported bugs with their standard C++ library several years ago (e.g.,
std::numeric_limits<double>::signalling_NaN()returns infinity, not a NaN) and these bugs are not even addressed in the Knowledge Base, much less fixed in any of the service packs).Jonathan Gilligan wrote: the fact that the Intel compiler blows the doors off VC++ is great news. Microsoft needs serious competition to keep it from getting fat and lazy, and it sounds as though Intel is going to provide a much-needed thorn in the side of MS's VC++ developers to get them off their rear ends as far as attention to numeric performance goes. A very refreshing attitude. I agree completely. MS gets a lot right, but *are* too lazy (or financially unmotivated) to fix what they get wrong. "Thank you, thank you very much" Elvis.
-
Stay away form it. It can't be deinstalled, the code generation is not selectable to P3 or P4 (only in ASM statements). The code generator has been rebuild but generates bugs in exception handling in debug builds (Not the C++ exception is the bug, but the exception handling code itself generates another OS exception) Im shure its not a bug in my exception handling code (does nothing at the moment only beeing catched to not bubble up in callers). This has been working OK without PP, so I blame the PP for it. Compiling on a PC without PP results in a clean run. I am a signature virus! Help me spread and copy me to your sig! Ooops I am infected
Ugh! Trust me. Would installing service pack 5 over the top remove it? Simon "Does the room have a CView?" Sonork ID 100.10024
-
Open Magazine has a story at http://www.open-mag.com/754088105111.htm reporting that "Intel's C++ Compilers for Linux and Windows smoke GNU C and MS Visual C++ in number crunching benchmarks." For real-world numerically intensive computations, they find that Intel's C++ compiler is 37% faster than VC++ 6 and 25% faster than VC++ 7 beta. Linux bashers should note that Intel is over 47% faster than Gnu C++ and that code compiled with the Intel compiler runs about 10% faster under Windows XP than under Linux. They report that they have noted for several years that VC++ under Windows is significantly better than GCC under Linux but had not had a way to apportion blame between the compiler and the operating system. Now they find that even using the same compiler, Linux just can't keep up with Windows for high-performance scientific computing! This is particularly interesting given that Linux is particularly favored over Microsoft in the science and engineering communities, where it is widely believed that it is faster than Windows for numerically intensive computations. To me (a true VC++ zealot, if a cranky one), the fact that the Intel compiler blows the doors off VC++ is great news. Microsoft needs serious competition to keep it from getting fat and lazy, and it sounds as though Intel is going to provide a much-needed thorn in the side of MS's VC++ developers to get them off their rear ends as far as attention to numeric performance goes. (How lazy are they? I reported bugs with their standard C++ library several years ago (e.g.,
std::numeric_limits<double>::signalling_NaN()returns infinity, not a NaN) and these bugs are not even addressed in the Knowledge Base, much less fixed in any of the service packs).Back in the mid 90's during a hardware class in college, we watched a video put out by Intel that covered compiler design and optimizations with their processors. Even then, they made the same claims against other compilers. If you are the one who designs the processors, I would surely hope your compiler would create the quickest code .
-
It is probably designed to make specific usage of the P3 and P4 instruction set... The VC++ 6 compiler is designed to work the same on ALL processors... There is a processor pack available for VC++ 6, anyone tried this one ??? -= Forgive my English =-
The processor pack only lets you write assembly language for specialized Pentium (and AMD) processors. It doesn't offer extended code-generation and optimization tools.
-
Stay away form it. It can't be deinstalled, the code generation is not selectable to P3 or P4 (only in ASM statements). The code generator has been rebuild but generates bugs in exception handling in debug builds (Not the C++ exception is the bug, but the exception handling code itself generates another OS exception) Im shure its not a bug in my exception handling code (does nothing at the moment only beeing catched to not bubble up in callers). This has been working OK without PP, so I blame the PP for it. Compiling on a PC without PP results in a clean run. I am a signature virus! Help me spread and copy me to your sig! Ooops I am infected
Yes. I agree stay away from the processor pack. It caused me stack corruption after an exception occured in ADO code. Also there is no uninstall so I had to re-install VC++ on several PCs.