Re: Re: Re: mathematica 64-bit for ms windows

*To*: mathgroup at smc.vnet.net*Subject*: [mg85674] Re: [mg85576] Re: Re: mathematica 64-bit for ms windows*From*: John Fultz <jfultz at wolfram.com>*Date*: Tue, 19 Feb 2008 01:52:55 -0500 (EST)*Reply-to*: jfultz at wolfram.com

On Thu, 14 Feb 2008 01:00:12 -0500 (EST), J=E1nos wrote: > > > On Feb 13, 2008, at 4:04 AM, David Park wrote: > >> But be aware that if you are buying a new computer, from Dell at >> least, and >> you specify a 64 bit microprocessor, you will not necessarily get a >> 64 bit >> operating system, and may not even be able to install a 64-bit >> operating >> system. So if you are looking to use 64-bit Mathematica check out very >> carefully before purchase that you will indeed have a 64-bit operating >> system. >> >> As I understand it, the main advantage of a 64-bit operating system >> is that >> you can address more memory, beyond 2GB, but that it is not >> significantly >> faster - and may even be slower. (Others may correct me or expand >> on this.) >> >> -- >> David Park >> djmpark at comcast.net >> http://home.comcast.net/~djmpark/ >> >> >> "peter" <glassymeow at yahoo.com> wrote in message >> news:fopb36$bfq$1 at smc.vnet.net... >>> hi >>> while browsing about the mathematica products i have noticed that >>> there >>> is no something stated lucidly like: >>> mathematica 64-bit for windows >>> while there is such a statement (64-bit) regarding mathematica >>> for Linux >>> does that mean that there is a one package for windows in which the >>> installer will detect whether the windows OS is 32 bit or 64 bit >>> and will >>> install accordingly ?? >>> peter >>> > > My understanding is if the machine is truly 64-bit and the OS is 64- > bit in all aspect, then operation at the same Hertz should be twice > as fast, because of the bus. Unfortunately there are 64-bit machine > that are not really 64-bit all the way, and there are OSs that also > not 64-bit and sometimes not even 32-bit. > > J=E1nos > > > -------------------------------------------------------- > The more money has to be spent to fix a problem, the more likely it > is that the people with the money will agree to fix it. > /Alex Satrapa/ Your comparison is not apples-to-apples. Pointers also become 64-bit data types. Typical programs push a lot of pointers around, and their doubled size can contribute significantly to cache misses (on equivalently sized caches), which can be a big performance drain. And a doubled bus width does not, by itself, do anything to deal with the cost of failed branch prediction, which will be equally horrible on both types of CPUs. There are, to be certain, some advantages for 64-bit CPUs in performance, but with enough associated drawbacks that it's better to do actual benchmarking thanto make any kind of a priori assumptions if performance is your only measure. And, as you say, there may be yet more mitigating factors which depend upon the specific of hardware and OS choice (although I wouldn't suggest any of the modern OS's are anything less than fully 32-bit). Of course, if you need to map more memory than your 32-bit OS/CPU will allow, then the choice is obvious. Sincerely, John Fultz jfultz at wolfram.com User Interface Group Wolfram Research, Inc.