An embedded and charset-unspecified text was scrubbed... Name: not available Url: https://stat.ethz.ch/pipermail/r-sig-mac/attachments/20050617/f68d27db/attachment.pl
64bit Tiger performance vs 32 bit Wintel vs 32 bit Linux
3 messages · Shea, Ed, Simon Urbanek
1 day later
On Jun 17, 2005, at 3:16 PM, Shea, Ed wrote:
Has anyone experienced any performance improvements running with a 64bit tiger compiled R.2.* vs the windows and/or Linux versions (on comparable hardware)?
By definition using 64-bit addressing instead of 32-bit is slower, simply because all pointers are twice the size and hence twice the memory is moved/accessed when dealing with pointers. There is no penalty for running 32-bit programs on ppc64 therefore 64-bit programs are usually slightly slower than the same 32-bit programs and my tests with R confirm this. It's hard to make any cross-platform claims, because "comparable hardware" is a very vague term. If you tell me what you're interested in I could run some tests on G5 and amd64 ...
Aside from access to more memory, are there any other benefits?
AFAIK, no. Cheers, Simon
On Jun 18, 2005, at 3:31 PM, Simon Urbanek wrote:
Aside from access to more memory, are there any other benefits?
AFAIK, no.
I should add that I meant that there are no other benefits comparing 32-bit R and 64-bit R on Tiger. Of course comparing 32-bit R on ppc Tiger with x86 Windows or Linux is a different story with many differences ranging from the CPU architecture, memory management to the optimization and available BLAS libraries ... (of course you could expand this and compare Linux on a G5 etc.). Cheers, Simon