[linux-audio-user] Re: 64bit vs 32bit (was: Re: AMD64
nando at ccrma.Stanford.EDU
Wed Jul 12 17:40:42 EDT 2006
On Wed, 2006-07-12 at 18:20 +0200, Arnold Krille wrote:
> 2006/7/12, Roberto Gordo Saez <roberto.gordo at gmail.com>:
> > I've made tests mainly on big server applications, I have not made
> > tests on audio software, so I suggest everyone to dedicate some time
> > testing your own apps. It is simple: make two partitions, one of them
> > with a 32bit OS, and the other optimized for 64 bit. Do not use a
> > chroot nor mixed environment (32bit libs on 64bit system), because the
> > results can be slightly different.
> Okay, I don't have space (and time) for a second partition. And while
> I understand that my testing might not be the optimum test, I think
> your's isn't either: Because you compare fully using a 64bit system
> with using a 64bit system in 32bit-mode. It is like comparing a big
> motorcycle with the same motorcycle but slowed down. For sure the
> slowed-down version will not be as fast. Also german IT-newspaper c't
> has made some comparisons when opteron64 and intel xeons with 64bit
> went onto the market Their result was that the amd64 are faster than
> the intel64. And surprisingly enough the amd64 in 32-bit mode where
> almost as fast as in 64-bit mode while the intels where significantly
> slower in 32-bit mode than in 64-bit mode...
> I am thinking about some other test, as someone mentioned registers
> and memory and stuff, which I think are good to recognise, but for the
> "real" computing speed maybe a mathematical test would be better? A
> test that wouldn't need lots of memory and lots of registers and
> doesn't have lots of cache-misses?
> Like computing pi to a certain number of relevant numbers?
> And, yes, I am comparing apples with oranges. But that was the
> question at least here in the lab: Comparing single-turion64 with
> double Xeon32. And guess what: On rather simple code, without special
> SSE/MMX/etc optimisation, with the same compilerflags, the turion was
> significantly faster.
So, a single turion64 is faster than and double xoen32. That's all that
the test says. It does not say "64 bit is faster than 32 bit".
To answer that question to an approximation you would have to install
the 32 bit version of the exact same distro that was running in the
turion64 in the turion64 machine. Then run the test again.
> So I would like to prove this observation by a good test. But
> comparing a retarded 64bit system with a full 64bit system isn't the
> solution. (Who would for example shut down half of the heating of his
> toaster anyway?)
If the performance advantage of 64bit mode over 32bit mode (in the same
processor / distro version) is not big some users might prefer to stay
in 32 bit mode simply because there is stuff that will not (yet) run in
64 bit mode.
More information about the linux-audio-user