12 Jul
2006
12 Jul
'06
11:04 a.m.
On 7/11/06, Arnold Krille <arnold.krille(a)gmail.com> wrote:
Well, I _know_ that 64bit is faster: Here at work I
have dual-Xeon's
with 2.8GHz, my laptop is a Turion64 with 1.8GHz. The same computation
(evaluating scientific data from time-of-flight/lifetime experiments)
on the same data-files is significant slower on the work-pc's (almost
factor 2). As my threads are one thread for computation and others for
loading and saving the data, it is really a comparison of two single
processors, where the faster one does even the IO-work itself...
Personally, I would like to suggest everyone to actually make some
serious tests on 64bit hardware with some intensive sound applications
(or other programs that are main tasks for the computer). Latter, you
can decide which one you want for everyday use.
Surprisingly for myself, I've found that 64bit are NOT always much
faster, and sometimes are actually slower. It is true that most
applications may improve performance when running on 64bit. But
exceptions does exist, mainly because the smaller data chunks on code
with 4 byte pointers (32bit), which improves cache usage, and also
because there are still lots of applications heavily optimized for
32bit x86 code, whose performance is degraded when running on 64bit.
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.
There is one case in that 64bit is always recommended: if you want to
use big resource hungry applications that handle more than 3GB of RAM
per process. Linux does support more RAM even on 32bit machines (using
PAE extension), but the ugly limit on memory per process on 32bit mode
is still there. On 64bit mode processes can access up to 1TB. And yes,
they actually exists. I've seen some applications easy allocating more
than 50GB of RAM per process! But not on my main desktop computer, it
is still a Pentium II :-)
I'm interested on this, so if anyone has made some real performance
tests with commonly used audio or other multimedia apps on 32bit and
64bit, it would be great to know results.
12 Jul
12 Jul
12:20 p.m.
New subject: [linux-audio-user] Re: 64bit vs 32bit (was: Re: AMD64 question: update)
2006/7/12, Roberto Gordo Saez <roberto.gordo(a)gmail.com>om>:
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 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?)
Arnold
--
visit http://dillenburg.dyndns.org/~arnold/
---
Wenn man mit Raubkopien Bands wie Brosis oder Britney Spears wirklich
verhindern könnte, würde ich mir noch heute einen Stapel Brenner und
einen Sack Rohlinge kaufen.
1:03 p.m.
New subject: [linux-audio-user] Re: 64bit vs 32bit (was: Re: AMD64 question: update)
you get twice as many registers, and some, but not all data is twice as large. the end
result seems to be about 20-30% faster than running in 32bit mode (basing this on
osc-count in synths + genlop -t compiletimes). certain things do feel slower, namely
firefox and KDE, which could be due all the pointers.
the big showstopper is all the coolest audio apps ( i mean the ones that arent just
mediocre kde/qt knockoffs of what MacOS had in 1993 ) are not 64bit clean. someone should
fedex miller puckette and james mccartney and ge wang linux-loaded amd64 boxen and demand
a fix!
right now i'd go Turion X2. if you want all 3 of: mobile, 64bit, dual-core, only AMD
has an answer for you..
1:37 p.m.
New subject: [linux-audio-user] Re: 64bit vs 32bit (was: Re: AMD64 question: update)
On 7/12/06, Arnold Krille <arnold.krille(a)gmail.com> wrote:
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...
Hmm... I don't understand your point here. FYI: My message was to say
the motorcycle but slowed down is actually faster some times. I am not
native English speaker, so maybe it was not clear.
The 64 bit extensions are documented on White Paper on AMD site; I
think they are public, though I've obtained mine by other means (I've
read them because I work right now as a driver and hardware engineer,
at low level programming, in case you wonder). They are not to make
your processor faster. Faster computing can be archived as a
consequence, but it is not always the case. Even, I've read something
from AMD people saying that 32bit mode is faster (and even
recommended) for some tasks.
Instead of trying to guess things, you should make tests (this is a
mere suggestion, don't take it as a mandate or something...). I think
you will be surprised, as I was. I was also thinking that new 64bit
mode was always faster until did some tests. I've tested both AMD and
Intel.
That is the reason for being interested in hear some tests with audio
apps. Yes, I *think* that multimedia software should be faster on
64bit, but real tests would be appreciated to confirm. Merely saying
that it is faster because it should be does not make any sense to me,
since I've seen sometimes it is not true.
5:40 p.m.
New subject: [linux-audio-user] Re: 64bit vs 32bit (was: Re: AMD64 question: update)
On Wed, 2006-07-12 at 18:20 +0200, Arnold Krille wrote:
2006/7/12, Roberto Gordo Saez
<roberto.gordo(a)gmail.com>om>:
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.
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 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.
-- Fernando
6:21 p.m.
New subject: [linux-audio-user] Re: 64bit vs 32bit (was: Re: AMD64 question: update)
> 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?)
heres a test i did. compile glibc 2.3.6 (nptlonly) on 2 setups.
athlon64/turion64, both single-core 2.0ghz connected via distcc over tcp over wifi. make
-j4
time: 22 minutes
macbook pro. 1.83ghz (32bit) dual-core. make -j3
time: 23 minutes
seems like theres no real difference. except of course that if you run 64bit, you need to
have your entire depchain down to glibc installed twice if you want to watch youtube or
use PD :)
13 Jul
13 Jul
7:01 a.m.
New subject: [linux-audio-user] Re: 64bit vs 32bit (was: Re: AMD64 question: update)
Just a short note, since I notice that i don't really know much about
processor-internals to make deeper comments on "64bit vs 32bit"
anymore...
2006/7/13, carmen <_(a)whats-your.name>me>:
seems like theres no real difference. except of course
that if you run
64bit, you need to have your entire depchain down to glibc installed twice
if you want to watch youtube or use PD :)
PD runs in native 64bit mode just fine on my turion64 with gentoo.
And for every proprietary app, that needs 32bits, gentoo has the right
environment available and installs it via emerge without you doing
anything...
The last time an (audio) app didn't compile in 64bit was specimen
(back then when it was activly developed), I wrote a report to the
author and a day later it was running (basicly compilation stopped
because of -Werror and a warning about casting a pointer to an int of
different size).
I even marked some 32bit apps as stable for 64bit on my system.
So in my point the argument "some things don't (yet) run on 64bit"
doesn't count a cent. :-)
Arnold
--
visit http://dillenburg.dyndns.org/~arnold/
---
Wenn man mit Raubkopien Bands wie Brosis oder Britney Spears wirklich
verhindern könnte, würde ich mir noch heute einen Stapel Brenner und
einen Sack Rohlinge kaufen.
1:45 p.m.
New subject: [linux-audio-user] Re: 64bit vs 32bit (was: Re: AMD64 question: update)
PD runs in native 64bit mode just fine on my turion64
with gentoo.
you clearly didnt use it. besides launch it and say 'oh, it runs fine'.
i guess it is possible to use despite all the math errors and without the ability to store
and recall audio in RAM. but i dont consider that usable.
So in my point the argument "some things
don't (yet) run on 64bit"
doesn't count a cent. :-)
i gues PD, SC, and Chuck dont count a cent in your world. despite their 'free'
status. i place some value on them :)
2:08 p.m.
New subject: [linux-audio-user] Re: 64bit vs 32bit (was: Re: AMD64 question: update)
Hi,
carmen wrote:
i gues PD, SC, and Chuck dont count a cent in your
world. despite their 'free' status. i place some value on them :)
Are you sure SC doesn't run on 64bit yet?
The server in any case does, and I do believe Stefan Kersten has been
working on getting the lang to work too...
Not sure whether that came to an end, though...
sincerely,
Marije
4:16 p.m.
New subject: [linux-audio-user] Re: 64bit vs 32bit (was: Re: AMD64 question: update)
2006/7/13, carmen <_(a)whats-your.name>me>:
Well, I don't claim to be a second Frank Barknecht. I do just some
simple synthesis and midi-sequencing with it. And that runs fine so
far...
PD runs in
native 64bit mode just fine on my turion64 with gentoo.
you clearly didnt use it.
besides launch it and say 'oh, it runs fine'. i gues PD, SC, and Chuck dont count a cent in your
world. despite their
'free' status. i place some value on them :)
Hmm, PD counts more than a cent in my world. But in comparison to
ardour and all the other small tools to record and produce real
music(*), I don't need (and use) PD that much. But as far as I use it,
it works...
So long,
Arnold, quiting this thread, promised!
(*) Thats my very personal opinion, if you like and make electronic
music more than me, than don't get offended by my position.
--
visit http://dillenburg.dyndns.org/~arnold/
---
Wenn man mit Raubkopien Bands wie Brosis oder Britney Spears wirklich
verhindern könnte, würde ich mir noch heute einen Stapel Brenner und
einen Sack Rohlinge kaufen.
6529
days inactive
6530
days old
linux-audio-user@lists.linuxaudio.org
9 comments
5 participants
participants (5)
-
Arnold Krille -
carmen -
Fernando Lopez-Lezcano -
Marije Baalman -
Roberto Gordo Saez