Very Computer

In terms of speed, all other things being equal, would a dual cpu system
running two 400mhz processors be equivilent to a single 800mhz
processor? I haven't a clue how this works.

If you had to ask this question, then you should go with single CPU.
It's easier and cheaper. :-)
--

8-cpu Cluster, Hosting, NAS, Linux, LaTeX, Python, vim, mutt, tin

Quote:> In terms of speed, all other things being equal, would a dual cpu system
> running two 400mhz processors be equivilent to a single 800mhz
> processor? I haven't a clue how this works.

If you're not sure, then the answer is a clear "no," which should go
along with "you REALLY don't want to go there."

In the Intel world, the answer is almost certainly "no," and the
notion of "all other things being equal" is exactly the fallacy that
you'd be liable to fall into.  

"All other things being equal" is an invalid assumption.

On hardware designed for highly scalable SMP, the answer could be yes;
that tends only to happen on systems costing on the order of hundreds
of thousands of dollars, using MIPS, SPARC or Alpha processors.

If you have any practical need for an answer to this, then you
presumably should be contacting vendors who will be quite happy to
have you submit some of your FORTRAN code to do test runs on their
hardware.  If things turn out well, you'll be spending a million
dollars.
--

http://www.cbbrowne.com/info/linux.html
Rules of  the Evil Overlord  #225. "I will  explain to my  guards that
most people have their eyes in the front of their heads and thus while
searching  for someone  it makes  little sense  to draw  a  weapon and
slowly back down the hallway." <http://www.eviloverlord.com/>

Quote:> In terms of speed, all other things being equal, would a dual cpu system
> running two 400mhz processors be equivilent to a single 800mhz
> processor? I haven't a clue how this works.

The main problem is that almost all software is not optimized to take
advantage of more than one CPU. As a result, the single CPU configuration is
going to be faster.

Therefore one 800 MHz processor is always faster than two 400 MHz processors
(unless there is a lot of preempting).
--
  8:23pm  up 23 days, 21:51,  3 users,  load average: 0.67, 0.63, 0.57

A computer splits tasks into processes, which are further split into threads.  
A typical user only usually uses one process (or thread) at a time, actively
(non-idle processes).  I do believe that if you were compiling two large
programs in sequence on a 800 MHz processor, it would take the same amount of
time to run the two compilations at the same time (two processes) on two 400
MHz processors.

Someone correct me if I am wrong about anything (I am not an OS developer).

Brian
--
  8:16pm  up 23 days, 21:44,  3 users,  load average: 0.52, 0.51, 0.54

That is assuming that 800 Mhz processor is exactly twice as fast as a
400 Mhz processor. If I remember correctly this is not always the case :-)

The first invalid assumption is that an 800MHz CPU is exactly twice as
fast at everything as is the 400MHz CPU.  Things aren't normally that
simple.

The second problem: Memory usage.  The two 400MHz CPUs are sharing a
memory bus.  If they can share it really nicely, this may be a Very
Good Thing.  If they're fighting for memory accesses, then Very Bad
Thing...

Cache accesses are similarly important.  If there's any shared cache
(not uncommon on IA-32), then if they're sharing gracefully, they may
get a lot better than a 2x speeedup, but if they're both fighting to
have their way, uh, uh...

High end multiprocessor systems tend towards "NUMA" (non uniform
memory access), which lets the CPUs be more independent, which is
usually a win.  You won't get that on a cheap SMP Intel box.

In practice, cheap SMP means you're using a knock-off of last year's
technology, which means you've lost before you've started.  The dual
400MHz machine probably has slower memory, slower disk drive, and
other strikes against it.

Unless it's a SPARC, and the 800MHz machine is an Intel, in which case
comparability leapt out the window...
--

http://www.cbbrowne.com/info/wp.html
"...I'm not one  of those who think Bill Gates is  the devil. I simply
suspect that if Microsoft ever met up with the devil, it wouldn't need
an interpreter." -- Nicholas Petreley, InfoWorld, Sept 16, 1996

No.  A near-ideal program would be about 80% as fast.

--

"Last I checked, it wasn't the power cord for the Clue Generator that
was sticking up your ass." - John Novak, rasfwrj

Hey, ya gotta start somewhere... ;)

Quote:> If you have any practical need for an answer to this, then you
> presumably should be contacting vendors who will be quite happy to
> have you submit some of your FORTRAN code to do test runs on their
> hardware.  If things turn out well, you'll be spending a million
> dollars.

It's not a practical need, just a curiosity that I couldn't find a good
answer for.... but now I have a couple, so thanks. And I think I'll save
that million for now.

| In terms of speed, all other things being equal, would a dual cpu system
| running two 400mhz processors be equivilent to a single 800mhz
| processor? I haven't a clue how this works.

On teh other hand, I just bought a dual Athlon motherboard to upgrade my
home system. I got the MB, 2 Athlon MP1800+ CPUs (~1.6GHz each) and
512Mb of RAM for about what a 2.?GHz P4 would have cost.

If you pick the CPU's carefully, you can get fast enough CPU's on the
dual MB to more than offset the normal efficiency penalty, AND save
money.

--
Reverend Paul Colquhoun, ULC.    http://andor.dropbear.id.au/~paulcol
     Asking for technical help in newsgroups?  Read this first:
        http://www.tuxedo.org/~esr/faqs/smart-questions.html

In real life, I've seen about a 40% increase in speed using P5, PII, and
PIII CPUs.  Basically, dual CPUs aren't always used to 100% efficiency
because of all the reasons mentioned.

OTOH, dual CPUs allow one CPU to handle one bunch of tasks, while you
work on the other CPU.  not quite how it works, but it lets me run a HUGE
compile or another intensive app, and still have an *apparently* unloaded
machine.

So from a usability standpoint, I like dual CPUs since the machine
*feels* faster than a comparable faster CPU when loaded.  The fact is
that the intensive task would finish sooner on a faster single CPU
machine, but the machine would *feel* loaded - sluggish response at the
keyboard, etc.

The other thing that people don't mention is that dual CPUs allow the OS
to run on one CPU and the video to run on the other (again, a gross
misstatement, but accurate for the level here).  So for example, one CPU
is handling the software RAID, disk caching, syslog, etc., while the
other is responding to you.

I like dual CPUs for servers, since a runaway process will typically suck
down a single CPU, allowing the server to function while I frantically
track down the problem....

It really boils down to how you use the machine and what you expect, and
personal preference.

--Yan

No, it won't be.  And which direction it goes depends a _lot_upon
other information:  what kinds of memory, what kinds of work you
are doing.

First of all, an 800 MHz processor is *not* twice as fast as a 400 MHz.
The processor *core* is twice as fast, but the cache is not.  Nor is the
memory.  So except for in-core, on-processor-core programs (few and far
between) the 800 will be _slower_ than twice the 400.

For single-program workloads that are not parallelized, you can't take
advantage of the second processor anyway, so in that case the fast single
processor system will be significantly faster.

For those programs that have been parallelized, it depends upon how they
fit into processor cache (duplicated resource) and main memory (shared
resource), and how efficiently they have been parallelized.  For cache
resident well-parallelized programs, the dual 400 is faster.

For heavy multi-program workloads (e.g., software development with lots
of "make"s), the processors act independently, and the dual-400 system
will be significantly faster.

fwiw.

MCNC-Environmental Modeling Center               phone: (919)248-9241
North Carolina Supercomputing Center               fax: (919)248-9245
3021 Cornwallis Road                                  P. O. Box 12889
Research Triangle Park, N. C.  27709-2889                         USA
"My opinions are my own, and I've got *lots* of them!"

I agree with this.

I recent had purchased a dual system and have been very happy with it.
Of course it is not as fast a doubling the clock freaqueny but then
again with memory bandwith limitations maybe thats not such a good idea
anyway. The main advantage I see is in response time and stablity. I do
at times compile large jobs and do some image processing and having the
extra processor still allows pretty good response time. In addition to
response time I have noticed too, that a runaway process does not lockup
the system, just one CPU. I can always go in a kill the job with no
apparent preformance lose.

In the end its really about experimentation. I would never recommend for
someone ( a generic home user ) to get one since the advantages are
pretty specific but there are some noticable difference. At least I
think so. The smp support in linux is really pretty good. You would
think with good process control the above advantages would not be
present but we do not have prefect process control. No os does that I
know off.