#MP performnce
5 messages in this thread
As we plan for the future upgrade to NT…..
Does anyone have any benchmarks on what the rendering performance would be like
with multiple pentium type motherboards?
When we upgraded from 486 DX2 to 486 DX4 we got almost the full theoretical 33
percent improvement in rendering (we actually got 31 percent). Which leads me
to believe that dual pentiums under NT should do welll but there isn't any data
I can find.
Being an ex-Intel employee I am pretty sure Intel doesn't have any numbers
specific to rendering. We are pretty much a niche market to the desktop guys.
It would be nice if Adesk could run some stuff and be able to advise us.
Thanks,
Steve
Steve,
Why wouldn't a dual Pentium system be twice as fast a single Pentium system?
<< It would be nice if Adesk could run some stuff and be able to advise us. >>
Autodesk will not release benchmarks to the public.
Mark
>>Why wouldn't a dual Pentium system be twice as fast a single >>Pentium
system?
PMJI: a dual Pentium system (I use a ASUS PCI/EISA Dual Pentium/100 board)
can`t be as twice as fast because they use the same 2nd level cache, the same
bus (adress and data path) and the same memory.
The chipset on the board has to sync both prozessors, the 2nd level cache has
to support two processors and the bus has only a limited number of lines(I
don`t know a better word). These are only some things that slow down the system
a little bit. Depending on the chipset and the Operation System you`ll lose
some performance…
Michael
Michael,
<< PMJI: a dual Pentium system (I use a ASUS PCI/EISA Dual Pentium/100 board)
can`t be as twice as fast because they use the same 2nd level cache, the same
bus (adress and data path) and the same memory. >>
Thanks for the clarification.
Mark
The reason is they share the same bus. If all is a perfect situation with a
shared bus you would get 1.5x performance improvment from second Pentium in a 2
Pentium system.
The secondary cache is also an issue, but Intel has suggested for higher
performance systems to have one cache for each processor. (As opposed to the
lower cost 2 Pentium, 1 cache approach.) I'm not sure if anyone has taken this
approach.
If you look into the future a bit to MP Pentium systems, P6 systems, etc. the
real performance platforms will be x2 and x4 architectures(multiply
processors). I think that since the performance is a very application dependent
issue, it is appropriate for Adesk to provide some numbers.
If benchmarks aren't available, the people who build the higher performance
(and higher cost) boxes won't be rewarded and we will all have to live with
mediocrity.
Steve.