"Brian J Beesley" <[EMAIL PROTECTED]> wrote:
> I'd be VERY interested to hear of anyone running Prime95 on a
> Xeon - the Xeon has 512K, 1M or 2M of L2 cache running at core
> speed (at painful, extortionate and positively obscene prices,
> respectively) - benchmark comparisons with PII systems at the same processor/bus
>speed would be
> interesting!
[SNIP]
> On Win 9x and NT, you seem to get the best performance by
> loading the system right up so that the memory is exhausted, then
> starting Prime95, then stopping the unneccessary tasks. I think
> this forces the system to load the important bits of Prime95 into
> contiguous physical memory, thereby improving the performance of
> the cache. The performance improvement is around 2% e.g.
> iteration benchmark for 320K FFT (exponent ~6M) on my PII-333
> improves from 0.251/0.252 to 0.246/0.247 as a consequence of
> loading this way. (I wrote a dedicated swap thrashing program to
> load the system up). I find you usually need to reload Prime95
> when starting a new assignment in order to keep the performance
> as high as possible.
>
> Regards
> Brian Beesley
First off, for Mr. Beesley and any others who may be interested, I am running Prime95
on a single
processor 400Mhz Xeon with 512K of 400Mhz L2 cache and 256MB RAM, and with exponents at
approximately 6 million, I typically get 0.138 seconds per iteration, using Windows NT
4.00
According to http://www2.tripnet.se/~nlg//benchmk.htm, a standard PII/400 with 64MB
RAM running
Windows 95 gets 0.184 seconds per iteration, for exponents in the rough range of 6
million. I would
be particularly curious of the performance of Prime95 on a PII/400 under Windows NT
4.00 with more
RAM, or on any PII/450 or 450 Mhz Xeon(s).
I don't know if this is at all meaningful, but when I was originally setting up
Prime95 on my
computer, and was asked to enter the type of processor by radio buttons and the Mhz
speed in the
numerical field, the computer had already chosen 562 Mhz. Because of this
non-standard number I
intuit that this is not a default, but rather that the software was trying to make a
guess at my
processor's speed, and assuming basic Pentium II or Pentium Pro architecture thought
that it runs at
562 Mhz. I wish to defer any conclusion from this to someone who has actual knowledge
about this
feature of Prime95, as the best I can do is make a guess. If it does try to estimate
the computer
speed, I also do not know whether it thinks my Dell Workstation 610 to be a 562Mhz
PPro or PII. At
the very least, I would be interested to hear whether any other users have experienced
a close
approximation of their processor speeds by Prime95.
As I am composing this message I decided to quit Prime 95, have Mathematica try to
factor the
Mersenne number I am currently working on, and restart Prime95. =) And indeed there
has been an
improvement in performance, to 0.136 seconds per iteration consistently. When I have
some more time
I'll play around with this theory, and by a much better method than the quick idea
described above.
I would be most interested in receiving Mr. Beesley's swap thrashing program, as it
would be a much
easier and surely more efficient and complete method of thus increasing performance!
Regards,
--Samuel Lipoff
