ATHLON vs XEON: number crunching

Ivan Oleynik oleynik at chuma.cas.usf.edu
Wed Jun 19 18:26:57 PDT 2002


Bill,

As in my first query, I did not try to say that the CPU clock is something
I want to compare. I just picked up current state-of-the-art Intel and
Athlon processors and tried to asnwer the question which one is faster for
my particular application. I got the code compiled with the same options
on all the machines I have an access and run the code. I used the exection
time of the code run on 4 year old Origin 2k 195 Mhz (this is the CPU unit
of old SGI Origin supercomputer in Oxford that has been just dismantled)
as 1 unit. Then I got the following:

1. Origin 2000, 195 MHz          ::: 1
2. HPUX 9000/785                 ::: 2  (1 year old)
3. P-III 1.3 GHz, 512k SC        ::: 3
4. Athlons (MP 1200, 1900, 2100) ::: 3
4. Xeon 2.2 GHz, 512k SC         ::: 4.5

My experience with this code tells that it samples a good portion of real
numerical stuff such as FFT, blas, lapack including the most intensive
eigenvalue-eigenvector type of problems. I always talk to my colleagues
that use these machines and my estimates are pretty much the same as
their, Of course, we all do not do any graphics processing or data mining.

I tried to find rationale in SPEC results, but sometimes it is very
difficult since most of benchmarks were done under Windows not Linux.

Thus, I am not trying to argue with the statement that CPU clock is not
the only factor responsible effective number crunching. And I absolutely
agree with you that everyone has to find a rationale based on his/her own
needs.

Ivan
         
------------------------------------------------------------------------

On Wed, 19 Jun 2002, Bill Broadley wrote:

> > What I found is that Xeon 2.2 GHz is 1.5 times faster! than any athlons I
> > tested. But the most strange thing is all these athlons: MP 1200, MP 1900,
> > MP 2100 give approximately the same timing within 5%. This is completely
> > above my comprehension.
> 
> A few benchmarking rules/tips:
> 
> #1 You can't use Mhz to compare across architectures.
> #2 Within architectures often performance does not scale linearly with clock
>    speed.
> #3 Any comparable cpu's can be shown to be faster OR slower then another 
>    if you pick the right benchmark.
> #4 Because a given benchmark shows X is faster than Y, that does NOT
>    mean X will be faster then Y for your application.
> #5 Bottlenecks outside the cpu often do not scale AT ALL with clock speed.
>    So if it's memory bandwidth a 2.0 Ghz can be 0% faster then a 1.5 Ghz
>    cpu.
> 
> The golden rule of benchmarking:
> #1 Use the application that justifies the purchase of the machine to
>    compare price/performance.  Only then can you be assured of getting
>    the most performance for your money.
> 
> > Did someone encounter such a strange pattern and what can be a source of
> > this behavior?
> 
> For things with certain memory access patterns P4's enjoy a significant 
> advantage in memory bandwidth.  On the other hand the athlon enjoys
> a performance lead on many other types of scientific codes.
> 
> Take specfp2000, a collection of 14 scientific applications (NOT 
> microbenchmarks).  The 2.0 Ghz P4 gets 669, the MP2000 gets 642.  So
> they are very similar right?  Nope, the reality is that at some parts
> of the specfp2000 suite the p4 is 1.64 times faster then the Athlon.
> At other benchmarks the Athlon is 1.56 times faster then the P4.
> 
> Which of the 14 parts of the speccpu 2000 suite is your application?
> Hard to see, see the golden rule #1.
> 
> -- 
> Bill Broadley
> Mathematics/Institute of Theoretical Dynamics
> UC Davis
> 
------------------------------------------------------------------------
Ivan I. Oleynik                       E-mail : oleynik at chuma.cas.usf.edu
Department of Physics
University of South Florida
4202 East Fowler Avenue                  Tel : (813) 974-8186
Tampa, Florida 33620-5700                Fax : (813) 974-5813
------------------------------------------------------------------------




More information about the Beowulf mailing list