|
ScienceMark
2 :::...
Science Mark is a favourite with us and I'm not sure why
more people don't use it. Our experience has been that it's
fast and accurate and gives repeatable results.
BlasBench:::...
If
you care, this is what they say about Blas:
Scientists
and engineers use large matrix multiplication to determine
peak MFLOPS. BlasBench measures the MFLOPS of your machine
via large matrix multiplication. The benchmark can be run
in single or double precision. Real world matrices are often
large and don't fit into the processor's cache. Science
Mark 2’s optimized implementations of matrix multiplication
in single and double precision, from an algorithm point
of view, perform remarkably better than the blind man's
matrix multiplication routine. BlasBench uses building blocks
called kernels at its lowest level that determine overall
MFLOPS performance. These kernels are written in C and a
wide assortment of assembly forms (SSE Vector, SSE2 Vector,
SSE2 Scalar, x87, 3DNow! Vector) to support any processor
architecture in the x86 market.
3dNow!......
Running
3dNow! highlight's the FX55's superiority with all matrix
configurations other than the 64x64 matrix where falls significantly
behind the 4000+.
4000+
Results
FX55
Results
SSE....
In
this case we see a massive performance advantage from the
FX55 using a 256x256 matrix size, but for some reason the
FX55 is extremely weak again working with the 64x64 matrix
with just over half the performance of the 4000+. This can't
be cache related due to the size and is certainly a curious
result.
4000+
Results
FX55
Results
In
light of the fact that these are essentially the same processor
with the same ammount of L2 cache and the same memory controller,
I find it odd that they handle these calculations so differently.
This is something we really need to look further into.
Scipher
Benchmark:::...
Cipher
is part of ScienceMark 2's benchmark suite. There's nothing
between the two sets of results in Scipher that can't be
explained by the difference in frequencies.
4000+
Results
- Lower is Better
FX55
Results
- Lower is Better
Primordia:::...
This
code calculates the Quantum Mechanical Hartree-Fock Orbitals
for each electron in any element of the periodic table.
A Self-Conistent loop is performed. At each step in the
loop the hartree, exchange, and the correlation potentials
for each orbital are evaluated. Got that? Good, now could
you explain it to me please?
The
FX55 pulls off a 24 second lead in calculation time once
again demostrating its prowess. 24 seconds may not sound
like much but remember thats saved from a total calculation
time of just 350 seconds. My my reckoning, if that calculation
was a little more complex and had to run for 24 hours, that
would translate to a saving of about 99 minutes. Stick that
CPU in a number crunching box running 24-7 and you have
a 600 hour advantage in the year!
4000+
Results
- Lower is Better
FX55
Results
- Lower is Better
|
