sullivan@msor.exeter.ac.uk (Rob Sullivan) (07/13/89)
My department (theoretical solid state physics) is interested in buying a new
machine in the near future. It would be very useful if someone could provide a
reliable set of benchmarks for the DN10000, which we would expect to use as a
three processor system. Benchmarks for the Stellar DS2500 and DS2000 would also
be extremely valuable.
Thanks in advance,
Rob...hidinger@peanuts.nosc.mil (Ron Hidinger) (07/17/89)
I don't know how reliable these benchmarks are but here are some
results obtained with a single cpu loaner from the local office.
Not sure how much memory it had, but swapping was not a factor.
Time was measured using /bin/time, except the graphics benchmark.
Ron Hidinger
hidinger@nosc.mil
--------------------------------------------------------------
The results are in seconds. The numerical integration involves
single and double precision floating point for the most part.
The FFT's are single precision. The beam pattern involves
a 2 dimensional FFT and false color intensity plot, about 5000
trapezoids. This was a single processor machine with the
extra graphics hardware. The pascal compiler was used which
according to the salesman is already one revision behind.
HOST solve 6 DOF PDE FFT's 2D beam pattern
------- --------------- ----- ---------------
DN300 --
DN320 105.8
DN330 59.0
DN460 44.1
DN3000 60.6
DN3000 (SR9.6) 53.5
DN4000 25.9
DN3500 18.6 26.5 ~16
DN3550 18.9
DN4500 13.2
DN10000 1.3 2.9 ~4
turner@dover.sps.mot.com (Robert Turner) (07/17/89)
In article <SULLIVAN.89Jul13161740@msor0.msor.exeter.ac.uk>, sullivan@msor.exeter.ac.uk (Rob Sullivan) writes: > reliable set of benchmarks for the DN10000, First there is no such animal as a "reliable set of benchmarks". But to answer your question, we have seen from 4 to 15 times faster wall clock times than a DN4000. Our DN10k is a multi-processor, four CPUs. When running three jobs in parallel each job ran less than 10% longer on wall clock time. Note all programs and all data resided in the single DN10K's disk. Your real question should be what will be the easiest machine to use in our Apollo environment. The answer hands down is a DN10K. For us, it was a plug and play, and on the ring in a couple of days (due to newness of SR10). Robert