To gather the results we used two dual Pentium II, 450 MHz machines (Asus P2B-D boards) connected via fast ethernet (3COM 3C905B-TX cards) and a 100 MBit/s switch.
The tested versions of the different implementations are:
 |
NT-MPICH | : 1.2 |
|
MPICH.NT | : 1.2.1 (Feb. 16, 2001) |
|
WMPI | : 1.5 beta Unfortunately there seems to be a bug in WMPI 1.53 which made it impossible to run the benchmark with this version. |
|
MPI/Pro | : 1.6 |
You might also want to take a look at our main performance page.
 |
| The chart above shows the results of the PMB PingPong benchmark. Larger values are better. |
 |
| The chart above shows the results of the PMB PingPing benchmark. Larger values are better. |
 |
| The chart above shows the results of the PMB Sendrecv benchmark. Larger values are better. |
 |
| The chart above shows the results of the PMB Exchange benchmark. Larger values are better. |
 |
| The chart above shows the results of the PMB Allreduce benchmark. Smaller values are better. |
 |
| The chart above shows the results of the PMB Reduce benchmark. Smaller values are better. |
 |
| The chart above shows the results of the PMB Reducescatter benchmark. Smaller values are better. |
 |
| The chart above shows the results of the PMB Allgather benchmark. Smaller values are better. |
 |
| The chart above shows the results of the PMB Allgatherv benchmark. Smaller values are better. |
 |
| The chart above shows the results of the PMB Alltoall benchmark. Smaller values are better. |
 |
| The chart above shows the results of the PMB Bcast benchmark. Smaller values are better. |