Reference Guide

Altair Radioss Performance
18 Dell EMC Ready Solutions for HPC Digital Manufacturing with AMD EPYC™ Processors—Altair Performance
These benchmarks were carried out on a cluster of eight servers, each with dual 7452 processors. The
results are presented in relative performance compared with the single node results. The parallel speedup for
the large Taurus model is nearly linear up to eight nodes (512 cores). However, the parallel speedup for the
smaller neon model starts to fall off above two nodes (128 cores). It should be noted that the Neon
benchmark model is very small by today’s standards. The time required to carry out the full 80msec
simulation on a single node is only 1338 seconds. As such, there is no expectation to see a good parallel
speedup with this model at four or more nodes.
Like AcuSolve, it is also possible to use Radioss in hybrid parallel mode. Figure 9 shows the parallel
performance for both Radioss benchmarks models using (1,2,4) shared memory threads on up to 8 compute
nodes.
1.0
2.0
4.0
8.0
64(1) 128(2) 256(4) 512(8)
Performance Relative to sinlge node
Number of Cores (Number of Nodes)
Figure 8: Radioss Parallel Performance
Neon
Taurus
1.0
2.0
4.0
8.0
64(1) 128(2) 256(4) 512(8)
Performance Relative to 48 Cores
Number of Cores (Number of Nodes)
Figure 9: RADIOSS Hybrid Parallel Scaling on EBB
N-1 N-2 N-4
T-1 T-2