This page summarizes AWS ParallelCluster benchmark results for Relion 4.0.
<Environment>
AWS ParallelCluster : ver3.0.3
AWS Region: us-east-1
Relion : ver4.0-beta-2
<Analysis Data>
<Instance Type>
[A] g5-vcpu192-gpu8 (g5.48xlarge)
[B] g5-vcpu96-gpu4 (g5.24xlarge)
[C] g5-vcpu48-gpu4 (g5.12xlarge)
[D] g5-vcpu64-gpu1 (g5.16xlarge)
[E] g5-vcpu32-gpu1 (g5.8xlarge)
[F] g4dn-vcpu96-gpu8 (g4dn.metal)
[G] g4dn-vcpu48-gpu4 (g4dn.12xlarge)
[H] g4dn-vcpu32-gpu1 (g4dn.8xlarge)
[ I ] c6i-vcpu128-gpu0 (c6i.32xlarge)
[ J ] c6id-vcpu128-gpu0 (c6id.32xlarge)
[K] c5d-vcpu96-gpu0 (c5.24xlarge)
[L] g5g-vcpu64-gpu0 (g5g.16xlarge)
[M] g5g-vcpu32-gpu1 (g5g.8xlarge)
<Relion Compiler>
[01] GCC + IntelMPI
[02] Intel + IntelMPI (CPU compile)
[03] GCC + OpenMPI
Results
2. Total cost vs Process time
(Enlarged view near the origin)
3. Process time vs Number of GPUs
<Analysis Data>
<Instance Type>
[A] g5-vcpu192-gpu8 (g5.48xlarge)
[B] g5-vcpu96-gpu4 (g5.24xlarge)
[C] g5-vcpu48-gpu4 (g5.12xlarge)
[D] g5-vcpu64-gpu1 (g5.16xlarge)
[E] g5-vcpu32-gpu1 (g5.8xlarge)
[F] g4dn-vcpu96-gpu8 (g4dn.metal)
[G] g4dn-vcpu48-gpu4 (g4dn.12xlarge)
[H] g4dn-vcpu32-gpu1 (g4dn.8xlarge)
[ I ] c6i-vcpu128-gpu0 (c6i.32xlarge)
[K] c5d-vcpu96-gpu0 (c5.24xlarge)
[L] g5g-vcpu64-gpu0 (g5g.16xlarge)
[M] g5g-vcpu32-gpu1 (g5g.8xlarge)
<Relion Compiler>
[01] GCC + IntelMPI
[02] Intel + IntelMPI (CPU compile)
[03] GCC + OpenMPI
Results
2. Total cost vs Process time
(Enlarged view near the origin)
3. Process time vs Number of GPUs
<Analysis Data>
<Instance Type>
[A] g5-vcpu192-gpu8 (g5.48xlarge)
[B] g5-vcpu96-gpu4 (g5.24xlarge)
[C] g5-vcpu48-gpu4 (g5.12xlarge)
[D] g5-vcpu64-gpu1 (g5.16xlarge)
[E] g5-vcpu32-gpu1 (g5.8xlarge)
[F] g4dn-vcpu96-gpu8 (g4dn.metal)
[G] g4dn-vcpu48-gpu4 (g4dn.12xlarge)
[H] g4dn-vcpu32-gpu1 (g4dn.8xlarge)
[ I ] c6i-vcpu128-gpu0 (c6i.32xlarge)
[ J ] c6id-vcpu128-gpu0 (c6id.32xlarge)
[K] c5d-vcpu96-gpu0 (c5.24xlarge)
[L] g5g-vcpu64-gpu0 (g5g.16xlarge)
[M] g5g-vcpu32-gpu1 (g5g.8xlarge)
<Relion Compiler>
[01] GCC + IntelMPI
[02] Intel + IntelMPI (CPU compile)
[03] GCC + OpenMPI
Results
2. Total cost vs Process time
Enlarged view near the origin
3. Process time vs Number of GPUs
<Analysis Data>
<Instance Type>
[A] c6i-vcpu128-gpu0 (c6i.32xlarge)
[B] c6id-vcpu128-gpu0 (c6id.32xlarge)
[C] c5d-vcpu96-gpu0 (c5d.24xlarge)
<Relion Compiler>
[01] Intel + IntelMPI (CPU compile)
[02] Intel + IntelMPI (GPU compile)
[03] GCC + IntelMPI
Results
2. Total cost vs Process time
3. Process time vs Number of cores