From 03a108d1e9443f70610782042a6facbab37d2fac Mon Sep 17 00:00:00 2001 From: Galen Shipman Date: Thu, 14 Sep 2023 20:31:27 -0700 Subject: [PATCH] update intro with SSNI equation, fix references and spelling --- doc/sphinx/0_intro/introduction.rst | 22 ++++++++++++++++++---- doc/sphinx/1_branson/branson.rst | 4 ++-- doc/sphinx/3_vibe/vibe.rst | 4 ++-- doc/sphinx/5_mlmd/mlmd.rst | 12 ++++++------ 4 files changed, 28 insertions(+), 14 deletions(-) diff --git a/doc/sphinx/0_intro/introduction.rst b/doc/sphinx/0_intro/introduction.rst index 06a29f2f..4840094c 100644 --- a/doc/sphinx/0_intro/introduction.rst +++ b/doc/sphinx/0_intro/introduction.rst @@ -216,11 +216,25 @@ Optional results: Scaled Single Node Improvement ============================== -One element of evaluation will focus on scaled single node improvement: +One element of evaluation will focus on scaled single node improvement (SSNI). SSNI is defined as follows: + +Given two platforms using one as a reference, SSNI is defined as a weighted geometric mean using the following equation. + +.. math:: + + SSNI = N(\prod_{i=1}^{M}(S_i)^{w_i})^\frac{1}{\sum_{i=1}^{M}{W_i}} + + +Where: + +* N = Number of nodes on ATS-5 system / Number of nodes on reference system, + +* M = total number of Benchmarks, + +* S = application speedup; Figure of Merit on ATS-5 system / Figure of Merit on reference system (Crossroads); S must be greater than 1, + +* w = weighting factor. -* Weighted average of Single node performance improvement * number of nodes -* Multiple node configurations can be proposed and may vary on single node performance improvement and number of nodes -* Baseline will be SSNI of Crossroads SPR-HBM Approvals diff --git a/doc/sphinx/1_branson/branson.rst b/doc/sphinx/1_branson/branson.rst index 8ca2b806..a12fe026 100644 --- a/doc/sphinx/1_branson/branson.rst +++ b/doc/sphinx/1_branson/branson.rst @@ -9,7 +9,7 @@ This is the documentation for the ATS-5 Benchmark Branson - 3D hohlraum single n Purpose ======= -From their [site]_: +From their [Branson]_: Branson is not an acronym. @@ -191,4 +191,4 @@ Verification of Results References ========== -.. [site] Alex R. Long, 'Branson', 2023. [Online]. Available: https://github.com/lanl/branson. [Accessed: 22- Feb- 2023] +.. [Branson] Alex R. Long, 'Branson', 2023. [Online]. Available: https://github.com/lanl/branson. [Accessed: 22- Feb- 2023] diff --git a/doc/sphinx/3_vibe/vibe.rst b/doc/sphinx/3_vibe/vibe.rst index 7a048f9e..25306743 100644 --- a/doc/sphinx/3_vibe/vibe.rst +++ b/doc/sphinx/3_vibe/vibe.rst @@ -7,7 +7,7 @@ This is the documentation for the ATS-5 Benchmark, Parthenon-VIBE. Purpose ======= -The Parthenon-VIBE benchmark [site]_ solves the Vector Inviscid Burgers' Equation on a block-AMR mesh. +The Parthenon-VIBE benchmark [Parthenon-VIBE]_ solves the Vector Inviscid Burgers' Equation on a block-AMR mesh. This benchmark is configured to use three levels of mesh resolution and mesh blocks of size 16^3. This AMR configuration is meant to mimic applications which require high resolution, the ability to capture sharp and dynamic interfaces, while balancing global memory footprint and the overhead of "ghost" cells. This configuration should not be changed as it would violate the intent of the benchmark. @@ -227,4 +227,4 @@ Verification of Results References ========== -.. [site] Jonah Miller, 'Parthenon', 2023. [Online]. Available: https://github.com/parthenon-hpc-lab/parthenon. [Accessed: 20- Mar- 2023] +.. [Parthenon-VIBE] Jonah Miller, 'Parthenon', 2023. [Online]. Available: https://github.com/parthenon-hpc-lab/parthenon. [Accessed: 20- Mar- 2023] diff --git a/doc/sphinx/5_mlmd/mlmd.rst b/doc/sphinx/5_mlmd/mlmd.rst index d80b30c7..142f5470 100644 --- a/doc/sphinx/5_mlmd/mlmd.rst +++ b/doc/sphinx/5_mlmd/mlmd.rst @@ -2,12 +2,12 @@ MLMD ****** -This is the documentation for the ATS-5 MLMD Benchmark for HIPPYNN driven kokkos-Lammps Molecular Dynamics. +This is the documentation for the ATS-5 MLMD Benchmark for HIPPYNN [HIPPYNN]_ driven kokkos-Lammps Molecular Dynamics. Purpose ======= -To benchmark perfromance of Lammps driven Molecular Dynamics. The problem configured in this test is a small Ag model built using the data included in the Allegro paper (DOI: https://doi.org/10.1038/s41467-023-36329-y) +To benchmark performance of Lammps [Lammps]_ driven Molecular Dynamics. The problem configured in this test is a small Ag model built using the data included in the Allegro paper (DOI: https://doi.org/10.1038/s41467-023-36329-y) Characteristics =============== @@ -16,7 +16,7 @@ Problem ------- This is a set of simulations on 1,022,400 silver atoms at a 5fs time step. -This benchmark is comparible to the Allegro Ag simulation: https://www.nature.com/articles/s41467-023-36329-y +This benchmark is comparable to the Allegro Ag simulation: https://www.nature.com/articles/s41467-023-36329-y This is a general purpose proxy for many MD simulations. @@ -24,7 +24,7 @@ This test will evaluate Pytorch performance, GPU performance, and MPI performanc Figure of Merit --------------- -The figure of merrit is the throughput of the MD simulations, whch is reported by lammps as 'Matom-step/s'. +The figure of merit is the throughput of the MD simulations, which is reported by Lammps as 'Matom-step/s'. Building ======== @@ -241,5 +241,5 @@ Verification of Results References ========== -.. [site] Nicolas Lubbers, "HIPPYNN" 2021. [Online]. Available: https://github.com/lanl/hippynn. [Accessed: 6- Mar- 2023] -.. [site] Axel Kohlmeyer et. Al, "Lammps". [Online]. Available: https://github.com/lammps/lammps. [Accessed: 6- Mar- 2023] +.. [HIPPYNN] Nicolas Lubbers, "HIPPYNN" 2021. [Online]. Available: https://github.com/lanl/hippynn. [Accessed: 6- Mar- 2023] +.. [Lammps] Axel Kohlmeyer et. Al, "Lammps". [Online]. Available: https://github.com/lammps/lammps. [Accessed: 6- Mar- 2023]