Update fig and readme.

Author: ambrad

screamv1-frontier-feb2023/figs/figs.hy

@@ -7,7 +7,7 @@
  (assoc matplotlib.rcParams "savefig.dpi" 300))
 
 (defn get-context [&optional [eul False] [threaded True]]
-  (sv prefix (+ "frontier-v1-scaling1-rocm5?-"
+  (sv prefix (+ "frontier-v1-*-nnodes"
                 (if eul "eul-" "")
                 (if threaded "" "nothrd-"))
       timers (, "CPL:RUN_LOOP" "CPL:ATM_RUN" "a:tl-sc prim_run_subcycle_c"
@@ -20,7 +20,7 @@
   {:prefix prefix
    :machine-name "Frontier"
    :compset "ne1024pg2_ne1024pg2.F2010-SCREAMv1"
-   :glob-data (+ "../data/" prefix "nnodes*-model_timing_stats")
+   :glob-data (+ "../data/" prefix "*-model_timing_stats")
    :timers0 (cut timers 0 1)
    :timers1 (cut timers 0 2)
    :timers2 (cut timers 0 5)
@@ -195,14 +195,14 @@
   (if cpu-sizing
     (cond [(= timer-set :timersa)
            (sv y [20 30 40 50 60 70 80 90 100 125 150 175 200 250 300 350
-                  400 450 500 550 600]
+                  400 450 500 550 600 700]
                yscale 90)
-           (pl.ylim (, 20 600))]
+           (pl.ylim (, 20 750))]
           [(= timer-set :timersbcut)
            (sv y [80 90 100 125 150 175 200 250 300 350 400 450 500 600 700 800 900
                   1000 1200 1400 1600 1800 2000 2500 3000 3500]
                yscale 500)
-           (pl.ylim (, 80 3500))])
+           (pl.ylim (, 80 3800))])
     (cond [(= timer-set :timers2)
            (sv y [50 100 150 200 300 400 500 600 700 800 900 1000
                   1200 1400 1600 1800]

screamv1-frontier-feb2023/figs/perf-table.tex

@@ -4,8 +4,8 @@
 Frontier & 512 & 58.3 & 64.8 & 82.3 & 288.8 & 306.0 & 337.6 \\
  & 1024 & 103.7 & 118.3 & 149.9 & 562.0 & 596.0 & 620.5 \\
  & 2048 & 175.6 & 197.6 & 252.1 & 1012.0 & 1069.2 & 1043.6 \\
- & 4096 & 261.5 & 311.2 & 395.4 & 1776.0 & 1568.6 & 1786.4 \\
- & 8192 & 297.4 & 384.1 & 550.6 & 3171.3 & 2820.8 & 2991.9 \\
+ & 4096 & 283.2 & 332.8 & 424.0 & 1760.2 & 1886.0 & 1693.5 \\
+ & 8192 & 419.5 & 514.5 & 645.3 & 3173.7 & 3349.6 & 2911.1 \\
 \hline
 Summit & 1024 & 70.0 & 78.2 & 100.8 & 401.4 & 465.9 & 382.3 \\
  & 2048 & 109.4 & 132.9 & 169.4 & 744.1 & 870.9 & 679.3 \\

screamv1-frontier-feb2023/figs/v1paper-sdpd-vs-nnode.pdf

@@ -1,4 +1,5 @@
-This directory contains data and code for the Frontier window 10-20 Feb 2023.
+This directory contains data and code for the Frontier window 10-20 Feb 2023,
+plus later additional Frontier runs.
 
 The run*sh scripts are one-off run scripts. They take as an argument the number
 of nodes to run on. run-rocm54.sh and run-rocm54-eul.sh use the tag
@@ -7,7 +8,11 @@ run-rocm51.sh uses the tag
     https://github.com/E3SM-Project/scream/releases/tag/archive%2Fscreamv1-frontier-feb2023-rocm51
 The rocm/5.4, cce/15.0.0 configuration has issues with BLAS in the LND
 component. The rocm/5.1, cce/14.0.2 configuration seems fine. As a result, our
-figures use the rocm51-annotated data.
+figures use the rocm51-annotated data. Later, we were able to redo the
+large-scale simulations. These use the branch
+    https://github.com/E3SM-Project/scream/tree/sarats/frontier-gb which is
+essentially the same rocm/5.1 configuration. These are the
+frontier-v1-scream-gb-o3-ne1024 data sets.
 
 jobmonitor.py is a tool to monitor a single job. If e3sm.exe terminates but the
 job hangs, jobmonitor.py will kill it, thus minimizing hanging time.
@@ -28,8 +33,23 @@ The figs/ directory contains hy (version 0.20 running on any python3) code to
 summarize and plot the data.
 
 The figures use the following subset of the data:
-  frontier-v1-scaling1-rocm51-nnodes512.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.1271946.230210-212030-model_timing_stats
-  frontier-v1-scaling1-rocm51-nnodes1024.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.1273705.230216-000516-model_timing_stats
-  frontier-v1-scaling1-rocm51-nnodes2048.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.1273525.230215-230716-model_timing_stats
-  frontier-v1-scaling1-rocm54-nnodes4096.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.1272650.230213-074922-model_timing_stats
-  frontier-v1-scaling1-rocm54-nnodes8192.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.1272541.230212-204538-model_timing_stats
+  Frontier:
+    frontier-v1-scaling1-rocm51-nnodes512.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.1271946.230210-212030-model_timing_stats
+    frontier-v1-scaling1-rocm51-nnodes1024.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.1273705.230216-000516-model_timing_stats
+    frontier-v1-scaling1-rocm51-nnodes2048.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.1273525.230215-230716-model_timing_stats
+    frontier-v1-scream-gb-o3-ne1024-nnodes4096.ne1024pg2_ne1024pg2.F2010-SCREAMv1-1389464.230729-022636-model_timing_stats
+    frontier-v1-scream-gb-o3-ne1024-nnodes8192.ne1024pg2_ne1024pg2.F2010-SCREAMv1-1389460.230730-003528-model_timing_stats
+  Summit:
+    screamv1-summit-oct2022/data/scream-v1-scaling2-nnodes1024.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.2495303.221008-023937-model_timing_stats
+    screamv1-summit-oct2022/data/scream-v1-scaling2-nnodes2048.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.2495304.221009-093803-model_timing_stats
+    screamv1-summit-oct2022/data/scream-v1-scaling2-nnodes3072.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.2495590.221008-072336-model_timing_stats
+    screamv1-summit-oct2022/data/scream-v1-scaling2-nnodes4096.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.2497059.221010-173053-model_timing_stats
+    screamv1-summit-oct2022/data/scream-v1-scaling2-nnodes4608.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.2495935.221009-090433-model_timing_stats
+  Perlmutter CPU
+    screamv1-pm-cpu-mar2023/data/pm-cpu-v1-scaling1-nnodes1536.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.6746630.230401-112731-model_timing_stats
+    screamv1-pm-cpu-mar2023/data/pm-cpu-v1-scaling1-nnodes2048.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.6907074.230404-030231-model_timing_stats
+  Perlmutter GPU
+    screamv1-pm-gpu-mar2023/data/pm-gpu-v1-scaling1-nnodes384.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.8132452.230425-101318-model_timing_stats
+    screamv1-pm-gpu-mar2023/data/pm-gpu-v1-scaling1-nnodes512.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.5993462.230309-205142-model_timing_stats
+    screamv1-pm-gpu-mar2023/data/pm-gpu-v1-scaling1-nnodes1024.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.8421021.230505-110231-model_timing_stats
+    screamv1-pm-gpu-mar2023/data/pm-gpu-v1-scaling1-nnodes1536.ne1024pg2_ne1024pg2.F2010-SCREAMv1-timing.8168038.230430-014707-model_timing_stats