some new scaling numbers on Frontier

Exec/science/flame_wave/scaling/frontier/frontier-scaling-2024-07-04-subch_simple.txt

@@ -0,0 +1,34 @@
+# scaling numbers for the 3D XRB -- using subch_simple
+# using the same inputs.He.25cm.static.1000Hz as previously
+# modules:
+#
+#   module load PrgEnv-gnu
+#   module load cray-mpich/8.1.27
+#   module load craype-accel-amd-gfx90a
+#   module load amd-mixed/6.0.0
+#   module unload darshan-runtime
+#
+# build info:
+#
+# EOS: /ccs/home/zingale/Microphysics/EOS/helmholtz
+# NETWORK: /ccs/home/zingale/Microphysics/networks/subch_simple
+# CONDUCTIVITY: /ccs/home/zingale/Microphysics/conductivity/stellar
+# INTEGRATOR: VODE
+# SCREENING: screen5
+#
+# Castro       git describe: 24.07
+# AMReX        git describe: 24.07-3-g7dc2081a9
+# Microphysics git describe: 24.07-1-g0a96241b
+
+
+# nodes  rocm      mag_grid_size   avg time /   std dev
+#                          step
+#  48      6.0          128      --- crashes due to not enough GPU memory ---
+  64      6.0          128           167.502   0.419448
+ 128      6.0          128            85.4082  0.29416
+ 256      6.0          128            46.4092  0.876868
+ 384      6.0          128            37.5447  0.138806
+ 512      6.0          128            25.5446  0.123848
+1024      6.0          128            17.3517  0.0857189
+#2048      6.0          128
+

Exec/science/flame_wave/scaling/frontier/frontier-scaling-2024-07-04.txt

@@ -0,0 +1,39 @@
+# new scaling numbers for the 3D XRB
+# using the same inputs.He.25cm.static.1000Hz as previously
+# modules:
+#
+#   module load PrgEnv-gnu
+#   module load cray-mpich/8.1.27
+#   module load craype-accel-amd-gfx90a
+#   module load amd-mixed/6.0.0
+#   module unload darshan-runtime
+#
+# build info:
+#
+# EOS: /ccs/home/zingale/Microphysics/EOS/helmholtz
+# NETWORK: /ccs/home/zingale/Microphysics/networks/iso7
+# CONDUCTIVITY: /ccs/home/zingale/Microphysics/conductivity/stellar
+# INTEGRATOR: VODE
+# SCREENING: screen5
+#
+# Castro       git describe: 24.07
+# AMReX        git describe: 24.07-3-g7dc2081a9
+# Microphysics git describe: 24.07-1-g0a96241b
+
+# nodes  rocm      mag_grid_size   avg time /   std dev
+#                          step
+  48      6.0          128           69.3646   0.286297
+  64      6.0          128           50.2029   0.681104
+ 128      6.0          128           28.4001   0.221368
+ 256      6.0          128           15.7771   0.133056
+ 384      6.0          128           12.8213   0.37406
+ 512      6.0          128            9.6077   0.331669
+1024      6.0          128            6.66329  0.103599
+#2048      6.0          128
+
+
+
+# in contrast to the previous run, we've disabled all inlining with
+# ROCm to get around some compiler bugs, so that might explain some
+# slight slowdown here.
+

Exec/science/flame_wave/scaling/frontier/frontier-scaling-rkc-2024-07-04.txt

@@ -0,0 +1,39 @@
+# new scaling numbers for the 3D XRB
+# using the same inputs.He.25cm.static.1000Hz as previously
+# modules:
+#
+#   module load PrgEnv-gnu
+#   module load cray-mpich/8.1.27
+#   module load craype-accel-amd-gfx90a
+#   module load amd-mixed/6.0.0
+#   module unload darshan-runtime
+#
+# build info:
+#
+# EOS: /ccs/home/zingale/Microphysics/EOS/helmholtz
+# NETWORK: /ccs/home/zingale/Microphysics/networks/iso7
+# CONDUCTIVITY: /ccs/home/zingale/Microphysics/conductivity/stellar
+# INTEGRATOR: RKC
+# SCREENING: screen5
+#
+# Castro       git describe: 24.07
+# AMReX        git describe: 24.07-3-g7dc2081a9
+# Microphysics git describe: 24.07-1-g0a96241b
+
+# nodes  rocm      mag_grid_size   avg time /   std dev
+#                          step
+  48      6.0          128           54.7583   0.287289
+  64      6.0          128           39.4336   0.32115
+ 128      6.0          128           22.8627   0.22725
+ 256      6.0          128           12.9558   0.12379
+ 384      6.0          128           10.3259   0.123708
+ 512      6.0          128            7.9563   0.104311
+1024      6.0          128            5.65801  0.109306
+#2048      6.0          128
+
+
+
+# in contrast to the previous run, we've disabled all inlining with
+# ROCm to get around some compiler bugs, so that might explain some
+# slight slowdown here.
+

Exec/science/flame_wave/scaling/frontier/frontier_scaling.py

@@ -9,13 +9,13 @@
 plt.rc("axes", linewidth=1.5)
 plt.rc("lines", markeredgewidth=1.5)
 
-frontier_data = np.loadtxt("frontier-scaling-2023-04-06.txt")
+frontier_data = np.loadtxt("frontier-scaling-2024-07-04.txt")
 
 frontier_nodes = frontier_data[:, 0]
 frontier_times = frontier_data[:, 3]
 frontier_std = frontier_data[:, 4]
 
-frontier_rkc_data = np.loadtxt("frontier-scaling-rkc-2023-05-31.txt")
+frontier_rkc_data = np.loadtxt("frontier-scaling-rkc-2024-07-04.txt")
 
 frontier_rkc_nodes = frontier_rkc_data[:, 0]
 frontier_rkc_times = frontier_rkc_data[:, 3]
@@ -41,7 +41,7 @@ def trend_line(c, t):
 
 ax.errorbar(frontier_nodes, frontier_times, yerr=frontier_std, ls="None", marker="x", label="Frontier (ROCm 5.3)")
 ax.errorbar(frontier_rkc_nodes, frontier_rkc_times, yerr=frontier_rkc_std, ls="None", marker="x", label="Frontier (RKC integrator)")
-ax.errorbar(summit_nodes, summit_times, yerr=summit_std, ls="None", marker="x", label="Summit (CUDA 11.4)")
+ax.errorbar(summit_nodes, summit_times, yerr=summit_std, ls="None", marker="^", label="Summit (CUDA 11.4)")
 
 c, t = trend_line(frontier_nodes, frontier_times)
 ax.plot(c, t, alpha=0.5, linestyle=":")