From 452807e4116c872b9d3bc6d2eb905f01a13b4497 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Bob=20Dr=C3=B6ge?= Date: Fri, 4 Sep 2026 16:58:37 +0200 Subject: [PATCH 1/6] use 4 cores for hipBLASLt --- eb_hooks.py | 3 +++ 1 file changed, 3 insertions(+) diff --git a/eb_hooks.py b/eb_hooks.py index 6b284ca7..fb79e6af 100644 --- a/eb_hooks.py +++ b/eb_hooks.py @@ -2475,6 +2475,9 @@ def set_maximum(parallel, max_value): # Deucalion has 32GB HBM for 48 cores per node CPU_TARGET_A64FX: (divide_by_factor, 4), # software-specific limits + 'hipBLASLt': { + '*': (set_maximum, 4), + }, 'libxc': { '*': (divide_by_factor, 2), CPU_TARGET_A64FX: (set_maximum, 6), From e3579bc8730f06a927305b2118fb1f8311cbd644 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Bob=20Dr=C3=B6ge?= Date: Sun, 6 Sep 2026 09:44:30 +0200 Subject: [PATCH 2/6] use single core for hipBLASLt --- eb_hooks.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/eb_hooks.py b/eb_hooks.py index fb79e6af..84d263db 100644 --- a/eb_hooks.py +++ b/eb_hooks.py @@ -2476,7 +2476,7 @@ def set_maximum(parallel, max_value): CPU_TARGET_A64FX: (divide_by_factor, 4), # software-specific limits 'hipBLASLt': { - '*': (set_maximum, 4), + '*': (set_maximum, 1), }, 'libxc': { '*': (divide_by_factor, 2), From a1f67233fca2dd8fab9f6f137bb5536015d13446 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Bob=20Dr=C3=B6ge?= Date: Mon, 7 Sep 2026 16:51:05 +0200 Subject: [PATCH 3/6] set divide_by_factor to 2 for rocBLAS, comment hipBLASLt limits --- eb_hooks.py | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/eb_hooks.py b/eb_hooks.py index 84d263db..de50e29e 100644 --- a/eb_hooks.py +++ b/eb_hooks.py @@ -2475,9 +2475,9 @@ def set_maximum(parallel, max_value): # Deucalion has 32GB HBM for 48 cores per node CPU_TARGET_A64FX: (divide_by_factor, 4), # software-specific limits - 'hipBLASLt': { - '*': (set_maximum, 1), - }, + #'hipBLASLt': { + # '*': (set_maximum, 1), + #}, 'libxc': { '*': (divide_by_factor, 2), CPU_TARGET_A64FX: (set_maximum, 6), @@ -2497,6 +2497,9 @@ def set_maximum(parallel, max_value): 'QuantumESPRESSO': { CPU_TARGET_A64FX: (set_maximum, 6), }, + 'rocBLAS': { + '*': (divide_by_factor, 2), + }, 'TensorFlow': { '*': (divide_by_factor, 2), CPU_TARGET_A64FX: (set_maximum, 8), From 97940ed6a21557bba627add218d25611bc7864c7 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Bob=20Dr=C3=B6ge?= Date: Tue, 8 Sep 2026 09:43:28 +0200 Subject: [PATCH 4/6] set maximum number of cores to 8 for rocBLAS --- eb_hooks.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/eb_hooks.py b/eb_hooks.py index de50e29e..0dc08205 100644 --- a/eb_hooks.py +++ b/eb_hooks.py @@ -2498,7 +2498,7 @@ def set_maximum(parallel, max_value): CPU_TARGET_A64FX: (set_maximum, 6), }, 'rocBLAS': { - '*': (divide_by_factor, 2), + '*': (set_maximum, 8), }, 'TensorFlow': { '*': (divide_by_factor, 2), From 94ceccd303f2c470a35fa2607468a057e5bba693 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Bob=20Dr=C3=B6ge?= Date: Tue, 8 Sep 2026 15:54:34 +0200 Subject: [PATCH 5/6] also max 8 cores for hipBLASLt --- eb_hooks.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/eb_hooks.py b/eb_hooks.py index 0dc08205..d8c8db30 100644 --- a/eb_hooks.py +++ b/eb_hooks.py @@ -2475,9 +2475,9 @@ def set_maximum(parallel, max_value): # Deucalion has 32GB HBM for 48 cores per node CPU_TARGET_A64FX: (divide_by_factor, 4), # software-specific limits - #'hipBLASLt': { - # '*': (set_maximum, 1), - #}, + 'hipBLASLt': { + '*': (set_maximum, 8), + }, 'libxc': { '*': (divide_by_factor, 2), CPU_TARGET_A64FX: (set_maximum, 6), From f152b51df761000f4f84ae920bd303648d9a18c7 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Bob=20Dr=C3=B6ge?= Date: Wed, 9 Sep 2026 20:54:45 +0200 Subject: [PATCH 6/6] reduce number of cores for rocBLAS to 4 --- eb_hooks.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/eb_hooks.py b/eb_hooks.py index d8c8db30..e0ae39e9 100644 --- a/eb_hooks.py +++ b/eb_hooks.py @@ -2498,7 +2498,7 @@ def set_maximum(parallel, max_value): CPU_TARGET_A64FX: (set_maximum, 6), }, 'rocBLAS': { - '*': (set_maximum, 8), + '*': (set_maximum, 4), }, 'TensorFlow': { '*': (divide_by_factor, 2),