{compiler,toolchain}[GCCcore-13.2.0] nvidia-compilers, NVHPC 25.7 CUDA-12.4 by adammccartney · Pull Request #24436 · easybuilders/easybuild-easyconfigs

adammccartney · 2025-11-01T12:30:59Z

Draft! -> intended for use in a discussion

Based on #23125,
this tries out one possible variation of an easyconfig for the nvidia compilers and NVHPC
toolchain. That would be suitable for use with EESSI/2023.06

Requires:
+ easybuilders/easybuild-easyblocks#3788
+ easybuilders/easybuild-framework#4927

Based on [easybuilders#23125](easybuilders#23125), this tries out one possible variation of an easyconfig for the nvidia-compilers. This is to make the module compatible with the defaults in EESSI/2023.06. It swaps out the SYSTEM toolchain for GCCcore-13.2.0 and uses that same gcccore version in the other deps. Also default to CUDA-12.4. Requires: + easybuilders/easybuild-easyblocks#3788 + easybuilders/easybuild-framework#4927

github-actions · 2025-11-01T12:31:36Z

Updated software `NVHPC-25.1.eb`

Diff against NVHPC-25.3-CUDA-12.8.0.eb

easybuild/easyconfigs/n/NVHPC/NVHPC-25.3-CUDA-12.8.0.eb

diff --git a/easybuild/easyconfigs/n/NVHPC/NVHPC-25.3-CUDA-12.8.0.eb b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.1.eb
index ccafff8d0a..7c7b0c741f 100644
--- a/easybuild/easyconfigs/n/NVHPC/NVHPC-25.3-CUDA-12.8.0.eb
+++ b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.1.eb
@@ -1,6 +1,5 @@
 name = 'NVHPC'
-version = '25.3'
-versionsuffix = '-CUDA-%(cudaver)s'
+version = '25.1'
 
 homepage = 'https://developer.nvidia.com/hpc-sdk/'
 description = """Complete toolchain based on NVIDIA HPC SDK. Includes C, C++ and FORTRAN
@@ -17,19 +16,18 @@ sources = [local_tarball_tmpl % '%(arch)s']
 checksums = [
     {
         local_tarball_tmpl % 'aarch64':
-            'a2b86cf5141c0a9b0925999521693981451a8d2403367c36c46238163be6f2bb',
+            '0e1d694d54d44559155024d5bab4ca6764eba52d3f27b89f5c252416976e0360',
         local_tarball_tmpl % 'x86_64':
-            'e2b2c911478a5db6a15d1fd258a8c4004dbfccf6f32f4132fe142a24fb7e6f8f',
+            '0813791f8363f4c493db7891b00396ce522cb73910279b8f18a440aedda6727c',
     }
 ]
 
 dependencies = [
-    ('CUDA', '12.8.0'),
-    ('nvidia-compilers', version, versionsuffix),
+    ('nvidia-compilers', version),
 ]
 
-# NVHPC needs CUDA to work. This easyconfig uses external CUDA from EasyBuild.
-# The default CUDA version used by NVHPC will be the version of loaded CUDA module
+# NVHPC needs CUDA to work. This easyconfig uses CUDA bundled in NVHPC
+# The default CUDA version used by NVHPC is the one defined in nvidia-compilers
 
 # The default Compute Capability used by NVHPC is the one defined in nvidia-compilers (if any)

Diff against NVHPC-25.1-CUDA-12.6.0.eb

easybuild/easyconfigs/n/NVHPC/NVHPC-25.1-CUDA-12.6.0.eb

diff --git a/easybuild/easyconfigs/n/NVHPC/NVHPC-25.1-CUDA-12.6.0.eb b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.1.eb
index d34825a8c2..7c7b0c741f 100644
--- a/easybuild/easyconfigs/n/NVHPC/NVHPC-25.1-CUDA-12.6.0.eb
+++ b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.1.eb
@@ -1,6 +1,5 @@
 name = 'NVHPC'
 version = '25.1'
-versionsuffix = '-CUDA-%(cudaver)s'
 
 homepage = 'https://developer.nvidia.com/hpc-sdk/'
 description = """Complete toolchain based on NVIDIA HPC SDK. Includes C, C++ and FORTRAN
@@ -24,12 +23,11 @@ checksums = [
 ]
 
 dependencies = [
-    ('CUDA', '12.6.0'),
-    ('nvidia-compilers', version, versionsuffix),
+    ('nvidia-compilers', version),
 ]
 
-# NVHPC needs CUDA to work. This easyconfig uses external CUDA from EasyBuild.
-# The default CUDA version used by NVHPC will be the version of loaded CUDA module
+# NVHPC needs CUDA to work. This easyconfig uses CUDA bundled in NVHPC
+# The default CUDA version used by NVHPC is the one defined in nvidia-compilers
 
 # The default Compute Capability used by NVHPC is the one defined in nvidia-compilers (if any)

Diff against NVHPC-24.11-CUDA-12.6.0.eb

easybuild/easyconfigs/n/NVHPC/NVHPC-24.11-CUDA-12.6.0.eb

diff --git a/easybuild/easyconfigs/n/NVHPC/NVHPC-24.11-CUDA-12.6.0.eb b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.1.eb
index 546b28e714..7c7b0c741f 100644
--- a/easybuild/easyconfigs/n/NVHPC/NVHPC-24.11-CUDA-12.6.0.eb
+++ b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.1.eb
@@ -1,73 +1,34 @@
 name = 'NVHPC'
-version = '24.11'
-versionsuffix = '-CUDA-%(cudaver)s'
+version = '25.1'
 
 homepage = 'https://developer.nvidia.com/hpc-sdk/'
-description = """C, C++ and Fortran compilers included with the NVIDIA HPC SDK (previously: PGI)"""
+description = """Complete toolchain based on NVIDIA HPC SDK. Includes C, C++ and FORTRAN
+compilers (nvidia-compilers), an MPI implementation based on OpenMPI (NVHPCX)
+and math libraries based on OpenBLAS and ScaLAPACK."""
 
 toolchain = SYSTEM
 
-local_tarball_tmpl = 'nvhpc_2024_%%(version_major)s%%(version_minor)s_Linux_%s_cuda_multi.tar.gz'
 # By downloading, you accept the HPC SDK Software License Agreement
 # https://docs.nvidia.com/hpc-sdk/eula/index.html
-# accept_eula = True
+local_tarball_tmpl = 'nvhpc_2025_%%(version_major)s%%(version_minor)s_Linux_%s_cuda_multi.tar.gz'
 source_urls = ['https://developer.download.nvidia.com/hpc-sdk/%(version)s/']
 sources = [local_tarball_tmpl % '%(arch)s']
 checksums = [
     {
         local_tarball_tmpl % 'aarch64':
-            'f2f64e5dec5e90dad5e12a31a992172b0aa19abf872ef1c54a1a437c7008eefb',
+            '0e1d694d54d44559155024d5bab4ca6764eba52d3f27b89f5c252416976e0360',
         local_tarball_tmpl % 'x86_64':
-            '0c27d66ed0e2d3007d30ac904922a9abf96475197dc0f4dcc6316d235a1dc0c3',
+            '0813791f8363f4c493db7891b00396ce522cb73910279b8f18a440aedda6727c',
     }
 ]
 
-local_gccver = '13.3.0'
 dependencies = [
-    ('GCCcore', local_gccver),
-    ('binutils', '2.42', '', ('GCCcore', local_gccver)),
-    # This is necessary to avoid cases where just libnuma.so.1 is present in the system and -lnuma fails
-    ('numactl', '2.0.18', '', ('GCCcore', local_gccver)),
-    ('CUDA', '12.6.0', '', SYSTEM),
+    ('nvidia-compilers', version),
 ]
 
-module_add_cuda = False
+# NVHPC needs CUDA to work. This easyconfig uses CUDA bundled in NVHPC
+# The default CUDA version used by NVHPC is the one defined in nvidia-compilers
 
-# specify default CUDA version that should be used by NVHPC
-# should match one of the CUDA versions that are included with this NVHPC version
-# (see install_components/Linux_x86_64/$version/cuda/) where $version is the NVHPC version
-# this version can be tweaked from the EasyBuild command line with
-# --try-amend=default_cuda_version="11.0" (for example)
-default_cuda_version = '%(cudaver)s'
+# The default Compute Capability used by NVHPC is the one defined in nvidia-compilers (if any)
 
-# NVHPC EasyBlock supports some features, which can be set via CLI or this easyconfig.
-# The following list gives examples for the easyconfig
-#
-# NVHPC needs CUDA to work. Two options are available: 1) Use NVHPC-bundled CUDA, 2) use system CUDA
-# 1) Bundled CUDA
-#    If no easybuild dependency to CUDA is present, the bundled CUDA is taken. A version needs to be specified with
-#      default_cuda_version = "11.0"
-#    in this easyconfig file; alternatively, it can be specified through the command line during installation with
-#      --try-amend=default_cuda_version="10.2"
-# 2) CUDA provided via EasyBuild
-#    Use CUDA as a dependency, for example
-#      dependencies = [('CUDA', '11.5.0')]
-#    The parameter default_cuda_version still can be set as above.
-#    If not set, it will be deduced from the CUDA module (via $EBVERSIONCUDA)
-#
-# Define a NVHPC-default Compute Capability
-#   cuda_compute_capabilities = "8.0"
-# Can also be specified on the EasyBuild command line via --cuda-compute-capabilities=8.0
-# Only single values supported, not lists of values!
-#
-# Options to add/remove things to/from environment module (defaults shown)
-#   module_byo_compilers = False  # Remove compilers from PATH (Bring-your-own compilers)
-#   module_nvhpc_own_mpi = False  # Add NVHPC's own pre-compiled OpenMPI
-#   module_add_math_libs = False  # Add NVHPC's math libraries (which should be there from CUDA anyway)
-#   module_add_profilers = False  # Add NVHPC's NVIDIA Profilers
-#   module_add_nccl = False       # Add NVHPC's NCCL library
-#   module_add_nvshmem = False    # Add NVHPC's NVSHMEM library
-#   module_add_cuda = False       # Add NVHPC's bundled CUDA
-
-# this bundle serves as a compiler-only toolchain, so it should be marked as compiler (important for HMNS)
-moduleclass = 'compiler'
+moduleclass = 'toolchain'

Updated software `NVHPC-25.3.eb`

Diff against NVHPC-25.3-CUDA-12.8.0.eb

easybuild/easyconfigs/n/NVHPC/NVHPC-25.3-CUDA-12.8.0.eb

diff --git a/easybuild/easyconfigs/n/NVHPC/NVHPC-25.3-CUDA-12.8.0.eb b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.3.eb
index ccafff8d0a..4c6f1148c3 100644
--- a/easybuild/easyconfigs/n/NVHPC/NVHPC-25.3-CUDA-12.8.0.eb
+++ b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.3.eb
@@ -1,6 +1,5 @@
 name = 'NVHPC'
 version = '25.3'
-versionsuffix = '-CUDA-%(cudaver)s'
 
 homepage = 'https://developer.nvidia.com/hpc-sdk/'
 description = """Complete toolchain based on NVIDIA HPC SDK. Includes C, C++ and FORTRAN
@@ -24,12 +23,11 @@ checksums = [
 ]
 
 dependencies = [
-    ('CUDA', '12.8.0'),
-    ('nvidia-compilers', version, versionsuffix),
+    ('nvidia-compilers', version),
 ]
 
-# NVHPC needs CUDA to work. This easyconfig uses external CUDA from EasyBuild.
-# The default CUDA version used by NVHPC will be the version of loaded CUDA module
+# NVHPC needs CUDA to work. This easyconfig uses CUDA bundled in NVHPC
+# The default CUDA version used by NVHPC is the one defined in nvidia-compilers
 
 # The default Compute Capability used by NVHPC is the one defined in nvidia-compilers (if any)

Diff against NVHPC-25.1-CUDA-12.6.0.eb

easybuild/easyconfigs/n/NVHPC/NVHPC-25.1-CUDA-12.6.0.eb

diff --git a/easybuild/easyconfigs/n/NVHPC/NVHPC-25.1-CUDA-12.6.0.eb b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.3.eb
index d34825a8c2..4c6f1148c3 100644
--- a/easybuild/easyconfigs/n/NVHPC/NVHPC-25.1-CUDA-12.6.0.eb
+++ b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.3.eb
@@ -1,6 +1,5 @@
 name = 'NVHPC'
-version = '25.1'
-versionsuffix = '-CUDA-%(cudaver)s'
+version = '25.3'
 
 homepage = 'https://developer.nvidia.com/hpc-sdk/'
 description = """Complete toolchain based on NVIDIA HPC SDK. Includes C, C++ and FORTRAN
@@ -17,19 +16,18 @@ sources = [local_tarball_tmpl % '%(arch)s']
 checksums = [
     {
         local_tarball_tmpl % 'aarch64':
-            '0e1d694d54d44559155024d5bab4ca6764eba52d3f27b89f5c252416976e0360',
+            'a2b86cf5141c0a9b0925999521693981451a8d2403367c36c46238163be6f2bb',
         local_tarball_tmpl % 'x86_64':
-            '0813791f8363f4c493db7891b00396ce522cb73910279b8f18a440aedda6727c',
+            'e2b2c911478a5db6a15d1fd258a8c4004dbfccf6f32f4132fe142a24fb7e6f8f',
     }
 ]
 
 dependencies = [
-    ('CUDA', '12.6.0'),
-    ('nvidia-compilers', version, versionsuffix),
+    ('nvidia-compilers', version),
 ]
 
-# NVHPC needs CUDA to work. This easyconfig uses external CUDA from EasyBuild.
-# The default CUDA version used by NVHPC will be the version of loaded CUDA module
+# NVHPC needs CUDA to work. This easyconfig uses CUDA bundled in NVHPC
+# The default CUDA version used by NVHPC is the one defined in nvidia-compilers
 
 # The default Compute Capability used by NVHPC is the one defined in nvidia-compilers (if any)

Diff against NVHPC-24.11-CUDA-12.6.0.eb

easybuild/easyconfigs/n/NVHPC/NVHPC-24.11-CUDA-12.6.0.eb

diff --git a/easybuild/easyconfigs/n/NVHPC/NVHPC-24.11-CUDA-12.6.0.eb b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.3.eb
index 546b28e714..4c6f1148c3 100644
--- a/easybuild/easyconfigs/n/NVHPC/NVHPC-24.11-CUDA-12.6.0.eb
+++ b/easybuild/easyconfigs/n/NVHPC/NVHPC-25.3.eb
@@ -1,73 +1,34 @@
 name = 'NVHPC'
-version = '24.11'
-versionsuffix = '-CUDA-%(cudaver)s'
+version = '25.3'
 
 homepage = 'https://developer.nvidia.com/hpc-sdk/'
-description = """C, C++ and Fortran compilers included with the NVIDIA HPC SDK (previously: PGI)"""
+description = """Complete toolchain based on NVIDIA HPC SDK. Includes C, C++ and FORTRAN
+compilers (nvidia-compilers), an MPI implementation based on OpenMPI (NVHPCX)
+and math libraries based on OpenBLAS and ScaLAPACK."""
 
 toolchain = SYSTEM
 
-local_tarball_tmpl = 'nvhpc_2024_%%(version_major)s%%(version_minor)s_Linux_%s_cuda_multi.tar.gz'
 # By downloading, you accept the HPC SDK Software License Agreement
 # https://docs.nvidia.com/hpc-sdk/eula/index.html
-# accept_eula = True
+local_tarball_tmpl = 'nvhpc_2025_%%(version_major)s%%(version_minor)s_Linux_%s_cuda_multi.tar.gz'
 source_urls = ['https://developer.download.nvidia.com/hpc-sdk/%(version)s/']
 sources = [local_tarball_tmpl % '%(arch)s']
 checksums = [
     {
         local_tarball_tmpl % 'aarch64':
-            'f2f64e5dec5e90dad5e12a31a992172b0aa19abf872ef1c54a1a437c7008eefb',
+            'a2b86cf5141c0a9b0925999521693981451a8d2403367c36c46238163be6f2bb',
         local_tarball_tmpl % 'x86_64':
-            '0c27d66ed0e2d3007d30ac904922a9abf96475197dc0f4dcc6316d235a1dc0c3',
+            'e2b2c911478a5db6a15d1fd258a8c4004dbfccf6f32f4132fe142a24fb7e6f8f',
     }
 ]
 
-local_gccver = '13.3.0'
 dependencies = [
-    ('GCCcore', local_gccver),
-    ('binutils', '2.42', '', ('GCCcore', local_gccver)),
-    # This is necessary to avoid cases where just libnuma.so.1 is present in the system and -lnuma fails
-    ('numactl', '2.0.18', '', ('GCCcore', local_gccver)),
-    ('CUDA', '12.6.0', '', SYSTEM),
+    ('nvidia-compilers', version),
 ]
 
-module_add_cuda = False
+# NVHPC needs CUDA to work. This easyconfig uses CUDA bundled in NVHPC
+# The default CUDA version used by NVHPC is the one defined in nvidia-compilers
 
-# specify default CUDA version that should be used by NVHPC
-# should match one of the CUDA versions that are included with this NVHPC version
-# (see install_components/Linux_x86_64/$version/cuda/) where $version is the NVHPC version
-# this version can be tweaked from the EasyBuild command line with
-# --try-amend=default_cuda_version="11.0" (for example)
-default_cuda_version = '%(cudaver)s'
+# The default Compute Capability used by NVHPC is the one defined in nvidia-compilers (if any)
 
-# NVHPC EasyBlock supports some features, which can be set via CLI or this easyconfig.
-# The following list gives examples for the easyconfig
-#
-# NVHPC needs CUDA to work. Two options are available: 1) Use NVHPC-bundled CUDA, 2) use system CUDA
-# 1) Bundled CUDA
-#    If no easybuild dependency to CUDA is present, the bundled CUDA is taken. A version needs to be specified with
-#      default_cuda_version = "11.0"
-#    in this easyconfig file; alternatively, it can be specified through the command line during installation with
-#      --try-amend=default_cuda_version="10.2"
-# 2) CUDA provided via EasyBuild
-#    Use CUDA as a dependency, for example
-#      dependencies = [('CUDA', '11.5.0')]
-#    The parameter default_cuda_version still can be set as above.
-#    If not set, it will be deduced from the CUDA module (via $EBVERSIONCUDA)
-#
-# Define a NVHPC-default Compute Capability
-#   cuda_compute_capabilities = "8.0"
-# Can also be specified on the EasyBuild command line via --cuda-compute-capabilities=8.0
-# Only single values supported, not lists of values!
-#
-# Options to add/remove things to/from environment module (defaults shown)
-#   module_byo_compilers = False  # Remove compilers from PATH (Bring-your-own compilers)
-#   module_nvhpc_own_mpi = False  # Add NVHPC's own pre-compiled OpenMPI
-#   module_add_math_libs = False  # Add NVHPC's math libraries (which should be there from CUDA anyway)
-#   module_add_profilers = False  # Add NVHPC's NVIDIA Profilers
-#   module_add_nccl = False       # Add NVHPC's NCCL library
-#   module_add_nvshmem = False    # Add NVHPC's NVSHMEM library
-#   module_add_cuda = False       # Add NVHPC's bundled CUDA
-
-# this bundle serves as a compiler-only toolchain, so it should be marked as compiler (important for HMNS)
-moduleclass = 'compiler'
+moduleclass = 'toolchain'

adammccartney · 2025-11-01T12:34:01Z

sorry, added this by mistake!

lexming and others added 4 commits June 18, 2025 11:07

split NVHPC v25 easyconfigs into nvidia-compilers

f7fa3f3

clarify usage of CUDA compute capability in nvidia-compilers

9380584

clarify usage of CUDA compute capability in NVHPC v25

5c9f320

github-actions bot added change new update labels Nov 1, 2025

adammccartney closed this Nov 1, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

{compiler,toolchain}[GCCcore-13.2.0] nvidia-compilers, NVHPC 25.7 CUDA-12.4#24436

{compiler,toolchain}[GCCcore-13.2.0] nvidia-compilers, NVHPC 25.7 CUDA-12.4#24436
adammccartney wants to merge 4 commits intoeasybuilders:developfrom
adammccartney:nvidia-compilers257-GCCcore-13.2.0

adammccartney commented Nov 1, 2025 •

edited

Loading

Uh oh!

github-actions bot commented Nov 1, 2025

Uh oh!

adammccartney commented Nov 1, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

adammccartney commented Nov 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Nov 1, 2025

Updated software NVHPC-25.1.eb

Updated software NVHPC-25.3.eb

Uh oh!

adammccartney commented Nov 1, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

adammccartney commented Nov 1, 2025 •

edited

Loading

Updated software `NVHPC-25.1.eb`

Updated software `NVHPC-25.3.eb`