{lib}[GCCcore/12.2.0,foss/2022b] PyTorch v1.13.1, cuDNN v8.5.0.96, magma v2.7.1, ... w/ CUDA 11.7.0

easybuild/easyconfigs/c/cuDNN/cuDNN-8.5.0.96-CUDA-11.7.0.eb

@@ -0,0 +1,40 @@
+name = 'cuDNN'
+version = '8.5.0.96'
+versionsuffix = '-CUDA-%(cudaver)s'
+
+homepage = 'https://developer.nvidia.com/cudnn'
+description = """The NVIDIA CUDA Deep Neural Network library (cuDNN) is
+a GPU-accelerated library of primitives for deep neural networks."""
+
+toolchain = SYSTEM
+
+# note: cuDNN is tied to specific to CUDA versions,
+# see also https://docs.nvidia.com/deeplearning/cudnn/support-matrix/index.html#cudnn-cuda-hardware-versions
+local_short_ver = '.'.join(version.split('.')[:3])
+source_urls = ['https://developer.download.nvidia.com/compute/redist/cudnn/'
+               'v%s/local_installers/%%(cudashortver)s/' % local_short_ver]
+sources = ['%(namelower)s-linux-%(cudnnarch)s-%(version)s_cuda%(cudamajver)s-archive.tar.xz']
+checksums = [
+    {
+        '%(namelower)s-linux-x86_64-%(version)s_cuda%(cudamajver)s-archive.tar.xz':
+            '5454a6fd94f008728caae9adad993c4e85ef36302e26bce43bea7d458a5e7b6d',
+        '%(namelower)s-linux-ppc64le-%(version)s_cuda%(cudamajver)s-archive.tar.xz':
+            '00373c3d5e0b536a5557d0d0eb50706777f213a222b4030e1b71b1bec43d205f',
+        '%(namelower)s-linux-sbsa-%(version)s_cuda%(cudamajver)s-archive.tar.xz':
+            '86780abbecd4634e7363fad1d000ae23b7905a5f8383bddbf7332c6934791dde',
+    }
+]
+
+dependencies = [('CUDA', '11.7.0')]
+
+sanity_check_paths = {
+    'files': [
+        'include/cudnn.h', 'lib64/libcudnn_adv_infer_static.a', 'lib64/libcudnn_adv_train_static.a',
+        'lib64/libcudnn_cnn_infer_static.a', 'lib64/libcudnn_cnn_train_static.a',
+        'lib64/libcudnn_ops_infer_static.a', 'lib64/libcudnn_ops_train_static.a',
+        'lib64/libcudnn.%s' % SHLIB_EXT
+    ],
+    'dirs': ['include', 'lib64'],
+}
+
+moduleclass = 'numlib'

easybuild/easyconfigs/m/magma/magma-2.7.1-foss-2022b-CUDA-11.7.0.eb

@@ -0,0 +1,40 @@
+easyblock = 'CMakeMake'
+
+name = 'magma'
+version = '2.7.1'
+versionsuffix = '-CUDA-%(cudaver)s'
+
+homepage = 'https://icl.cs.utk.edu/magma/'
+description = """The MAGMA project aims to develop a dense linear algebra library similar to
+ LAPACK but for heterogeneous/hybrid architectures, starting with current Multicore+GPU systems."""
+
+toolchain = {'name': 'foss', 'version': '2022b'}
+toolchainopts = {'pic': True, 'openmp': True}
+
+source_urls = ['https://icl.cs.utk.edu/projectsfiles/%(name)s/downloads/']
+sources = [SOURCE_TAR_GZ]
+checksums = ['d9c8711c047a38cae16efde74bee2eb3333217fd2711e1e9b8606cbbb4ae1a50']
+
+builddependencies = [
+    ('CMake', '3.24.3'),
+]
+dependencies = [
+    ('CUDA', '11.7.0', '', SYSTEM),
+    ('UCX-CUDA', '1.13.1', versionsuffix),
+]
+
+# default CUDA compute capabilities to use (override via --cuda-compute-capabilities)
+cuda_compute_capabilities = ['3.5', '5.0', '6.0', '7.0', '7.5', '8.0', '8.6']
+
+# make sure both static and shared libs are built
+local_common_opts = '-DCMAKE_CUDA_FLAGS=-allow-unsupported-compiler -DGPU_TARGET="%%(cuda_sm_space_sep)s"'
+configopts = [
+    (local_common_opts + ' -DBUILD_SHARED_LIBS=%s ') % local_shared for local_shared in ('ON', 'OFF')
+]
+
+sanity_check_paths = {
+    'files': ['lib/libmagma.%s' % SHLIB_EXT, 'lib/libmagma.a'],
+    'dirs': ['include'],
+}
+
+moduleclass = 'math'

easybuild/easyconfigs/n/NCCL/NCCL-2.16.2-GCCcore-12.2.0-CUDA-11.7.0.eb

@@ -0,0 +1,33 @@
+name = 'NCCL'
+version = '2.16.2'
+versionsuffix = '-CUDA-%(cudaver)s'
+
+homepage = 'https://developer.nvidia.com/nccl'
+description = """The NVIDIA Collective Communications Library (NCCL) implements multi-GPU and multi-node collective
+communication primitives that are performance optimized for NVIDIA GPUs."""
+
+toolchain = {'name': 'GCCcore', 'version': '12.2.0'}
+
+github_account = 'NVIDIA'
+source_urls = [GITHUB_SOURCE]
+sources = ['v%(version)s-1.tar.gz']
+patches = ['NCCL-2.16.2_fix-cpuid.patch']
+checksums = [
+    {'v2.16.2-1.tar.gz': '7f7c738511a8876403fc574d13d48e7c250d934d755598d82e14bab12236fc64'},
+    {'NCCL-2.16.2_fix-cpuid.patch': '0459ecadcd32b2a7a000a2ce4f675afba908b2c0afabafde585330ff4f83e277'},
+]
+
+builddependencies = [('binutils', '2.39')]
+
+dependencies = [
+    ('CUDA', '11.7.0', '', SYSTEM),
+    ('UCX-CUDA', '1.13.1', versionsuffix),
+]
+
+prebuildopts = "sed -i 's/NVCUFLAGS  := /NVCUFLAGS  := -allow-unsupported-compiler /' makefiles/common.mk && "
+buildopts = "VERBOSE=1"
+
+# default CUDA compute capabilities to use (override via --cuda-compute-capabilities)
+cuda_compute_capabilities = ['3.5', '5.0', '6.0', '7.0', '7.5', '8.0', '8.6']
+
+moduleclass = 'lib'

easybuild/easyconfigs/p/PyTorch/PyTorch-1.13.1-foss-2022b-CUDA-11.7.0.eb

@@ -0,0 +1,197 @@
+name = 'PyTorch'
+version = '1.13.1'
+versionsuffix = '-CUDA-%(cudaver)s'
+
+homepage = 'https://pytorch.org/'
+description = """Tensors and Dynamic neural networks in Python with strong GPU acceleration.
+PyTorch is a deep learning framework that puts Python first."""
+
+toolchain = {'name': 'foss', 'version': '2022b'}
+
+source_urls = [GITHUB_RELEASE]
+sources = ['%(namelower)s-v%(version)s.tar.gz']
+patches = [
+    'PyTorch-1.7.0_disable-dev-shm-test.patch',
+    'PyTorch-1.10.0_fix-kineto-crash.patch',
+    'PyTorch-1.11.0_fix-fp16-quantization-without-fbgemm.patch',
+    'PyTorch-1.11.1_skip-test_init_from_local_shards.patch',
+    'PyTorch-1.12.0_fix-EmbeddingBag-without-fbgemm.patch',
+    'PyTorch-1.12.1_add-hypothesis-suppression.patch',
+    'PyTorch-1.12.1_fix-skip-decorators.patch',
+    'PyTorch-1.12.1_fix-test_cpp_extensions_jit.patch',
+    'PyTorch-1.12.1_fix-test_wishart_log_prob.patch',
+    'PyTorch-1.12.1_fix-TestTorch.test_to.patch',
+    'PyTorch-1.12.1_fix-use-after-free-in-tensorpipe-agent.patch',
+    'PyTorch-1.12.1_fix-vsx-loadu.patch',
+    'PyTorch-1.12.1_fix-vsx-vector-funcs.patch',
+    'PyTorch-1.12.1_skip-test_round_robin.patch',
+    'PyTorch-1.13.1_allow-GCC-12-for-CUDA-11.7.patch',
+    'PyTorch-1.13.1_disable-test-sharding.patch',
+    'PyTorch-1.13.1_fix-duplicate-kDefaultTimeout-definition.patch',
+    'PyTorch-1.13.1_fix-flaky-jit-test.patch',
+    'PyTorch-1.13.1_fix-fsdp-fp16-test.patch',
+    'PyTorch-1.13.1_fix-fsdp-tp-integration-test.patch',
+    'PyTorch-1.13.1_fix-gcc-12-missing-includes.patch',
+    'PyTorch-1.13.1_fix-gcc-12-warning-in-fbgemm.patch',
+    'PyTorch-1.13.1_fix-kineto-crash-on-exit.patch',
+    'PyTorch-1.13.1_fix-numpy-deprecations.patch',
+    'PyTorch-1.13.1_fix-protobuf-dependency.patch',
+    'PyTorch-1.13.1_fix-pytest-args.patch',
+    'PyTorch-1.13.1_fix-python-3.11-compat.patch',
+    'PyTorch-1.13.1_fix-test-ops-conf.patch',
+    'PyTorch-1.13.1_fix-warning-in-test-cpp-api.patch',
+    'PyTorch-1.13.1_fix-wrong-check-in-fsdp-tests.patch',
+    'PyTorch-1.13.1_increase-tolerance-test_jit.patch',
+    'PyTorch-1.13.1_increase-tolerance-test_ops.patch',
+    'PyTorch-1.13.1_increase-tolerance-test_optim.patch',
+    'PyTorch-1.13.1_install-vsx-vec-headers.patch',
+    'PyTorch-1.13.1_no-cuda-stubs-rpath.patch',
+    'PyTorch-1.13.1_remove-flaky-test-in-testnn.patch',
+    'PyTorch-1.13.1_skip-failing-grad-test.patch',
+    'PyTorch-1.13.1_skip-failing-singular-grad-test.patch',
+    'PyTorch-1.13.1_skip-test_find_unused_parameters-detail.patch',
+    'PyTorch-1.13.1_skip-test-requiring-online-access.patch',
+    'PyTorch-1.13.1_skip-tests-without-fbgemm.patch',
+    'PyTorch-1.13.1_workaround-gcc12-destructor-exception-bug.patch',
+    'PyTorch-2.0.1_avoid-test_quantization-failures.patch',
+]
+checksums = [
+    {'pytorch-v1.13.1.tar.gz': 'dbc229ee9750b02b514937d017744443a269ea0241ed3f32b9af0703589d25d4'},
+    {'PyTorch-1.7.0_disable-dev-shm-test.patch': '622cb1eaeadc06e13128a862d9946bcc1f1edd3d02b259c56a9aecc4d5406b8a'},
+    {'PyTorch-1.10.0_fix-kineto-crash.patch': 'dc467333b28162149af8f675929d8c6bf219f23230bfc0d39af02ba4f6f882eb'},
+    {'PyTorch-1.11.0_fix-fp16-quantization-without-fbgemm.patch':
+     'cc526130b6446bbbf5f0f7372d3aeee3e7d4c4d6e471524dff028b430b152934'},
+    {'PyTorch-1.11.1_skip-test_init_from_local_shards.patch':
+     '4aeb1b0bc863d4801b0095cbce69f8794066748f0df27c6aaaf729c5ecba04b7'},
+    {'PyTorch-1.12.0_fix-EmbeddingBag-without-fbgemm.patch':
+     '090598592283e3fc46ee08a68b6a6afe07be41b26514afba51834408bf1c98ed'},
+    {'PyTorch-1.12.1_add-hypothesis-suppression.patch':
+     'e71ffb94ebe69f580fa70e0de84017058325fdff944866d6bd03463626edc32c'},
+    {'PyTorch-1.12.1_fix-skip-decorators.patch': 'e3ca6e42b2fa592ea095939fb59ab875668a058479407db3f3684cc5c6f4146c'},
+    {'PyTorch-1.12.1_fix-test_cpp_extensions_jit.patch':
+     '1efc9850c431d702e9117d4766277d3f88c5c8b3870997c9974971bce7f2ab83'},
+    {'PyTorch-1.12.1_fix-test_wishart_log_prob.patch':
+     'cf475ae6e6234b96c8d1bf917597c5176c94b3ccd940b72f2e1cd0c979580f45'},
+    {'PyTorch-1.12.1_fix-TestTorch.test_to.patch': '75f27987c3f25c501e719bd2b1c70a029ae0ee28514a97fe447516aee02b1535'},
+    {'PyTorch-1.12.1_fix-use-after-free-in-tensorpipe-agent.patch':
+     '0bd7e88b92c4c6f0fecf01746009858ba19f2df68b10b88c41485328a531875d'},
+    {'PyTorch-1.12.1_fix-vsx-loadu.patch': '8bfe3c94ada1dd1f7974a1261a8b576fb7ae944050fa1c7830fca033831123b2'},
+    {'PyTorch-1.12.1_fix-vsx-vector-funcs.patch': 'caccbf60f62eac313896c1eaec78b08f5d0fdfcb907079087490bb13d1561aa2'},
+    {'PyTorch-1.12.1_skip-test_round_robin.patch': '63d4849b78605aa088fdff695637d9473ea60dee603a3ff7f788690d70c55349'},
+    {'PyTorch-1.13.1_allow-GCC-12-for-CUDA-11.7.patch':
+     '4c9a4247dcf6e0f62fda2e7283f7de6f7c801d5e61c39d27a91a287f9d363d68'},
+    {'PyTorch-1.13.1_disable-test-sharding.patch': 'df2074adeba47998ce2993d99ca64eb6f1c79ab7057f553b436efdec264d3572'},
+    {'PyTorch-1.13.1_fix-duplicate-kDefaultTimeout-definition.patch':
+     '882f8cfaf33490a4372928fb6673cbbfa40e5be1b64bf7e0cc2924d73cf872e8'},
+    {'PyTorch-1.13.1_fix-flaky-jit-test.patch': '71efdeb29b5e5b4982c9f5cb2182733654a34d52f85bb5487bc4d7d99b86101b'},
+    {'PyTorch-1.13.1_fix-fsdp-fp16-test.patch': '8ae68e60d6e1f92f50322b7f0381c7e65251fba32d7606e3a238a36a2f55b5cf'},
+    {'PyTorch-1.13.1_fix-fsdp-tp-integration-test.patch':
+     '31e2d63b54ae1a8c554575f46db79bf8bbda851b6ca0ffe623c4911207a3c2bc'},
+    {'PyTorch-1.13.1_fix-gcc-12-missing-includes.patch':
+     '18df8c61ecaa9fb659346c1e172828bca6b069f0145bb8f6a36b0a23b7bef0a6'},
+    {'PyTorch-1.13.1_fix-gcc-12-warning-in-fbgemm.patch':
+     '5c7be91a6096083a0b1315efe0001537499c600f1f569953c6a2c7f4cc1d0910'},
+    {'PyTorch-1.13.1_fix-kineto-crash-on-exit.patch':
+     'f1e6808ee8d91a2ad76e0caedb4685e5aec3008d5e2e3c3c3e88cbb25cbd71b4'},
+    {'PyTorch-1.13.1_fix-numpy-deprecations.patch': 'f461b570efe0434ddd806bf2fa7020eb213e3ed89d0eb4403e076f4276ba2a46'},
+    {'PyTorch-1.13.1_fix-protobuf-dependency.patch':
+     '8bd755a0cab7233a243bc65ca57c9630dfccdc9bf8c9792f0de4e07a644fcb00'},
+    {'PyTorch-1.13.1_fix-pytest-args.patch': 'd3e3c841cf8d73683750f29326f2be56ee0bb5df7ff522baf7d7c3f301a91ec2'},
+    {'PyTorch-1.13.1_fix-python-3.11-compat.patch': 'fa4eb0e27e00a90bb217b77c0023089c4659c03f37d781ab4a681bdcb4f0432f'},
+    {'PyTorch-1.13.1_fix-test-ops-conf.patch': 'df652eec7753864ebebbfeca546929a53e3fb8f24259d5c9b964266a8551198c'},
+    {'PyTorch-1.13.1_fix-warning-in-test-cpp-api.patch':
+     'bdde0f2105215c95a54de64ec4b1a4520528510663174fef6d5b900eb1db3937'},
+    {'PyTorch-1.13.1_fix-wrong-check-in-fsdp-tests.patch':
+     'cbb5ca9ad668a504a456a2cc02d7254b79ddfd9a971a1648f0508fb103a9fc89'},
+    {'PyTorch-1.13.1_increase-tolerance-test_jit.patch':
+     'b97913754a0ae0887b8137db0b0d57caff8c3d7bd96fe555ea27ea01ff14527a'},
+    {'PyTorch-1.13.1_increase-tolerance-test_ops.patch':
+     'c909fdfc2b12df457e1eb5514265ffec3eab653994949416f3f048668421e223'},
+    {'PyTorch-1.13.1_increase-tolerance-test_optim.patch':
+     'a079d824085eab89794f5ecfc67792f735ed8cfd3fe7db52e4dea62e583cfe06'},
+    {'PyTorch-1.13.1_install-vsx-vec-headers.patch':
+     '7b678f54bb947afd4767f5877ac424b4b94ce5db609ea20f5a869ccf4027035f'},
+    {'PyTorch-1.13.1_no-cuda-stubs-rpath.patch': '4c636059850fc9d1ecb27ce275f8aad5d5b6fdc19e35aff0c25b86cb3201352a'},
+    {'PyTorch-1.13.1_remove-flaky-test-in-testnn.patch':
+     'be83ff61fe2dedab6d49c232936d5622df81ab49154264490021c6c828e53315'},
+    {'PyTorch-1.13.1_skip-failing-grad-test.patch': '6681200f9509893cb9231b5c93ac9bc5e6d9d9ae4febefca52e7cbc843ba8f51'},
+    {'PyTorch-1.13.1_skip-failing-singular-grad-test.patch':
+     '72688a57b2bb617665ad1a1d5e362c5111ae912c10936bb38a089c0204729f48'},
+    {'PyTorch-1.13.1_skip-test_find_unused_parameters-detail.patch':
+     'c71a3385ce5fc447f908a3df78ade2143d97e2538cf03b530db4f6cc8b32c22b'},
+    {'PyTorch-1.13.1_skip-test-requiring-online-access.patch':
+     '61c3b7859dc06a9969981b07aa2789630de110d6d1d3633d27364be47af74712'},
+    {'PyTorch-1.13.1_skip-tests-without-fbgemm.patch':
+     '481e595f673baf8ae58b41697a6792b83048b0264aa79b422f48cd8c22948bb7'},
+    {'PyTorch-1.13.1_workaround-gcc12-destructor-exception-bug.patch':
+     'a09a2d7ebd428c65988729578bb3fa372565ba176ab9ed7abf11f6fcb15e903e'},
+    {'PyTorch-2.0.1_avoid-test_quantization-failures.patch':
+     '02e3f47e4ed1d7d6077e26f1ae50073dc2b20426269930b505f4aefe5d2f33cd'},
+]
+
+osdependencies = [OS_PKG_IBVERBS_DEV]
+
+builddependencies = [
+    ('CMake', '3.24.3'),
+    ('hypothesis', '6.68.2'),
+    # For tests
+    ('pytest-rerunfailures', '12.0'),
+    ('pytest-shard', '0.1.2'),
+]
+
+dependencies = [
+    ('CUDA', '11.7.0', '', SYSTEM),
+    ('Ninja', '1.11.1'),  # Required for JIT compilation of C++ extensions
+    ('Python', '3.10.8'),
+    ('protobuf', '23.0'),
+    ('protobuf-python', '4.23.0'),
+    ('pybind11', '2.10.3'),
+    ('SciPy-bundle', '2023.02'),
+    ('PyYAML', '6.0'),
+    ('MPFR', '4.2.0'),
+    ('GMP', '6.2.1'),
+    ('numactl', '2.0.16'),
+    ('FFmpeg', '5.1.2'),
+    ('Pillow', '9.4.0'),
+    ('cuDNN', '8.5.0.96', '-CUDA-%(cudaver)s', SYSTEM),
+    ('magma', '2.7.1', '-CUDA-%(cudaver)s'),
+    ('NCCL', '2.16.2', '-CUDA-%(cudaver)s'),
+    ('expecttest', '0.1.3'),
+]
+
+custom_opts = ['CMAKE_CUDA_FLAGS=-allow-unsupported-compiler']
+
+excluded_tests = {
+    '': [
+        # This test seems to take too long on NVIDIA Ampere at least.
+        'distributed/test_distributed_spawn',
+        # Broken on CUDA 11.6/11.7: https://github.com/pytorch/pytorch/issues/75375
+        'distributions/test_constraints',
+        # no xdoctest
+        'doctests',
+        # failing on broadwell
+        # See https://github.com/easybuilders/easybuild-easyconfigs/issues/17712
+        'test_native_mha',
+        # intermittent failures on various systems
+        # See https://github.com/easybuilders/easybuild-easyconfigs/issues/17712
+        'distributed/rpc/test_tensorpipe_agent',
+    ]
+}
+
+runtest = 'cd test && PYTHONUNBUFFERED=1 %(python)s run_test.py --continue-through-error  --verbose %(excluded_tests)s'
+
+# The readelf sanity check command can be taken out once the TestRPATH test from
+# https://github.com/pytorch/pytorch/pull/87593 is accepted, since it is then checked as part of the PyTorch test suite
+local_libcaffe2 = "$EBROOTPYTORCH/lib/python%%(pyshortver)s/site-packages/torch/lib/libcaffe2_nvrtc.%s" % SHLIB_EXT
+sanity_check_commands = [
+    "readelf -d %s | egrep 'RPATH|RUNPATH' | grep -v stubs" % local_libcaffe2,
+]
+
+# Especially test_quantization has a few corner cases that are triggered by the random input values,
+# those cannot be easily avoided, see https://github.com/pytorch/pytorch/issues/107030
+# So allow a low number of tests to fail as the tests "usually" succeed
+max_failed_tests = 2
+
+tests = ['PyTorch-check-cpp-extension.py']
+
+moduleclass = 'ai'

easybuild/easyconfigs/p/PyTorch/PyTorch-1.13.1_allow-GCC-12-for-CUDA-11.7.patch

@@ -0,0 +1,29 @@
+The JIT generator for CUDA fails as GCC 12 isn't officially compatible with CUDA 11.7.
+We can make it compatible by passing `-allow-unsupported-compiler`
+but also need to tell the PyTorch code about the raised maximum compiler version.
+
+Author: Alexander Grund (TU Dresden)
+
+diff --git a/torch/utils/cpp_extension.py b/torch/utils/cpp_extension.py
+index 9ab43e5ccdd..15da97619be 100644
+--- a/torch/utils/cpp_extension.py
++++ b/torch/utils/cpp_extension.py
+@@ -56,7 +56,7 @@ CUDA_GCC_VERSIONS = {
+     '11.4': ((6, 0, 0), (11, 5, 0)),
+     '11.5': ((6, 0, 0), (11, 5, 0)),
+     '11.6': ((6, 0, 0), (11, 5, 0)),
+-    '11.7': ((6, 0, 0), (11, 5, 0)),
++    '11.7': ((6, 0, 0), (12, 3, 0)),
+ }
+
+ CUDA_CLANG_VERSIONS = {
+@@ -227,7 +227,8 @@ COMMON_NVCC_FLAGS = [
+     '-D__CUDA_NO_HALF_CONVERSIONS__',
+     '-D__CUDA_NO_BFLOAT16_CONVERSIONS__',
+     '-D__CUDA_NO_HALF2_OPERATORS__',
+-    '--expt-relaxed-constexpr'
++    '--expt-relaxed-constexpr',
++    '-allow-unsupported-compiler',
+ ]
+
+ COMMON_HIP_FLAGS = [

easybuild/easyconfigs/p/PyTorch/PyTorch-1.13.1_skip-test_find_unused_parameters-detail.patch

@@ -0,0 +1,48 @@
+In distributed/test_c10d_nccl the tests
+test_find_unused_parameters_kwarg_debug_detail and
+test_find_unused_parameters_kwarg_grad_is_view_debug_detail
+are failing often on some systems with the root error seemingly being
+> terminate called after throwing an instance of 'c10::Error'
+>  what():  CUDA error: driver shutting down
+
+Stacktrace:
+frame #0: c10::Error::Error(<snip>) + 0x8d (0x2ae861eff2cd in <snip>/torch/lib/libc10.so)
+frame #1: c10::detail::torchCheckFail(<snip>) + 0xd0 (0x2ae861ec64d1 in <snip>/torch/lib/libc10.so)
+frame #2: c10::cuda::c10_cuda_check_implementation(<snip>) + 0x352 (0x2ae861e948c2 in <snip>/torch/lib/libc10_cuda.so)
+frame #3: c10d::ProcessGroupNCCL::WorkNCCL::startedGPUExecutionInternal() const + 0x140 (0x2ae848587e80 in <snip>/torch/lib/libtorch_cuda.so)
+frame #4: c10d::ProcessGroupNCCL::WorkNCCL::isStarted() + 0x58 (0x2ae84858a1b8 in <snip>/torch/lib/libtorch_cuda.so)
+frame #5: c10d::ProcessGroupNCCL::workCleanupLoop() + 0x3c8 (0x2ae84858ee18 in <snip>/torch/lib/libtorch_cuda.so)
+
+Just skip the tests to avoid failing the testsuite.
+
+Author: Alexander Grund (TU Dresden)
+
+diff --git a/test/distributed/test_c10d_nccl.py b/test/distributed/test_c10d_nccl.py
+index 6a0858eebf8..7340a89db82 100644
+--- a/test/distributed/test_c10d_nccl.py
++++ b/test/distributed/test_c10d_nccl.py
+@@ -12,7 +12,7 @@ import time
+ from contextlib import contextmanager
+ from datetime import timedelta
+ from itertools import product
+-from unittest import mock
++from unittest import mock, skip
+
+ import torch
+ import torch.distributed as c10d
+@@ -1460,6 +1460,7 @@ class DistributedDataParallelTest(
+
+     # TODO: Combine the following tests once https://github.com/pytorch/pytorch/issues/55967
+     # is resolved.
++    @skip("Debug level DETAIL fails on some systems/CUDA versions")
+     @requires_nccl()
+     @skip_if_lt_x_gpu(2)
+     @with_dist_debug_levels(levels=["DETAIL"])
+@@ -1478,6 +1479,7 @@ class DistributedDataParallelTest(
+     def test_find_unused_parameters_kwarg_debug_off(self):
+         self._test_find_unused_parameters_kwarg()
+
++    @skip("Debug level DETAIL fails on some systems/CUDA versions")
+     @requires_nccl()
+     @skip_if_lt_x_gpu(2)
+     @with_dist_debug_levels(levels=["DETAIL"])