Skip to content

Commit 4267c4f

Browse files
committed
Merge branch 'fixConfigDoc' into develop
2 parents a2a0189 + 46d1973 commit 4267c4f

1 file changed

Lines changed: 32 additions & 8 deletions

File tree

doc/source/reference/makefile.rst

Lines changed: 32 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -130,32 +130,56 @@ Finally, *Idefix* can be configured to run on Mi250 by enabling HIP and the desi
130130
131131
MPI (multi-GPU) can be enabled by adding ``-DIdefix_MPI=ON`` as usual.
132132

133-
Jean Zay at IDRIS, Nvidia V100 and A100 GPUs
134-
--------------------------------------------
133+
Jean Zay at IDRIS, Nvidia V100/A100/H100 GPUs
134+
---------------------------------------------
135135

136-
We recommend the following modules and environement variables on Jean Zay:
136+
We recommend the following modules and environement variables on Jean Zay V100/A100:
137137

138138
.. code-block:: bash
139139
140+
module load arch/a100 # ONLY forA100
140141
module load cuda/12.1.0
141142
module load gcc/12.2.0
142143
module load openmpi/4.1.1-cuda
143-
module load cmake/3.18.0
144+
module load cmake/3.25.2
145+
146+
While for H100:
147+
148+
.. code-block:: bash
149+
150+
module load arch/h100
151+
module load cmake/3.30.1
152+
module load cuda/12.1.0
153+
module load openmpi/4.1.5-cuda
144154
145155
*Idefix* can then be configured to run on Nvidia V100 with the following options to ccmake:
146156

147157
.. code-block:: bash
148158
149-
-DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_VOLTA70=ON -DKokkos_ENABLE_IMPL_CUDA_MALLOC_ASYNC=OFF
159+
-DKokkos_ENABLE_CUDA=ON -DKokkos_ARCH_VOLTA70=ON
150160
151161
While Ampere A100 GPUs are enabled with
152162

153163
.. code-block:: bash
154164
155-
-DKokkos_ENABLE_CUDA=ON -DKokkos_ENABLE_AMPERE80=ON -DKokkos_ENABLE_IMPL_CUDA_MALLOC_ASYNC=OFF
165+
-DKokkos_ENABLE_CUDA=ON -DKokkos_ARCH_AMPERE80=ON
166+
167+
And for H100 GPUS:
168+
169+
.. code-block:: bash
170+
171+
-DKokkos_ENABLE_CUDA=ON -DKokkos_ARCH_HOPPER90=ON
172+
173+
174+
MPI (multi-GPU) can be enabled by adding ``-DIdefix_MPI=ON`` as usual.
175+
176+
177+
.. warning::
178+
179+
As of *Idefix* 2.1.02, we automatically disable Cuda Malloc async (``-DKokkos_ENABLE_IMPL_CUDA_MALLOC_ASYNC=OFF``). However, earlier versions of
180+
*Idefix* requires this flag when calling cmake to prevent a bug when using PSM2 with async Cuda malloc possibly leading to openmpi crash or hangs on Jean Zay.
181+
156182

157-
MPI (multi-GPU) can be enabled by adding ``-DIdefix_MPI=ON`` as usual. The malloc async option is here to prevent a bug when using PSM2 with async
158-
Cuda malloc possibly leading to openmpi crash or hangs on Jean Zay.
159183

160184
.. _setupSpecificOptions:
161185

0 commit comments

Comments
 (0)