October 2021 - Ecp-aurora-sdk-discuss

New Aurora SDK available in JLSE
by Chan-nui, Christopher 28 Oct '21

28 Oct '21

Hello, We are pleased to announce that a new Aurora SDK has been deployed in JLSE. This version oneapi/release/2021.10.30.001 Additionally oneapi/eng-compiler/2021.10.30.002 includes engineering drops of the compilers, MKL is NOT supported with the engineering compilers pending an updated release. THERE IS AN IMPORTANT CHANGE IN MODULE NAMING We are now separating out the types of releases by name in order to make it clearer what is contained in each release. For example, to load the latest full release use 'oneapi/release', to load the latest engineering compilers 'oneapi/eng-compiler'. oneapi/<TYPE> loads the default module for this type of OneAPI SDK (Tested with Argonne Testset at the time of release) oneapi/<TYPE>/<VER> loads specific version of OneAPI SDK Version (Tested with Argonne Testset at the time of release) oneapi/<TYPE>/latest loads the default for the type of OneAPI SDK+Latest UMD (Untested) Currently defined types are: release Use this channel for the OneAPI SDK with MKL and is considered the most stable version you can get. eng-compiler Bleeding edge compilers for those that are waiting for fixes on compiler bugs not solved by 'release' channel. tile-as-device Tile as device module supported only in OpenMP IMPORTANT INFORMATION ABOUT GPU SCALING Please take note of the type of behavior you are trying to offload to the GPU, implicit scaling (Default) : Uses both the tiles of the GPU to scale the workload. ZE_AFFINITY_MASK will not work and one might observe hangs if these options are combined. explicit scaling export EnableWalkerPartition=-1 : Set this if your workload is designed to use each tile of the GPU separately. or you want to use only of the tiles of the GPU via ZE_AFFINITY_MASK.Use the oneapi module to load the appropriate drivers for your system. For additional details please refer to the latest release notes in the Wiki: https://wiki.jlse.anl.gov/display/inteldga/DGA+SDK+Release+Notes Public Documentation ==================== * ALCF oneAPI overview - https://www.alcf.anl.gov/support-center/aurora/oneapi-overview * Intel GPGPU Software Information - https://dgpu-docs.intel.com/ * Intel oneAPI Documentation - https://software.intel.com/content/www/us/en/develop/tools/oneapi/documenta… Please let us know of any issues with the release or the environment settings. -- Christopher Chan-Nui christopher.chan-nui(a)intel.com

1 0

Upcoming oneapi update
by Chan-nui, Christopher 28 Oct '21

28 Oct '21

At 5pm CDT the oneapi will be updated. Running jobs will not be affected. Default modulefiles will be updated. -- Christopher Chan-Nui christopher.chan-nui(a)intel.com

1 0

New Release of the Deep Learning Frameworks now available in JLSE
by Ramer, Rebecca 08 Oct '21

08 Oct '21

Hi, We would like to announce availability of new frameworks module for pytorch and tensorflow for use in the JLSE. This release is based on the 2021.3.1 NDA version of the AiKit. The modules support single tile runs and distributed training runs using horovod. Information on how to setup and work with this release can be found here for pytorch: https://wiki.jlse.anl.gov/display/inteldga/AIKit+NDA+-+2021.3.1 and here for tensorflow: https://wiki.jlse.anl.gov/display/inteldga/AIKit+NDA+-+2021.3.1+-+Tensorflow At this time I am still updating the run details for the Tensorflow workloads, I should be completed shortly. Please let us know of any issues encountered. Thanks, Rebecca Ramer rebecca.ramer(a)intel.com<mailto:[email protected]>

1 0