diff --git a/doc.zih.tu-dresden.de/docs/archive/knl_nodes.md b/doc.zih.tu-dresden.de/docs/archive/knl_nodes.md index 78e4cabc7b4f40574e084834d69175cdd9fa29ef..ff7d4d9168e0a904e3e7258a18236cc65a9ea27b 100644 --- a/doc.zih.tu-dresden.de/docs/archive/knl_nodes.md +++ b/doc.zih.tu-dresden.de/docs/archive/knl_nodes.md @@ -1,57 +1,53 @@ # Intel Xeon Phi (Knights Landing) -Xeon Phi nodes are **Out of Service**! +!!! warning + + This page is deprceated. The Xeon Phi nodes are **out of service**. The nodes `taurusknl[1-32]` are equipped with -- Intel Xeon Phi procesors: 64 cores Intel Xeon Phi 7210 (1,3 GHz) +- Intel Xeon Phi processors: 64 cores Intel Xeon Phi 7210 (1,3 GHz) - 96 GB RAM DDR4 - 16 GB MCDRAM - `/scratch`, `/lustre/ssd`, `/projects`, `/home` are mounted Benchmarks, so far (single node): -- HPL (Linpack): 1863.74 GFlops -- SGEMM (single precision) MKL: 4314 GFlops +- HPL (LINPACK): 1863.74 GFLOPS +- SGEMM (single precision) MKL: 4314 GFLOPS - Stream (only 1.4 GiB memory used): 431 GB/s Each of them can run 4 threads, so one can start a job here with e.g. -```Bash -srun -p knl -N 1 --mem=90000 -n 1 -c 64 a.out +```console +marie@login$ srun -p knl -N 1 --mem=90000 -n 1 -c 64 a.out ``` -In order to get their optimal performance please re-compile your code -with the most recent Intel compiler and explicitely set the compiler -flag `-xMIC-AVX512`. +In order to get their optimal performance please re-compile your code with the most recent Intel +compiler and explicitly set the compiler flag `-xMIC-AVX512`. -MPI works now, we recommend to use the latest Intel MPI version -(intelmpi/2017.1.132). To utilize the OmniPath Fabric properly, make -sure to use the "ofi" fabric provider, which is the new default set by -the module file. +MPI works now, we recommend to use the latest Intel MPI version (intelmpi/2017.1.132). To utilize +the OmniPath Fabric properly, make sure to use the "ofi" fabric provider, which is the new default +set by the module file. -Most nodes have a fixed configuration for cluster mode (Quadrant) and -memory mode (Cache). For testing purposes, we have configured a few -nodes with different modes (other configurations are possible upon -request): +Most nodes have a fixed configuration for cluster mode (Quadrant) and memory mode (Cache). For +testing purposes, we have configured a few nodes with different modes (other configurations are +possible upon request): | Nodes | Cluster Mode | Memory Mode | |:-------------------|:-------------|:------------| | `taurusknl[1-28]` | Quadrant | Cache | -| `taurusknl29` | Quadrant | Flat | +| `taurusknl29` | Quadrant | Flat | | `taurusknl[30-32]` | SNC4 | Flat | -They have SLURM features set, so that you can request them specifically -by using the SLURM parameter `--constraint` where multiple values can -be linked with the & operator, e.g. `--constraint="SNC4&Flat"`. If you -don't set a constraint, your job will run preferably on the nodes with -Quadrant+Cache. +They have Slurm features set, so that you can request them specifically by using the Slurm parameter +`--constraint` where multiple values can be linked with the & operator, e.g. +`--constraint="SNC4&Flat"`. If you don't set a constraint, your job will run preferably on the nodes +with Quadrant+Cache. -Note that your performance might take a hit if your code is not -NUMA-aware and does not make use of the Flat memory mode while running -on the nodes that have those modes set, so you might want to use -`--constraint="Quadrant&Cache"` in such a case to ensure your job does not -run on an unfavorable node (which might happen if all the others are -already allocated). +Note that your performance might take a hit if your code is not NUMA-aware and does not make use of +the Flat memory mode while running on the nodes that have those modes set, so you might want to use +`--constraint="Quadrant&Cache"` in such a case to ensure your job does not run on an unfavorable +node (which might happen if all the others are already allocated). -[Knl Best Practice Guide](https://prace-ri.eu/training-support/best-practice-guides/best-practice-guide-knights-landing/) +[KNL Best Practice Guide](https://prace-ri.eu/training-support/best-practice-guides/best-practice-guide-knights-landing/) diff --git a/doc.zih.tu-dresden.de/mkdocs.yml b/doc.zih.tu-dresden.de/mkdocs.yml index fe7e7459dc9da48dea514805706f4d7d7972c813..6069771f5ceefef39881088750272ef074104b11 100644 --- a/doc.zih.tu-dresden.de/mkdocs.yml +++ b/doc.zih.tu-dresden.de/mkdocs.yml @@ -128,6 +128,7 @@ nav: - System Titan: archive/system_titan.md - System Triton: archive/system_triton.md - System Venus: archive/system_venus.md + - KNL Nodes: archive/knl_nodes.md - UNICORE Rest API: archive/unicore_rest_api.md - Vampir Trace: archive/vampir_trace.md - Windows Batchjobs: archive/windows_batch.md diff --git a/doc.zih.tu-dresden.de/wordlist.aspell b/doc.zih.tu-dresden.de/wordlist.aspell index ab4da040e64199ed321389665808c9fefe255807..011c877881dd378b7fdf3f0aabfe30dc20cd8243 100644 --- a/doc.zih.tu-dresden.de/wordlist.aspell +++ b/doc.zih.tu-dresden.de/wordlist.aspell @@ -10,6 +10,7 @@ CPU CPUs CUDA CXFS +DDR DFG EasyBuild fastfs @@ -19,6 +20,7 @@ Flink Fortran GFLOPS gfortran +GiB gnuplot Gnuplot GPU @@ -27,6 +29,7 @@ Haswell HDFS Horovod HPC +HPL icc icpc ifort @@ -36,6 +39,8 @@ Itanium jpg Jupyter Keras +KNL +LINPACK LoadLeveler lsf LSF @@ -82,6 +87,7 @@ scancel scontrol scp SGI +SGEMM SHA SHMEM SLES @@ -100,4 +106,5 @@ Theano tmp Trition Vampir +Xeon ZIH