by

Compiling GAMESS-v2020.2 with Intel MPI

GAMESS Download Site can be found at https://www.msg.chem.iastate.edu/GAMESS/download/dist.source.shtml

Compiling GAMESS

% tar -zxvf gamess-current.tar.gz
% cd gamess
% ./config

You have to answer the following question on

  • Machine Type? – I chose “linux64“
  • GAMESS directory? -I chose “/usr/local/gamess“
  • GAMESS Build Directory – I chose “/usr/local/gamess“
  • Version? [00] – I chose default [00]
  • Choice of Fortran Compilers – I chose “ifort”
  • Version Number of ifort – I chose “18” (You can check by issuing the command ifort -V)
  • Standard Math Library – I chose “mkl”
  • Path of MKL – I chose “/usr/local/intel/2018u3/compilers_and_libraries_2018.3.222/linux/mkl“
  • Type “Proceed” next
  • Communication Library – I chose “mpi” (I’m using Infiniband)
  • Enter MPI Library – I chose “impi“
  • Enter Location of impi – I chose “/usr/local/intel/2018u3/impi/2018.3.222“
  • Build experimental support of LibXC – I chose “no“
  • Build Beta Version of Active Space CCT3 and CCSD3A – I chose “no“
  • Build LIBCCHEM – I chose “no“
  • Build GAMESS with OpenMP thread support – I chose “yes”

Once done, you should see

Your configuration for GAMESS compilation is now in
     /usr/local/gamess/install.info
Now, please follow the directions in
     /usr/local/gamess/machines/readme.unix

Compiling ddi

Edit DDI Node Sizes by editing /usr/local/gamess/ddi/compddi
Look at Line 90 and 91. You may want to edit MAXCPUS and MAXNODES. Once done, you can compile ddi

% ./compddi >& compddi.log &

Compiling GAMESS

The compilation will take a while. So relax…..

% ./compall >& compall.log &

Linking Executable Form of GAMESS with the command

./lked gamess 01 >& lked.log &

Edit the Scratch Directory setting at rungms

% vim rungms
set SCR=/scratch/$USER
by

Finding Top Processes using Highest Memory and CPU Usage in Linux

Read this Article from Find Top Running Processes by Highest Memory and CPU Usage in Linux. This is a quick way to view processes that consumed the largest RAM and CPU

ps -eo pid,ppid,cmd,%mem,%cpu --sort=-%mem | head
   PID   PPID CMD                         %MEM %CPU
414699 414695 /usr/local/ansys_inc/v201/f 20.4 98.8
 30371      1 /usr/local/pbsworks/pbs_acc  0.2  1.0
 32241      1 /usr/local/pbsworks/pbs_acc  0.2  4.0
 30222      1 /usr/local/pbsworks/pbs_acc  0.2  0.6
  7191      1 /usr/local/pbsworks/dm_exec  0.1  0.8
 30595      1 /usr/local/pbsworks/pbs_acc  0.1  3.1
 30013      1 /usr/local/pbsworks/pbs_acc  0.1  0.3
 29602  29599 nginx: worker process        0.1  0.2
 29601  29599 nginx: worker process        0.1  0.3

The -o is to specify the output format. The -e is to select all processes. In order to sort in descending format, it hsould be –sort=–%mem

Interesting.

by

Storage Performance Basics for Deep Learning

This is an interesting write-up from James Mauro from Nvidia on Storage Performance Basics for Deep Learning.

“The complexity of the workloads plus the volume of data required to feed deep-learning training creates a challenging performance environment. Deep learning workloads cut across a broad array of data sources (images, binary data, etc), imposing different disk IO load attributes, depending on the model and a myriad of parameters and variables.”

For Further Reads… Do take a look at https://developer.nvidia.com/blog/storage-performance-basics-for-deep-learning/

by

GTC 2021 Keynote with NVIDIA CEO Jensen Huang

NVIDIA CEO Jensen announced NVIDIA’s first data center CPU, Grace, named after Grace Hopper, a U.S. Navy rear admiral and computer programming pioneer. Grace is a highly specialized processor targeting largest data intensive HPC and AI applications as the training of next-generation natural-language processing models that have more than one trillion parameters.

Further accelerating the infrastructure upon which hyperscale data centers, workstations, and supercomputers are built, Huang announced the NVIDIA BlueField-3 DPU.

The next-generation data processing unit will deliver the most powerful software-defined networking, storage and cybersecurity acceleration capabilities.

Where BlueField-2 offloaded the equivalent of 30 CPU cores, it would take 300 CPU cores to secure, offload, and accelerate network traffic at 400 Gbps as BlueField-3— a 10x leap in performance, Huang explained.