System Updates

2021-07-06

Add / Update / Delete	Software	Version	Previous version
Update	SingularityPRO	3.7-4	3.7-1

2021-06-30

Add / Update / Delete	Software	Version	Previous version
Add	cuDNN	8.2.1
Add	NCCL	2.9.9-1

2021-05-10

Add / Update / Delete	Software	Version	Previous version
Add	cuDNN	8.2.0
Add	NCCL	2.9.6-1

The documentation has been revised with the addition of a compute node (A) with NVIDIA A100.

2021-04-07

Add / Update / Delete	Software	Version	Previous version
Update	NVIDIA Tesla Driver	460.32.03	440.33.01
Update	OFED	5.0-2.1.8.0	4.4-1.0.0.0
Update	Univa Grid Engine	8.6.17	8.6.6
Update	SingularityPRO	3.7-1	3.5-6
Update	BeeOND	7.2.1	7.2
Update	Docker	19.03.15	17.12.0
Update	Scality S3 Connector	7.4.8.1	7.4.8
Add	gcc	9.3.0
Add	pgi	20.4
Add	nvhpc	20.11 21.2
Add	cmake	3.19
Add	go	1.15
Add	julia	1.5
Add	lua	5.3.6 5.4.2
Add	python	2.7.18 3.6.12 3.7.10 3.8.7
Add	R	4.0.4
Add	CUDA	11.0.3 11.1.1 11.2.2
Add	cuDNN	8.1.1
Add	NCCL	2.8.4-1
Add	openmpi	4.0.5
Add	mvapich2-gdr	2.3.5
Add	mvapich2	2.3.5
Add	hadoop	3.3
Add	spark	3.0
Add	aws-cli	2.1
Add	fuse-sshfs	3.7.1
Add	s3fs-fuse	1.87
Add	sregistry-cli	0.2.36
Delete	intel	2017.8.262 2018.5.274 2019.5.281
Delete	pgi	17.10 18.10 19.1 19.10 20.1
Delete	nvhpc	20.9
Delete	cmake	3.16 3.17
Delete	go	1.12 1.13
Delete	intel-advisor	2017.5 2018.4 2019.5
Delete	intel-inspector	2017.4 2018.4 2019.5
Delete	intel-itac	2017.0.4 2018.0.4 2019.0.5
Delete	intel-mkl	2017.0.4 2018.0.4 2019.0.5
Delete	intel-vtune	2017.6 2018.4 2019.6
Delete	julia	1.3 1.4
Delete	python	2.7.15 3.4.8 3.5.5 3.7.6
Delete	R	3.5.0 3.6.3
Delete	cuda	10.0.130
Delete	cudnn	7.1.3 7.5.0 7.6.0 7.6.1 7.6.2 7.6.3 7.6.4 8.0.2
Delete	nccl	2.3.4-1 2.3.5-2 2.4.2-1 2.4.7-1 2.8.3-1
Delete	intel-mpi	2017.4 2018.4 2019.5
Delete	openmpi	4.0.3
Delete	mvapich2-gdr	2.3.3 2.3.4
Delete	mvapich2	2.3.3 2.3.4
Delete	hadoop	2.9 2.10 3.1
Delete	singularity	2.6.1
Delete	spark	2.3 2.4
Delete	aws-cli	1.16.194 1.18 2.0
Delete	fuse-sshfs	2.10
Delete	s3fs-fuse	1.85
Delete	sregistry-cli	0.2.31

2021-03-13

Add / Update / Delete	Software	Version	Previous version
Update	SingularityPRO	3.5-6	3.5-4
Update	DDN Lustre	2.12.6_ddn13-1	2.10.7_ddn14-1

2020-12-15

Add / Update / Delete	Software	Version	Previous version
Add	go	1.14
Add	intel	2020.4.304
Add	intel-advisor	2020.3
Add	intel-inspector	2020.3
Add	intel-itac	2020.0.3
Add	intel-mkl	2020.0.4
Add	intel-mpi	2019.9
Add	intel-vtune	2020.3
Add	nvhpc	20.9
Add	cuDNN	8.0.5
Add	NCCL	2.8.3-1
Update	BeeOND	7.2	7.1.5
Update	Scality S3 Connector	7.4.8	7.4.6.3

Additional Feature: SSH Access to Compute Nodes

We have added the feature to enable SSH login to the compute nodes. See SSH Access to Compute Nodes for details.

2020-10-09

Add / Update / Delete	Software	Version	Previous version
Update	SingularityPRO	3.5-4	3.5-2

2020-08-31

Add / Update / Delete	Software	Version	Previous version
Update	Scality S3 Connector	7.4.6.3	7.4.5.4

2020-07-31

Add / Update / Delete	Software	Version
Add	SingularityPRO	3.5-2
Add	cuDNN	8.0.2
Add	NCCL	2.7.8-1
Add	mvapich2-gdr	2.3.4
Add	mvapich2	2.3.4

2020-06-01

Add / Update / Delete	Software	Version	Previous version
Update	BeeOND	7.1.5	7.1.4

2020-04-21

Update MVAPICH2-GDR 2.3.3

MVAPICH2-GDR 2.3.3 for gcc 4.8.5 was updated to the fixed version about the following issue.

MPI_Allreduce provided by MVAPICH2-GDR may raise floating point exceptions

On the other hand, MVAPICH2-GDR 2.3.3 for PGI was uninstalled. If you need MVAPICH2-GDR for PGI, please contact Customer Support.

2020-04-03

Add / Update / Delete	Software	Version	Previous version
Update	DDN GRIDScaler	4.2.3.20	4.2.3.17
Update	Scality S3 Connector	7.4.5.4	7.4.5.0
Update	libfabric	1.7.0-1	1.5.3-1
Add	intel	2018.5.274 2019.5.281
Add	pgi	19.1 19.10 20.1
Add	R	3.6.3
Add	cmake	3.16 3.17
Add	go	1.12 1.13
Add	intel-advisor	2017.5 2018.4 2019.5
Add	intel-inspector	2017.4 2018.4 2019.5
Add	intel-itac	2017.0.4 2018.0.4 2019.0.5
Add	intel-mkl	2017.0.4 2018.0.4 2019.0.5
Add	intel-vtune	2017.6 2018.4 2019.6
Add	julia	1.0 1.3 1.4
Add	openjdk	1.8.0.242 11.0.6.10
Add	python	3.7.6 3.8.2
Add	gdrcopy	2.0
Add	nccl	2.6.4-1
Add	intel-mpi	2017.4 2018.4 2019.5
Add	mvapich2-gdr	2.3.3
Add	mvapich2	2.3.3
Add	openmpi	3.1.6 4.0.3
Add	hadoop	2.9 2.10 3.1
Add	spark	2.3 2.4
Add	aws-cli	1.18 2.0
Delete	gcc	7.3.0
Delete	intel	2018.2.199 2018.3.222 2019.3.199
Delete	pgi	18.5 19.3
Delete	go	1.11.2
Delete	intel-mkl	2017.8.262 2018.2.199 2018.3.222 2019.3.199
Delete	openjdk	1.6.0.41 1.8.0.161
Delete	cuda	9.0/9.0.176.2 9.0/9.0.176.3
Delete	gdrcopy	1.2
Delete	intel-mpi	2018.2.199
Delete	mvapich2-gdr	2.3rc1 2.3 2.3a 2.3.1 2.3.2
Delete	mvapich2	2.3rc2 2.3 2.3.2
Delete	openmpi	1.10.7 2.1.3 2.1.5 3.0.3 3.1.0 3.1.2 3.1.3
Delete	hadoop	2.9.1 2.9.2
Delete	spark	2.3.1 2.3.2 2.4.0

2019-12-17

Add / Update / Delete	Software	Version	Previous version
Update	DDN Lustre	2.10.7_ddn14-1	2.10.5_ddn7-1
Update	BeeOND	7.1.4	7.1.3
Update	Scality S3 Connector	7.4.5.0	7.4.4.4
Update	NVIDIA Tesla Driver	440.33.01	410.104
Add	CUDA	10.2.89
Add	cuDNN	7.6.5
Add	NCCL	2.5.6-1

Other fixes are as follows:

Add Memory-Intensive Node

2019-11-06

Add / Update / Delete	Software	Version	Previous version
Add	GCC	7.3.0, 7.4.0
Add	sregistry-cli	0.2.31

Other fixes are as follows:

Fixed cuda/* modules to set the paths to extras/CUPTI.
Fixed python/3.4, python/3.5, and python/3.6 to solve the problem that error occurred when executing shutil.copytree on Home area.

2019-10-04

Add / Update / Delete	Software	Version	Previous version
Update	Univa Grid Engine	8.6.6	8.6.3
Update	DDN GRIDScaler	4.2.3.17	4.2.3.15
Update	BeeOND	7.1.3	7.1.2
Add	CUDA	10.1.243
Add	cuDNN	7.6.3 7.6.4
Add	NCCL	2.4.8-1
Add	MVAPICH2-GDR	2.3.2
Add	MVAPICH2	2.3.2
Add	fuse-sshfs	2.10

Other fixes are as follows:

Add CUDA 10.1 support to cuDNN 7.5.0, 7.5.1, 7.6.0, 7.6.1, 7.6.2
Add CUDA 10.1 support to NCCL 2.4.2-1, 2.4.7-1
Add CUDA 10.0 and 10.1 support to GDRCopy 1.2
Add CUDA 10.1 support to Open MPI 2.1.6
Increase /tmp capacity of interactive nodes from 26GB to 12TB
Add process monitoring and process cancellation mechanism on the interactive node

Start process monitoring on the interactive nodes

Process monitoring started on the interactive nodes. High load or lengthy tasks on the interactive nodes will be killed by the process monitoring system, so use the compute nodes with the qrsh/qsub command.

Change the job submission and execution limits

We changed the job submission and execution limits as follows.

Limitations	Current limits	Previous limits
The maximum number of tasks within an array job	75000	1000
The maximum number of any user's running jobs at the same time	200	0(unlimited)

About known issues

The status of following known issues were changed to close.

A comupte node can execute only up to 2 jobs each resource type "rt_G.small" and "rt_C.small" (normally up to 4 jobs ).

2019-08-01

Add / Update / Delete	Software	Version
Add	cuDNN	7.6.2
Add	NCCL	2.4.7-1
Add	s3fs-fuse	1.85

Other fixes are as follows:

Add CUDA 10.0 support to Open MPI 1.10.7, 2.1.5, 2.1.6

2019-07-10

Add / Update / Delete	Software	Version
Add	CUDA	10.0.130.1
Add	cuDNN	7.5.1 7.6.0 7.6.1
Add	aws-cli	1.16.194

2019-04-05

Add / Update / Delete	Software	Version	Previous version
Update	CentOS	7.5	7.4
Update	Univa Grid Engine	8.6.3	8.5.4
Update	Java	1.7.0_171	1.7.0_141
Update	Java	1.8.0_161	1.8.0_131
Add	DDN Lustre	2.10.5_ddn7-1
Update	NVIDIA Tesla Driver	410.104	396.44
Add	CUDA	10.0.130
Add	Intel Compiler	2019.3
Add	PGI	18.10 19.3

Other fixes are as follows:

Migrate Home area from GPFS to DDN Lustre

2019-03-14

Add / Update / Delete	Software	Version
Add	Intel Compiler	2017.8 2018.3
Add	PGI	17.10
Add	Open MPI	2.1.6
Add	cuDNN	7.5.0
Add	NCCL	2.4.2-1
Add	Intel MKL	2017.8 2018.3

Other fixes are as follows:

Add PGI 17.10 support to MVAPICH2-GDR 2.3
Add PGI support to Open MPI 2.1.5, 2.1.6, 3.1.3
Change the default version of Open MPI to 2.1.6
Fix typo in MVAPICH2 modules, wrong top directory

2019-01-31

User/Group/Job names are now masked when displaying the result of 'qstat'

We changed the job scheduler configuration, so that User/Group/Job names are masked from the result of qstat command. These columns are shown only for your own jobs, otherwise these columns are masked by '*'. An example follows:

[username@es1 ~]$ qstat -u '*' | head
job-ID     prior   name       user         state submit/start at     queue                          jclass                         slots ja-task-ID
------------------------------------------------------------------------------------------------------------------------------------------------
    123456 0.28027 run.sh     username     r     01/31/2019 12:34:56 gpu@g0001                                                        80
    123457 0.28027 ********** **********   r     01/31/2019 12:34:56 gpu@g0002                                                        80
    123458 0.28027 ********** **********   r     01/31/2019 12:34:56 gpu@g0003                                                        80
    123450 0.28027 ********** **********   r     01/31/2019 12:34:56 gpu@g0004                                                        80

2018-12-18

Add / Update / Delete	Software	Version
Add	cuDNN	7.4.2
Add	NCCL	2.3.7-1
Add	Open MPI	3.0.3 3.1.3
Add	MVAPICH2-GDR	2.3
Add	Hadoop	2.9.2
Add	Spark	2.3.2 2.4.0
Add	Go	1.11.2
Add	Intel MKL	2018.2.199

cuDNN 7.4.2

The NVIDIA CUDA Deep Neural Network library (cuDNN) 7.4.2 was installed.

To set up user environment:

$ module load cuda/9.2/9.2.148.1
$ module load cudnn/7.4/7.4.2

NCCL 2.3.7-1

The NVIDIA Collective Communications Library (NCCL) 2.3.7-1 was installed.

To set up user environment:

$ module load cuda/9.2/9.2.148.1
$ module load nccl/2.3/2.3.7-1

Open MPI 3.0.3, 3.1.3

Open MPI (without --cuda option) 3.0.3, 3.1.3 were installed.

To set up user environment:

$ module load openmpi/3.1.3

MVAPICH2-GDR 2.3

MVAPICH2-GDR 2.3 was installed.

To set up user environment:

$ module load cuda/9.2/9.2.148.1
$ module load mvapich/mvapich2-gdr/2.3

Hadoop 2.9.2

Apache Hadoop 2.9.2 was installed.

To set up user environment:

$ module load openjdk/1.8.0.131
$ module load hadoop/2.9.1

Spark 2.3.2, 2.4.0

Apache Spark 2.3.2, 2.4.0 were installed.

To set up user environment:

$ module load spark/2.4.0

Go 1.11.2

Go Programming Language 1.11.2 was installed.

To set up user environment:

$ module load go/1.11.2

Intel MKL 2018.2.199

Intel Math Kernel Library (MKL) 2018.2.199 was installed.

To set up user environment:

$ module load intel-mkl/2018.2.199

2018-12-14

Add / Update / Delete	Software	Version	Previous version
Update	Singularity	2.6.1	2.6.0
Delete	Singularity	2.5.2

Singularity 2.6.1 was installed. The usage is as follows:

$ module load singularity/2.6.1
$ singularity run image_path

The release note will be found:

Singularity 2.6.1

And, we uninstalled version 2.5.2 and 2.6.0 because severe security issues (CVE-2018-19295) were reported. If you are using Singularity with specifying version number, such as singularity/2.5.0 or singularity/2.6.0, please modify your job scripts to specify singularity/2.6.1.

ex) module load singularity/2.5.2 -> module load singularity/2.6.1