Versions Compared

Old Version 30

changes.mady.by.user Indy Siva

Saved on

compared with

New Version Current

changes.mady.by.user Indy Siva

Saved on

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

Table of Contents
 


Introduction

...


This is a special node with a separate batching system

...

be listed in the group "aspen" and must use this for accounting in their scripts: #PBS -W group_list=aspen -A aspen 


Special note

...


  1. There is a space on /lscratch for each user. This is a fast SSD and hence it would be advantageous to copy the data to this folder and run the job from it.

    No Format
    As the home directory is shared across all nodes, you can transfer files first to gowonda and you will see it in your home directory on all nodes including n060. If you need to use the local scratch on n060 (it is not shared with gowonda), then move the folder or files from your home directory to your /lscratch/snumber . For performance, it is best to use /scratch space for all computation.

e.g: on n060, run this: mv mydataFolder /lscratch/snumber
  2. There are 5 GPUs for dljun queue and 1 GPu for use by dlyaq queue. To use this, use attribute: ngupus=1 together with the queue name (see sample pbs script above). All jobs will be queued and when a resource becomes available, it will be run on that queue.
  3. There is a space /project/deeplearning for all members of the group "deeplearning". And there is a space /project/aspen for all members of group "aspen"

...


No Format
qstat -q

server: n060

Queue            Memory CPU Time Walltime Node   Run   Que   Lm  State
---------------- ------ -------- -------- ---- ----- ----- ----  -----
workq              --      --       --     --      0     0   --   D R
dl                 --      --    100:00:0    1     0     0   --   D R
dljun              --      --    300:00:0    1     0     0   --   E R
dlyao              --      --    300:00:0    1     0     0   --   E R
                                               ----- -----
                                                   0     0

...


Quick run for testing
 No Format
qsub -q dljun@n060 -W group_list=deeplearning -A deeplearning  -- /bin/date
qstat -1an @n060


...

Sample Interactive PBS run
 No Format
Interactive run all queues have been disabled except on iworkq
Usage is as follows:
qsub -I -q dljun@n060iworkq -W group_list=deeplearning -A deeplearning
OR
qsub -I -q dlyao@n060iworkq  -W group_list=aspen -A aspen
 
...


Sample pbs script: To use in queue dljun
 No Format
cat sample.pbs.script-dljun to run on queue named dlyao

#!/bin/bash -l
#PBS -m abe
## Mail to user
#PBS -M YourEmail@griffith.edu.au
#PBS -V
## Job name
#PBS -N  JunTest
#PBS -q dljun@n060
#####PBS -q dlyao@n060
####Other options #PBS -q dlyao@n060 or #PBS -q workq@n060
#PBS -W group_list=deeplearning -A deeplearning
###Other options group_list=aspen -A aspen
### Number of nodes:Number of CPUs:Number of threads per node
#PBS -l select=1:ncpus=1:ngpus=1:mem=12gb,walltime=100:00:00
#PBS -j oe
### Add current shell environment to job (comment out if not needed)
#PBS -V
# The job's working directory
echo Working directory is $PBS_O_WORKDIR
cd $PBS_O_WORKDIR
source $HOME/.bashrc
module list
echo "Starting job"
echo Running on host `hostname`
echo Time is `date`
echo Directory is `pwd`
gpustat
nvidia-smi
echo "Done with job"


Another Sample pbs script: To use in queue dlyao
...

 No Format
#!/bin/bash -l
#PBS -m abe
## Mail to user
#PBS -M YOURNAME@griffith.edu.au
#PBS -V
## Job name
#PBS -N  YaoJobMyName
#PBS -q dlyao@n060
####Other options #PBS -q dlyao@n060 or #PBS -q workq@n060
#PBS -W group_list=aspen -A aspen
##### Other option##s PBS -W group_list=deeplearning -A deeplearning
###Other options group_list=aspen -A aspen
### Number of nodes:Number of CPUs:Number of threads per node
#PBS -l select=1:ncpus=1:ngpus=1:mem=12gb,walltime=100:00:00
#PBS -j oe
### Add current shell environment to job (comment out if not needed)
#PBS -V
# The job's working directory
echo Working directory is $PBS_O_WORKDIR
cd $PBS_O_WORKDIR
source $HOME/.bashrc
module list
echo "Starting job"
echo Running on host `hostname`
echo Time is `date`
echo Directory is `pwd`
gpustat
nvidia-smi
sleep 100
echo "Done with job"
 
...


Specifications
Hardware: HPE Proliant HPE XL270d Gen 10 Node CTO server, 
Intel(R) Xeon(R) Gold 6140 CPU @ 2.30GHz
The OS is Centos 7.6 and the batching system is PBS 18.2
This node has 6 nvidia GPU cards (HPE NVIDIA Tesla V100-32GB PCle) 

 No Format
 nvidia-smi
Wed Dec 12 08:28:49 2018
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.79       Driver Version: 410.79       CUDA Version: 10.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Tesla V100-PCIE...  On   | 00000000:14:00.0 Off |                    0 |
| N/A   32C    P0    26W / 250W |      0MiB / 32480MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   1  Tesla V100-PCIE...  On   | 00000000:15:00.0 Off |                    0 |
| N/A   33C    P0    25W / 250W |      0MiB / 32480MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   2  Tesla V100-PCIE...  On   | 00000000:39:00.0 Off |                    0 |
| N/A   33C    P0    25W / 250W |      0MiB / 32480MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   3  Tesla V100-PCIE...  On   | 00000000:3A:00.0 Off |                    0 |
| N/A   33C    P0    28W / 250W |      0MiB / 32480MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   4  Tesla V100-PCIE...  On   | 00000000:88:00.0 Off |                    0 |
| N/A   34C    P0    27W / 250W |      0MiB / 32480MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   5  Tesla V100-PCIE...  On   | 00000000:89:00.0 Off |                    0 |
| N/A   33C    P0    26W / 250W |      0MiB / 32480MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+


 gpustat
n060.default.domain  Wed Dec 12 08:29:10 2018
[0] Tesla V100-PCIE-32GB | 32'C,   0 % |     0 / 32480 MB |
[1] Tesla V100-PCIE-32GB | 33'C,   0 % |     0 / 32480 MB |
[2] Tesla V100-PCIE-32GB | 33'C,   0 % |     0 / 32480 MB |
[3] Tesla V100-PCIE-32GB | 33'C,   0 % |     0 / 32480 MB |
[4] Tesla V100-PCIE-32GB | 34'C,   0 % |     0 / 32480 MB |
[5] Tesla V100-PCIE-32GB | 33'C,   0 % |     0 / 32480 MB |


These are the specs for each GPU card:
GPU Architecture NVIDIA Volta
NVIDIA Tensor Cores 640
NVIDIA CUDA® Cores 5,120
Double-Precision Performance 7 TFLOPS
Single-Precision Performance 14 TFLOPS
Tensor Performance 112 TFLOPS
GPU Memory 16 GB HBM2
Memory Bandwidth 900 GB/sec
ECC YESInterconnect Bandwidth* 32 GB/sec
System Interface PCIe Gen3
Form Factor  PCIe Full Height/Length
Max Power Comsumption 250 W
Thermal Solution Passive
Compute APIs CUDA, DirectCompute,OpenCLTM, OpenACC
...

Configuration
...

pbs node configuration
 No Format
Qmgr: p n n060
#
# Create nodes and set their properties.
#
#
# Create and define node n060
#
create node n060 Mom=n060.default.domain
set node n060 state = free
set node n060 resources_available.arch = linux
set node n060 resources_available.host = n060
set node n060 resources_available.mem = 197554184kb
set node n060 resources_available.ncpus = 70
set node n060 resources_available.ngpus = 6
set node n060 resources_available.vnode = n060
set node n060 resv_enable = True
...
 No Format
 pbs queue configuration

Qmgr: p q dljun
#
# Create queues and set their attributes.
#
#
# Create and define queue dljun
#
create queue dljun
set queue dljun queue_type = Execution
set queue dljun Priority = 20
set queue dljun acl_user_enable = True
set queue dljun acl_users = redacted
set queue dljun acl_users += redacted
<redacted>
set queue dljun resources_max.ncpus = 56
set queue dljun resources_max.ngpus = 5
set queue dljun resources_max.nodect = 1
set queue dljun resources_max.walltime = 300:00:00
set queue dljun resources_default.ncpus = 1
set queue dljun resources_default.nodect = 1
set queue dljun resources_default.nodes = 1
set queue dljun resources_default.walltime = 100:00:00
set queue dljun acl_group_enable = True
set queue dljun acl_groups = deeplearning
set queue dljun enabled = True
set queue dljun started = True


# Create and define queue dlyao
#
create queue dlyao
set queue dlyao queue_type = Execution
set queue dlyao Priority = 20
set queue dlyao acl_user_enable = True
set queue dlyao acl_users = redacted
set queue dlyao acl_users += redacted
<redacted>
set queue dlyao resources_max.ncpus = 12
set queue dlyao resources_max.ngpus = 1
set queue dlyao resources_max.nodect = 1
set queue dlyao resources_max.walltime = 300:00:00
set queue dlyao resources_default.ncpus = 1
set queue dlyao resources_default.nodect = 1
set queue dlyao resources_default.nodes = 1
set queue dlyao resources_default.walltime = 100:00:00
set queue dlyao acl_group_enable = True
set queue dlyao acl_groups = aspen
set queue dlyao enabled = True
set queue dlyao started = True
...

Installed Applications
 

 No Format
 'module avail" will list currently installed application
 
e.g: module load anaconda/5.3.1py3
conda info --envs
source activate keras
pip install soundfile

...
 No Format
Qs: Regarding the output, there are some print lines in my code that help me to monitor how my program is working. like the error of model and so
on.  So is there any way to see this kind of online output on the terminal or log files while the job is being processed by the cluster?

Ans: There are a few ways of doing this. 
1. You may run an interactive pbs job with "-I" option. 
For example: qsub -I -q dljun@n060iworkq -W group_list=deeplearning -A deeplearning -l select=1:ncpus=1:ngpus=1:mem=12gb,walltime=100:00:00 
After this you will be given a shell and then you can run your command: 
module load anaconda/5.3.1py3
module load cuda/10.0
source activate tensorflow-gpu
python3 /export/home/s5108500/lscratch/Nick/DeepModels/keypoints/baseline_main.py

2. Alternatively, submit the job. Run the script named watch_jobs.sh
It will ask for the compute node name and the pbs job number and basically will run this command:
tail -f /var/spool/pbs/spool/$JOBNO.n060.*

e.g:
sh watch_jobs.sh

n060:
                                                            Req'd  Req'd   Elap
Job ID          Username Queue    Jobname    SessID NDS TSK Memory Time  S Time
--------------- -------- -------- ---------- ------ --- --- ------ ----- - -----
58.n060         s2594054 dljun    IndyTestDL  45304   1   1   12gb 100:0 R 00:11 n060/0
===========================
Please enter Node Number e.g: n060
n060
Please enter Job number  e.g 9066
58
===========================

|   5  Tesla V100-PCIE...  On   | 00000000:89:00.0 Off |                    0 |
| N/A   33C    P0    26W / 250W |      0MiB / 32480MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

?

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+
...
 No Format
Check if this returns correctly
 
 /usr/local/cuda-10.0/samples/bin/x86_64/linux/release/deviceQuery
 
>>>>>>>
/usr/local/cuda-10.0/samples/bin/x86_64/linux/release/deviceQuery Starting...
 CUDA Device Query (Runtime API) version (CUDART static linking)
Detected 6 CUDA Capable device(s)
Device 0: "Tesla V100-PCIE-32GB"
  CUDA Driver Version / Runtime Version          10.0 / 10.0
  CUDA Capability Major/Minor version number:    7.0
  Total amount of global memory:                 32480 MBytes (34058272768 bytes)
  (80) Multiprocessors, ( 64) CUDA Cores/MP:     5120 CUDA Cores
  GPU Max Clock rate:                            1380 MHz (1.38 GHz)
  Memory Clock rate:                             877 Mhz
  Memory Bus Width:                              4096-bit
  L2 Cache Size:                                 6291456 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 7 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Enabled
  Device supports Unified Addressing (UVA):      Yes
  Device supports Compute Preemption:            Yes
  Supports Cooperative Kernel Launch:            Yes
  Supports MultiDevice Co-op Kernel Launch:      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 20 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
Device 1: "Tesla V100-PCIE-32GB"
  CUDA Driver Version / Runtime Version          10.0 / 10.0
  CUDA Capability Major/Minor version number:    7.0
  Total amount of global memory:                 32480 MBytes (34058272768 bytes)
  (80) Multiprocessors, ( 64) CUDA Cores/MP:     5120 CUDA Cores
  GPU Max Clock rate:                            1380 MHz (1.38 GHz)
  Memory Clock rate:                             877 Mhz
  Memory Bus Width:                              4096-bit
  L2 Cache Size:                                 6291456 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 7 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Enabled
  Device supports Unified Addressing (UVA):      Yes
  Device supports Compute Preemption:            Yes
  Supports Cooperative Kernel Launch:            Yes
  Supports MultiDevice Co-op Kernel Launch:      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 21 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
Device 2: "Tesla V100-PCIE-32GB"
  CUDA Driver Version / Runtime Version          10.0 / 10.0
  CUDA Capability Major/Minor version number:    7.0
  Total amount of global memory:                 32480 MBytes (34058272768 bytes)
  (80) Multiprocessors, ( 64) CUDA Cores/MP:     5120 CUDA Cores
  GPU Max Clock rate:                            1380 MHz (1.38 GHz)
  Memory Clock rate:                             877 Mhz
  Memory Bus Width:                              4096-bit
  L2 Cache Size:                                 6291456 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 7 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Enabled
  Device supports Unified Addressing (UVA):      Yes
  Device supports Compute Preemption:            Yes
  Supports Cooperative Kernel Launch:            Yes
  Supports MultiDevice Co-op Kernel Launch:      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 57 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
Device 3: "Tesla V100-PCIE-32GB"
  CUDA Driver Version / Runtime Version          10.0 / 10.0
  CUDA Capability Major/Minor version number:    7.0
  Total amount of global memory:                 32480 MBytes (34058272768 bytes)
  (80) Multiprocessors, ( 64) CUDA Cores/MP:     5120 CUDA Cores
  GPU Max Clock rate:                            1380 MHz (1.38 GHz)
  Memory Clock rate:                             877 Mhz
  Memory Bus Width:                              4096-bit
  L2 Cache Size:                                 6291456 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 7 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Enabled
  Device supports Unified Addressing (UVA):      Yes
  Device supports Compute Preemption:            Yes
  Supports Cooperative Kernel Launch:            Yes
  Supports MultiDevice Co-op Kernel Launch:      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 58 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
Device 4: "Tesla V100-PCIE-32GB"
  CUDA Driver Version / Runtime Version          10.0 / 10.0
  CUDA Capability Major/Minor version number:    7.0
  Total amount of global memory:                 32480 MBytes (34058272768 bytes)
  (80) Multiprocessors, ( 64) CUDA Cores/MP:     5120 CUDA Cores
  GPU Max Clock rate:                            1380 MHz (1.38 GHz)
  Memory Clock rate:                             877 Mhz
  Memory Bus Width:                              4096-bit
  L2 Cache Size:                                 6291456 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 7 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Enabled
  Device supports Unified Addressing (UVA):      Yes
  Device supports Compute Preemption:            Yes
  Supports Cooperative Kernel Launch:            Yes
  Supports MultiDevice Co-op Kernel Launch:      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 136 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
Device 5: "Tesla V100-PCIE-32GB"
  CUDA Driver Version / Runtime Version          10.0 / 10.0
  CUDA Capability Major/Minor version number:    7.0
  Total amount of global memory:                 32480 MBytes (34058272768 bytes)
  (80) Multiprocessors, ( 64) CUDA Cores/MP:     5120 CUDA Cores
  GPU Max Clock rate:                            1380 MHz (1.38 GHz)
  Memory Clock rate:                             877 Mhz
  Memory Bus Width:                              4096-bit
  L2 Cache Size:                                 6291456 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 7 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Enabled
  Device supports Unified Addressing (UVA):      Yes
  Device supports Compute Preemption:            Yes
  Supports Cooperative Kernel Launch:            Yes
  Supports MultiDevice Co-op Kernel Launch:      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 137 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
> Peer access from Tesla V100-PCIE-32GB (GPU0) -> Tesla V100-PCIE-32GB (GPU1) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU0) -> Tesla V100-PCIE-32GB (GPU2) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU0) -> Tesla V100-PCIE-32GB (GPU3) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU0) -> Tesla V100-PCIE-32GB (GPU4) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU0) -> Tesla V100-PCIE-32GB (GPU5) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU1) -> Tesla V100-PCIE-32GB (GPU0) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU1) -> Tesla V100-PCIE-32GB (GPU2) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU1) -> Tesla V100-PCIE-32GB (GPU3) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU1) -> Tesla V100-PCIE-32GB (GPU4) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU1) -> Tesla V100-PCIE-32GB (GPU5) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU2) -> Tesla V100-PCIE-32GB (GPU0) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU2) -> Tesla V100-PCIE-32GB (GPU1) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU2) -> Tesla V100-PCIE-32GB (GPU3) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU2) -> Tesla V100-PCIE-32GB (GPU4) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU2) -> Tesla V100-PCIE-32GB (GPU5) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU3) -> Tesla V100-PCIE-32GB (GPU0) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU3) -> Tesla V100-PCIE-32GB (GPU1) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU3) -> Tesla V100-PCIE-32GB (GPU2) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU3) -> Tesla V100-PCIE-32GB (GPU4) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU3) -> Tesla V100-PCIE-32GB (GPU5) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU4) -> Tesla V100-PCIE-32GB (GPU0) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU4) -> Tesla V100-PCIE-32GB (GPU1) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU4) -> Tesla V100-PCIE-32GB (GPU2) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU4) -> Tesla V100-PCIE-32GB (GPU3) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU4) -> Tesla V100-PCIE-32GB (GPU5) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU5) -> Tesla V100-PCIE-32GB (GPU0) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU5) -> Tesla V100-PCIE-32GB (GPU1) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU5) -> Tesla V100-PCIE-32GB (GPU2) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU5) -> Tesla V100-PCIE-32GB (GPU3) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU5) -> Tesla V100-PCIE-32GB (GPU4) : Yes
deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 10.0, CUDA Runtime Version = 10.0, NumDevs = 6
Result = PASS
 
>>>>>>>>
gpu issues - Sample torch.device.py 

 No Format
more log_device_placement.py
####https://www.tensorflow.org/guide/using_gpu
import tensorflow as tf
# Creates a graph.
a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
c = tf.matmul(a, b)
# Creates a session with log_device_placement set to True.
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
# Runs the op.
print(sess.run(c))


...
 No Format
https://www.tensorflow.org/tutorials
 
cat tensorflowTutorial.py
###########################
 
import tensorflow as tf
mnist = tf.keras.datasets.mnist

(x_train, y_train),(x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0

model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(input_shape=(28, 28)),
  tf.keras.layers.Dense(512, activation=tf.nn.relu),
  tf.keras.layers.Dropout(0.2),
  tf.keras.layers.Dense(10, activation=tf.nn.softmax)
])
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

model.fit(x_train, y_train, epochs=5)
model.evaluate(x_test, y_test)


...

gpu issues - sample pbs script
 No Format
Here is sample pbs scripts
 
Sample PBS script:
==================
cat pbs.tensor.01

#!/bin/bash 
#PBS -m abe
#PBS -M Youremail@griffith.edu.au
#PBS -V
#PBS -N testImage
#PBS -q dljun@n060
#PBS -W group_list=deeplearning -A deeplearning
#PBS -l select=1:ncpus=1:ngpus=1:mem=32gb,walltime=300:00:00
#PBS -j oe
module load anaconda/5.3.1py3
#conda info --envs
#source activate deeplearning
source activate tensorflow-gpu
##nvidia-debugdump -l
##nvidia-smi
###python main.py --cfg cfg/config3.yml --gpu 0
cd  $PBS_O_WORKDIR
python /export/home/s12345/lpbs/cuda/tensorflowTutorial.py
 
 
 
module load anaconda/5.3.1py3
#conda info --envs
#source activate deeplearning
source activate tensorflow-gpu
##nvidia-debugdump -l
##nvidia-smi
###python main.py --cfg cfg/config3.yml --gpu 0
cd  $PBS_O_WORKDIR
python /export/home/s12345/lpbs/cuda/tensorflowTutorial.py
 

How do I run multiple tensorflow scripts in the same job
 No Format
cat pbs.01

>>>>>>>>>>>>>>>>>>>
#!/bin/bash
#PBS -m abe
#PBS -M myemail@griffith.edu.au
#PBS -V
#PBS -N verc235
#PBS -q dljun@n060
#PBS -W group_list=deeplearning -A deeplearning
#PBS -l select=1:ncpus=16:ngpus=1:mem=32gb,walltime=300:00:00
#cd  $PBS_O_WORKDIR
GPUNUM=`echo $CUDA_VISIBLE_DEVICES`
module load anaconda/5.3.1py3
module load cuda/10.0
#conda info --envs
#source activate deeplearning
source activate tensorflow-gpu
##nvidia-debugdump -l
##nvidia-smi
GPUNUM=`echo $CUDA_VISIBLE_DEVICES`
MASTERDIR=/export/home/s1234/scratch/home/DeepXi/ver/c2/5
cd $MASTERDIR/5
python3 deepxi.py --train 1 --gpu $GPUNUM &
cd $MASTERDIR/10
python3 deepxi.py --train 1 --gpu $GPUNUM &
cd $MASTERDIR/15
python3 deepxi.py --train 1 --gpu $GPUNUM &
cd $MASTERDIR/20
python3 deepxi.py --train 1 --gpu $GPUNUM
>>>>>>>>
Submit the job like this:
qsub pbs.01

Reference
  1. https://www.pbsworks.com/pdfs/PBSAdminGuide18.2.pdf
  2. https://conf-ers.griffith.edu.au/download/attachments/21332198/xl270d_gen10.pdf?api=v2
  3. https://www.microway.com/hpc-tech-tips/nvidia-smi_control-your-gpus/
  4. https://weeraman.com/put-that-gpu-to-good-use-with-python-e5a437168c01
  5. https://stackoverflow.com/questions/48152674/how-to-check-if-pytorch-is-using-the-gpu
  6. https://discuss.pytorch.org/t/solved-make-sure-that-pytorch-using-gpu-to-compute/4870/14
  7. https://www.tensorflow.org/guide/using_gpu