Myriad is designed for high I/O, high throughput jobs that will run within a single node rather than multi-node parallel jobs.
Myriad accounts can be applied for via the Research Computing sign up process.
As Myriad is our most general-purpose system, everyone who signs up for a Research Computing account is given access to Myriad.
You will use your UCL username and password to ssh in to Myriad.
If using PuTTY, put
myriad.rc.ucl.ac.uk as the hostname and your
seven-character username (with no @ after) as the username when logging
uccaxxx. When entering your password in PuTTY no characters or
bulletpoints will show on screen - this is normal.
If you are outside the UCL firewall you will need to follow the instructions for Logging in from outside the UCL firewall.
Logging in to a specific node§
You can access a specific Myriad login node with:
ssh email@example.com ssh firstname.lastname@example.org
The main address will redirect you on to either one of them.
Copying data onto Myriad§
You will need to use an SCP or SFTP client to copy data onto Myriad. Please refer to the page on How do I transfer data onto the system?
The default quotas on Myriad are 150GB for home and 1TB for Scratch.
These are hard quotas: once you reach them, you will no longer be able to write more data. Keep an eye on them, as this will cause jobs to fail if they cannot create their .o or .e files at the start, or their output files partway through.
You can check both quotas on Myriad by running:
which will give you output similar to this:
Storage Used Quota % Used Path home 721.68 MiB 150.00 GiB 0% /home/uccaxxx scratch 52.09 MiB 1.00 TiB 0% /scratch/scratch/uccaxxx
You can apply for quota increases using the form at Additional Resource Requests.
Here are some tips for managing your quota and finding where space is being used.
|2 to 36||48hrs|
Interactive jobs run with
qrsh have a
maximum wallclock time of 2 hours.
Myriad contains three main node types: standard compute nodes, high memory nodes and GPU nodes. As new nodes as added over time with slightly newer processor variants, new letters are added.
|Type||Cores per node||RAM per node||tmpfs||Nodes|
|J||36 + 2 P100 GPUs||192GB||1500G||2|
|E,F||36 + 2 V100 GPUs||192GB||1500G||9|
You can tell the type of a node by its name: type H nodes are named
Here are the processors each node type has:
- F, H, I, J: Intel(R) Xeon(R) Gold 6140 CPU @ 2.30GHz
- B, D, E: Intel(R) Xeon(R) Gold 6240 CPU @ 2.60GHz
(If you ever need to check this, you can include
cat /proc/cpuinfo in your jobscript so
you get it in your job's .o file for the exact node your job ran on. You will get an entry
for every core).
Myriad has three types of GPU nodes, J, E and F. There are two J-type nodes each with two nVidia Tesla P100s. There is one F-type and eight E-type nodes, each with two nVidia Tesla V100s. The CPUs are slightly different on these two.
You can request one or two GPUs by adding them as a resource request to your jobscript:
# For 1 GPU #$ -l gpu=1 # For 2 GPUs #$ -l gpu=2
This will give you either type of GPU. If you need to specify one over the other, add a request for that type of node to your jobscript:
# request a V100 node only #$ -ac allow=EF
The GPU nodes page has some sample code for running GPU jobs if you need a test example.
Tensorflow is installed: type
module avail tensorflow to see the
Modules to load for the non-MKL GPU version:
module unload compilers mpi module load compilers/gnu/4.9.2 module load python3/recommended module load cuda/8.0.61-patch2/gnu-4.9.2 module load cudnn/6.0/cuda-8.0 module load tensorflow/1.4.1/gpu