Clusters/Partitions overview

Fig. 35 A view of PSMN clusters

Note

Clusters are the hardware assembly (including dedicated network, etc). Partitions are logical divisions within clusters.

Cluster E5

Hardware specifications per node:

login nodes	CPU Model	cores	RAM	ratio	infiniband	GPU	local scratch
c8220node1	E5-2670 0 @ 2.6GHz	16	64 GiB	4 GiB/core	56 GiB/s
	E5-2667v2 @ 3.3GHz	16	128 GiB	8 GiB/core	56 GiB/s
	E5-2667v2 @ 3.3GHz	16	256 GiB	16 GiB/core	56 GiB/s
e5-2667v4comp[1-2]	E5-2667v4 @ 3.2GHz	16	128 GiB	8 GiB/core	56 GiB/s
	E5-2667v4 @ 3.2GHz	16	256 GiB	16 GiB/core	56 GiB/s
	E5-2697Av4 @ 2.6GHz	32	256 GiB	16 GiB/core	56 GiB/s

Hint

Best use case: training, testing, sequential jobs, small parallel jobs (<32c)

• E5 scratch (general purpose scratch) is available, from all nodes of the partition, on the following path:

/scratch/E5N

• Choose ‘E5’ environment modules (See Modular Environment). See Using X2Go for data visualization for visualization servers informations.

Partition E5-short

This partition is open to everyone, 30 minutes walltime.

Partition E5

This partition is open to everyone, 4 hours walltime.

Partition E5-long

This partition is open to everyone, 8 days walltime.

Cluster Lake

Partition E5-GPU

Hardware specifications per node:

login nodes	CPU Model	cores	RAM	ratio	infiniband	GPU	local scratch
r730gpu01	E5-2637v3 @ 3.5GHz	8	128 GiB	16 GiB/core	56 GiB/s	RTX2080Ti	/scratch/Lake

This partition is open to everyone, 8 days walltime (and yes, it is in Lake cluster).

Hint

Best use case: GPU jobs, training, testing, sequential jobs, small parallel jobs (<32c)

• Choose ‘E5’ environment modules (See Modular Environment),

• Partition E5-GPU has access to Lake scratchs, and has no access to /scratch/E5N (see Partition Lake below),

• 8 days walltime.

Partition Lake

Hardware specifications per node:

login nodes	CPU Model	cores	RAM	ratio	infiniband	GPU	local scratch
m6142comp[1-2]	Gold 6142 @ 2.6GHz	32	384 GiB	12 GiB/core	56 GiB/s
	Gold 6142 @ 2.6GHz	32	384 GiB	12 GiB/core	56 GiB/s		/scratch/disk/
	Gold 5118 @ 2.3GHz	24	96 GiB	4 GiB/core	56 GiB/s
cl6242comp[1-2]	Gold 6242 @ 2.8GHz	32	384 GiB	12 GiB/core	56 GiB/s
cl5218comp[1-2]	Gold 5218 @ 2.3GHz	32	192 GiB	6 GiB/core	56 GiB/s
cl6226comp[1-2]	Gold 6226R @ 2.9GHz	32	192 GiB	6 GiB/core	56 GiB/s

This partition is open to everyone, 8 days walltime.

Hint

Best use case: medium parallel jobs (<384c)

• Lake scratchs are available, from all nodes of the partition, on the following paths:

/scratch/
    ├── Bio
    ├── Chimie
    ├── Lake    (general purpose scratch)
    └── Themiss

• ‘Bio’, ‘Chimie’ scratchs are meant for users of respective labs and teams,

• ‘Themiss’ scratch is reserved to Themiss ERC users,

• ‘Lake’ is for all users (general purpose scratch),

• Nodes c6420node[049-060],r740bigmem201 have a local scratch on /scratch/disk/ (120 days lifetime residency, --constraint=local_scratch),

• Choose ‘Lake’ environment modules (See Modular Environment). See Using X2Go for data visualization for visualization servers informations,

• 8 days walltime.

Partition Lake-bigmem

Note

This partition is subject to authorization, open a ticket to justify access.

Hint

Best use case: large memory jobs (<32c, 32c max), sequential jobs

Hardware specifications per node:

login nodes	CPU Model	cores	RAM	ratio	infiniband	GPU	local scratch
none	Gold 6226R @ 2.9GHz	32	1,5 TiB	46 GiB/core	56 GiB/s		/scratch/disk/

• Lake scratchs are available (see Partition Lake above), and a local scratch on /scratch/disk/ (120 days lifetime residency, --constraint=local_scratch),

• 8 days walltime.

Partition Lake-premium

This partition is open to everyone, 8 days walltime, but has a preemption mode (see Preemption mode), in which jobs with high priority will requeue jobs with normal priority. Ideal for burst campaigns (lots of short jobs) and workflows with multiple jobs.

Hint

Best use case: sequential jobs, medium parallel jobs (<384c)

Hardware specifications per node:

login nodes	CPU Model	cores	RAM	ratio	infiniband	GPU	local scratch
cl6242comp[1-2]	Gold 6242 @ 2.8GHz	32	384 GiB	12 GiB/core	56 GiB/s

• Lake scratchs are available (see Partition Lake above),

• 8 days walltime.

Cluster Epyc

Hardware specifications per node:

login nodes	CPU Model	cores	RAM	ratio	infiniband	GPU	local scratch
none	EPYC 7702 @ 2.0GHz	128	512 GiB	4 GiB/core	100 GiB/s	N/A	None

Partition Epyc

This partition is open to everyone, 8 days walltime.

Hint

Best use case: large parallel jobs (>256c)

• There is no login node available in the Epyc partition at the moment. Use an Interactive session for builds/tests,

• Epyc partition has access to Lake scratchs (see Partition Lake above),

• There is no specific environment modules (use ‘E5’),

• 8 days walltime.

Cluster Cascade

Hardware specifications per node:

login nodes	CPU Model	cores	RAM	ratio	infiniband	GPU	local scratch
s92node01	Platinum 9242 @ 2.3GHz	96	384 GiB	4 GiB/core	100 GiB/s	N/A	None

Partition Cascade

This partition is open to everyone, 8 days walltime.

Hint

Best use case: sequential jobs, large parallel jobs (>256c)

• Cascade scratch (general purpose scratch) is available, from all nodes of the partition, on the following path:

/scratch/
    ├── Cascade     (general purpose scratch)
    └── Cral

• ‘Cral’ scratch is meant for CRAL users and CRAL teams,

• Choose ‘Cascade’ environment modules (See Modular Environment),

• 8 days walltime.

Partition Cascade-premium

Hint

Best use case: sequential jobs, large parallel jobs (>256c)

Same as partition Cascade above, this partition is open to everyone, 8 days walltime, but with a preemption mode (see Preemption mode), in which jobs with high priority will requeue jobs with normal priority. Ideal for burst campaigns (lots of short jobs) and workflows with multiple sequential jobs.

Partition Cascade-GPU

Hardware specifications per node:

login nodes	CPU Model	cores	RAM	ratio	infiniband	GPU	local scratch
none	Platinum 9242 @ 2.3GHz	96	384 GiB	4 GiB/core	100 GiB/s	L4	None

Same as partition Cascade above, this partition is open to everyone, 8 days walltime.

Hint

Best use case: GPU jobs, training, testing, sequential jobs, large parallel jobs (>256c)