Computing Hardware & Partitions

If you are considering purchasing new hardware to integrate into the cluster, please contact the cluster administrator. Coordination is necessary to ensure the hardware complies with minimal requirements and to synchronize procurement efforts with other departments. The data center in Troja is available for hosting of servers only if they contribute to the shared computational resources immediately or by the end of an externally funded project (e.g. ERC, GAČR).

SLURM Essentials

  • Terminology: A cluster consists of nodes, which are grouped into partitions. Users submit jobs (consisting of tasks) to a scheduling queue.
  • GRES (Generic Resources): Used to request specialized hardware like GPUs. Use --gpus=count or --gres=gpu:type:count in submission commands.
  • Fair Share Priority: Users who have used the cluster less in recent weeks receive higher priority. The priority half-life is 14 days.
  • Preemption: Low-priority jobs in partitions like ffa-preempt can be interrupted by high-priority jobs.
    • REQUEUE: The job is killed and requeued; Slurm sends a signal one minute before termination.
    • SUSPEND: The job is paused, releasing CPUs but not memory.
  • Checkpointing: Essential for preemptable jobs. Use the scheckpoint command or the DMTCP system.
  • Important Commands: sinfo (status), squeue (queue info), srun (interactive), sbatch (batch), scancel (terminate job), seff (efficiency statistics).

NOTE: Cores in next tables are per socket; also HT are per core - you can also use sinfo -N -O NodeList,SocketCoreThread,Memory:.10,Features:.50 | uniq to get information about sockets, cores, SMT, memory, and features.

Chimera Specification

Nodes with the "ib" feature are interconnected with HDR InfiniBand.

Chimera Nodes

Node names CPU Sockets Cores HT RAM GRES Additional info
deepth1[3-4] AMD EPYC 7282 2 16 2 128 GB IB
utf-asus[1-3] AMD EPYC 7302 2 16 2 256 GB local_scratch, IB
utf-hm1 AMD EPYC 7763 2 64 2 2 TB local_scratch, swap, IB
node[0-3,6] Intel(R) Xeon(R) E5-2630 v4 2 10 2 256 GB local_scratch
ucjf-asus1 AMD EPYC 7532 2 32 2 512 GB IB
ucjf-asus1 AMD EPYC 7532 2 32 2 512 GB local_scratch, IB
ucjf-asusb[1-2] AMD EPYC 7352 2 24 2 256 GB local_scratch, IB
onos2 Intel(R) Xeon(R) CPU E5-2670 v3 2 12 2 128 GB
onos[3-4] Intel(R) Xeon(R) CPU E5-2670 v3 2 12 2 256 GB
c10t4[0-1] Intel(R) Xeon(R) CPU E5-2640 v3 2 8 2 64 GB
c11t4[2-3] Intel(R) Xeon(R) CPU E5-2640 v3 2 8 2 64 GB
c14t3[2-3] Intel(R) Xeon(R) CPU E5-2640 v3 2 8 2 64 GB
c23t46 Intel(R) Xeon(R) CPU E5-2630 v3 2 10 2 64 GB
c74t57 AMD EPYC 7451 2 24 1 128 GB gpu:quadro_m2000 vram:4G NVIDIA Quadro M2000 4 GB
c71t55,c70t54,c72t56,c73t38 AMD EPYC 7451 2 24 1 128 GB
c73t38 AMD EPYC 7451 2 24 1 128 GB
c42t47 Intel(R) Xeon(R) CPU E5-2650 v4 2 12 2 64 GB
c41t19 Intel(R) Xeon(R) CPU E5-2650 v4 2 12 1 64 GB
c40t17,c43t48 Intel(R) Xeon(R) CPU E5-2630 v4 2 10 1 64 GB
c27t15,c28t14 Intel(R) Xeon(R) CPU E5-2630 v4 2 10 2 64 GB
c67t02 Intel(R) Xeon(R) CPU E5-2630 v4 2 10 1 64 GB gpu:quadro_m2000 vram:4G NVIDIA Quadro M2000 4 GB
c64t03,c39t09 Intel(R) Xeon(R) CPU E5-2630 v4 2 10 2 64 GB
c65t04,c66t05 Intel(R) Xeon(R) CPU E5-2650 v4 2 12 2 64 GB
c77t06,c78t10,c75t34 AMD EPYC 7532 2 32 1 256 GB
c93t33 AMD EPYC 7543 2 32 1 512 GB 16 numas
c94t26 AMD EPYC 7543 2 32 1 256 GB 16 numas
c08t08 Intel(R) Xeon(R) CPU E5-2650 v2 2 8 1 64 GB no AVX2, FMA
mff-a2-[01-02] AMD EPYC 7543 2 32 2 512 GB gpu:T600:1 vram:4G IB, NVIDIA T600 4 GB
*-a2-[03-18] AMD EPYC 7543 2 32 2 512 GB IB
mff-a2-19 AMD EPYC 7543 2 32 2 512 GB gpu:L40:1 mps:100 vram:48G IB, NVIDIA Tesla L40 48 GB
ucjf-a4-01 AMD EPYC 9354 2 32 2 384 GB gpu:L40:1 mps:100 vram:48G IB, NVIDIA Tesla L40 48 GB
r[31-35] Intel(R) Xeon(R) Gold 6140 CPU 2 18 2 128 GB IB(EDR)
r[36-39,41-47] Intel(R) Xeon(R) Gold 6240 CPU 2 18 2 128 GB IB(EDR)
r40 Intel(R) Xeon(R) Gold 6240 CPU 4 18 2 512 GB IB(EDR)
r50 AMD EPYC 9654 2 96 2 1.5 TB gpu:MI210:2 vram:64G IB(EDR), AMD MI210
fuuk-1[2-4] AMD EPYC 9654 1 96 2 1.5 TB gpu:L40:4 vram:48G IB(EDR), 4x NVIDIA TESLA L40s 48 GB

Chimera node features

Feature description
ib node has infiniband
local_scratch nde has local scratch
avx2, sse42, fma, CPU support this instruction set
avx512_old CPUs supporting avx512f, avx512dq, avx512cd, avx512bw, avx512vl, avx512_vnni instruction sets (older Intel CPUs)
avx512 supports avx512_old and avx512ifma, avx512_bf16, avx512vbmi, avx512_vbmi2, avx512_bitalg,avx512_vpopcntdq instruction sets (new AMD / Intel CPUs)
#C Procesor has # CPU cores
#T Core has # SMT threads
#N Machine has # numa nodes
#S Machine has # sockets

Chimera Partitions

Name Nodes Priority Timelimit DefaultTime Preemption Oversubs Allowed accounts Intended use
ffa all without debug and subset of auuk, ucjf, mff-a2-[01-02] 5 12 hours 1 hour no FORCE:1 any free-for-all, any job
ffa-short all without debug and subset of auuk, ucjf, mff-a2-[01-02] 5 2 hours 30 minutes no FORCE:1 any default, free-for-all, any job
ffa-long mff-a2-19 5 07 days 1 day no FORCE:1 any free-for-all, any, QoS only 1/2 of the node
ffa-preempt* all without mff-a2-[01-02] 5 36 hours 1 hour REQUEUE NO any free-for-all, any job, with preemption
ffa-checkpoint* all without subset of ucjf , mff-a2-[01-02] 5 36 hours 1 hour REQUEUE NO any free-for-all, jobs with checkpointing, with preemption, grace time 60s
auuk subset of c[00-99]t[00-99], auuk-a2-[07-15,17-18] 40 8 hours 4 hours NO FORCE:1 auuk jobs for AUUK
auuk-preempt* c[00-99]t[00-99], auuk-a2-[07-15,17-18] 20 12 hours 8 hours REQUEUE FORCE:1 auuk jobs for AUUK with preemption
utf utf-* node[0-1,6], r4[1-7] 40 infinite 1 day no FORCE:1 utf jobs for UTF
utf-pejcha r-[31-40] 40 infinite 1 day no FORCE:1 utf-pejcha jobs for Pejcha's group
fuuk deepth1[3-4] 40 infinite 1 day no FORCE:1 fuuk jobs for FUUK
fuuk-barvik fuuk-1[2-4] 40 infinite 1 day no FORCE:1 fuuk-barvik jobs for Barvik's group
ucjf ucjf-* 40 infinite 1 day no FORCE:1 ucjf jobs for UCJF
math r[41-47,50] 40 infinite 1 day no FORCE:1 math jobs for mathematical section
debug onos[3-4] 100 12 hour 1 hour no FORCE:1 any for interactive jobs only, QoS debug (max CPUs:24 and memory:200GB)
checkpoint onos[3-4], c14t3[2-3] 50 10 minutes 5 minutes no FORCE:1 any for starting jobs with checkpoints
gpu-ffa mff-a2-19, ucjf-a4-01, fuuk-1[2-4] 10 12 hours 1 hour no FORCE:1 any only jobs that require gpu
edu mff-a2-[01-02] 20 3 hours 2 hours no FORECE:1 any for education, only one job per user

*preemptable partitions have lower fair share billing (TRESBillingWeights="CPU=") ; ffa-preempt 0.1; ffa-checkpoint 0.15; auuk-preempt 0.25

Parlab Specification

All nodes are interconnected by InfiniBand FDR (56 Gb/s) and 10 GbE. The front-end server is connected by 10 GbE to the external world. The latest version of OpenMPI is installed on all nodes.

Parlab Nodes

Node names CPU Sockets Cores HT RAM GRES Additional info
w[201-208] Intel Xeon Gold 6130 2 16 2 128 GB

Parlab Partitions

Name Nodes Priority Timelimit Preemption Allowed accounts Intended use
ffa all 10 12 hours REQUEUE any default, free-for-all, any job
mpi-homo-ffa w[201-208] 30 12 hours REQUEUE any free-for-all, MPI jobs on homogeneous nodes
mpi-homo-long w[201-208] 100 7 days REQUEUE ksi, kdss, kdsstudent, nprg042s, nprg058s, nprg054s debugging on newer CPUs, MPI debugging, long jobs
mpi-homo-short w[201-208] 150 2 hours REQUEUE ksi, kdss, kdsstudent, nprg042s, nprg058s, nprg054s executing short jobs on 2-socket system, MPI jobs on homogeneous nodes

Gpulab Specification

These nodes are integrated into Chimera with their own environment and 10 GbE interconnects.

Gpulab Nodes

Node names CPU Sockets Cores HT RAM GRES Additional info
volta01 Intel Xeon Silver 4110 2 8 2 256 GB gpu:V100:2, vram:16G 1x NVIDIA V100 PCIe 16 GB, latest V100 compatible CUDA (12.9.1)
volta02 Intel Xeon Silver 4110 2 8 2 192 GB gpu:V100:2, vram:16G 2x NVIDIA V100 PCIe 16 GB, latest V100 compatible CUDA (12.9.1)
volta03 Intel Xeon Silver 4110 2 8 2 192 GB gpu:V100:2, vram:16G 2x NVIDIA V100 PCIe 16 GB, latest V100 compatible CUDA (12.9.1)
volta04 Intel Xeon Gold 5218 2 16 2 384 GB gpu:V100:1, vram:32G 1x NVIDIA V100 SXM2 32 GB, latest V100 compatible CUDA (12.9.1)
volta05 Intel Xeon Gold 5218 2 16 2 384 GB gpu:V100:4, vram:32G 4x NVIDIA V100 SXM2 32 GB, latest V100 compatible CUDA (12.9.1)
ampere01 AMD EPYC 9554P 1 64 2 384 GB gpu:L40:4, vram:48G 4x NVIDIA L40 PCIe 48 GB, latest CUDA
ampere02 AMD EPYC 9374F 2 32 2 768 GB gpu:A100:2, vram:80G 2x NVIDIA A100 PCIe 80 GB, latest CUDA
ampere03 AMD EPYC 9554 1 64 2 384 GB gpu:L40:1, vram:40G 1x NVIDIA L40 PCIe 40 GB, latest CUDA
hopper01 AMD EPYC 9554 2 48 2 1024 GB gpu:H100:2, vram:80G 2x NVIDIA H100 PCIe 80 GB, latest CUDA
bw01 AMD EPYC 9454 1 48 1 768 GB gpu:H100:2, vram:80G 2x NVIDIA RTX PRO Blackwell Server Edition PCIe 96 GB, latest CUDA

Gpulab Partitions

Name Nodes Priority Timelimit Preemption Allowed accounts Intended use
ffa-gpulab all 10 12 hours REQUEUE any default, free-for-all, any job
gpu-ffa-gpulab volta[01-05],ampere[01-03],hopper[01] 20 12 hours REQUEUE any free-for-all, any GPU job
gpu-long volta[01-05],ampere[01-02],hopper[01] 100 7 days REQUEUE kdss, kdsstudent, ksviprj long GPU jobs
gpu-short volta[01-05],ampere[01-02],hopper[01] 150 2 hours REQUEUE kdss, kdsstudent, ksviprj debugging GPU tasks, executing short GPU jobs
gpu-long-ksi volta[01-05],ampere[01-02],hopper[01] 200 7 days REQUEUE ksi, ksiprj, ksibio long GPU jobs for KSI department
gpu-short-ksi volta[01-05],ampere[01-02],hopper[01] 250 2 hours REQUEUE ksi, ksiprj, ksibio debugging GPU tasks, executing short GPU jobs
gpu-short-teach volta[01-05],ampere[01-02],hopper[01] 300 2 hours REQUEUE nprg042s, nprg042t, nprg058s, nprg058t parallel computing courses
gpu-bio ampere[01-03],hopper[01] 800 7 days REQUEUE ksibio GPU jobs for Bioinformatics study program