CHPC Technical Specifications | High Performance Computing and Data Center

The CHPC cluster comprises 31 nodes, housing 1920 CPU cores and 5 GPUs, incorporating Dell PowerEdge R6525s and R7525s nodes and boasting a theoretical sustained single precision performance exceeding 170.69 TFLOPs.

Node CPU Specs

This table summarizes the CPU hardware specification for the nodes.

NodeName	CPUTot	ThreadsPerCore	CoresPerSocket	Sockets	Model name	CPU MHz	CPU max MHz	Partition	CPU TFLOPS
chpc-compute-12-3	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-4	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-5	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-6	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-7	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-8	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-9	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-10	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-11	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-12	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-13	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-14	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-15	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-16	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-17	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-18	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-19	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-20	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-21	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-22	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-23	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-compute-12-24	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.446	2395.446	main	2.45
chpc-gpu-10-0	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.454	2395.454	gpu	2.45
chpc-gpu-10-1	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.454	2395.454	gpu	2.45
chpc-gpu-10-2	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.454	2395.454	gpu	2.45
chpc-gpu-12-0	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.454	2395.454	gpu	2.45
chpc-gpu-12-1	64	1	32	2	AMD EPYC 7532 32-Core Processor	2395.454	2395.454	gpu	2.45
chpc-highmem-10-0	48	1	24	2	AMD EPYC 7352 24-Core Processor	2295.496	2295.496	highmem	1.76
chpc-highmem-10-1	48	1	24	2	AMD EPYC 7352 24-Core Processor	2295.496	2295.496	highmem	1.76
chpc-highmem-12-0	48	1	24	2	AMD EPYC 7352 24-Core Processor	2295.496	2295.496	highmem	1.76
chpc-highmem-12-1	48	1	24	2	AMD EPYC 7352 24-Core Processor	2295.496	2295.496	highmem	1.76

Node GPU Specs

This table summarizes the GPU hardware specification for the GPU nodes.

NodeName	Partition	Gres	GPU TFLOPS	GPU Deep Learning TFLOPS	GPU Memory (GB)	GPU CUDA Cores	GPU Tensor Cores	GPU Half Precision TFLOPS
chpc-gpu-10-0	gpu	a100:1	19.5	156.0	40.0	6912.0	432.0	312.0
chpc-gpu-10-1	gpu	a100:1	19.5	156.0	40.0	6912.0	432.0	312.0
chpc-gpu-10-2	gpu	a100:1	19.5	156.0	40.0	6912.0	432.0	312.0
chpc-gpu-12-0	gpu	a100:1	19.5	156.0	40.0	6912.0	432.0	312.0
chpc-gpu-12-1	gpu	a100:1	19.5	156.0	40.0	6912.0	432.0	312.0

Summary Plots

These visualizations present a summarized view of the compute resources within the CHPC cluster.

Node vs. CPU TFLOPS

This graphical representation offers insights into the theoretical CPU compute performance of individual nodes.

Node vs. GPU

This graphical representation offers insights into the theoretical GPU compute performance of individual nodes.

CPU Composition

This graphical representation offers insights into the CPU compositions of individual nodes.

QOS Specifications

The “main” QoS permits users to conduct calculations over a span of 5 days, capped at 1000 jobs or 1000 cores per user. Under the “long” QoS, users can access only 100 jobs or 100 cores for a duration of up to 14 days. For the “highmem” QoS, users can operate for one week with 100 jobs or 100 cores or allocate up to 6000 GB of RAM per user.

In the realm of the “gpu” QoS, users can leverage Nvidia A100 tensor core GPUs, with restrictions set to 3 jobs, 3 GPUs, or 192 cores, lasting up to 2 days. These GPUs deliver 9.7 TFLOPS of double-precision performance. However, when employed in machine learning applications, particularly neural networks with enabled sparsity, each GPU can achieve an impressive tensor performance of 780 TFLOPS individually.

QoS (rows) / Parition (columns) Limits on CHPC

QoS/Partition	Main	Highmem	GPU
Main	120 hours / job 1600 cores / group 1000 cores / job 1000 cores / user 1000 jobs / user	X	X
Long	336 hours / job 500 cores / group 300 cores / job 300 cores / user 300 jobs / user	X	X
Debug	15 mins / job 128 cores / group 32 cores / job 4 jobs / user	15 minutes / job 128 cores / group 32 cores / job 4 jobs / user	X
Highmem	X	168 hours / job 192 cores / group 100 cores / user 100 jobs / user 8TB memory / group 6 TB memory / user	X
GPU	X	X	48 hours / job 320 cores / group 192 cores / job 5 GPUs/ group 3 GPUs / user 3 jobs / user