Difference between revisions of "Large-Memory SMP Servers"

From UFRC
Jump to navigation Jump to search
(Created page with "The HPC Center within Research Computing currently maintains the following resources for calculations requiring large amounts of physical memory. # 21 Intel (E5462, 8x 2.8 GHz c...")
 
 
(32 intermediate revisions by 4 users not shown)
Line 1: Line 1:
The HPC Center within Research Computing currently maintains the following resources for calculations requiring large amounts of physical memory.
+
[[Category:Scheduler]]
 +
Research Computing currently maintains the following resources for calculations requiring large amounts of physical memory.
  
# 21 Intel (E5462, 8x 2.8 GHz cores) servers with 64 GB of RAM (physical memory)
+
{| class="wikitable" style="text-align: center;"
# 1 Intel (X7560, 16x 2.2 GHz cores) server with 128 GB of RAM.
+
|+ Large Memory Servers
# 1 AMD (Opteron 6174, 48x 2.2 GHz cores) server with 512 GB of RAM
+
|-
 +
! Nodes
 +
! Architecture
 +
! Vendor
 +
! Processor
 +
! Frequency
 +
! Cores
 +
! SLURM Memory(GB)
 +
|-
 +
| 1 || amd64 || AMD || EPYC 7702 || 2.0 || 128 || 4023
 +
|-
 +
| 2 || amd64 || Intel || E7-8850 || 3.2 || 192 || 1509
 +
|}
  
The 128 GB and 512 GB machines are available only via a dedicated queue (bigmem).  Before attempting to run jobs via the bigmem queue, you should request access to the bigmem queue via a support request (http://support.hpc.ufl.edu).
+
For more details see automatically generated data in [[Available Node Features]]
  
Note that memory is no less a consumable resource than processors.  Therefore, all resource requests are considered on the basis of "processor equivalents" as opposed to just processorsThis mechanism normalizes large-memory jobs to an equivalent number of CPUs (processor equivalants, PE) so that NCU allocations may be enforced. 
+
The big memory machines are available only via the HiPerGator "bigmem" partition. To submit jobs to this partition you will need to add the following directive to your job submission script.
 +
   #SBATCH --partition=bigmem
  
PE = MAX(ProcsRequestedByJob / TotalConfiguredProcs, MemoryRequestedByJob / TotalConfiguredMemory) * Total- ConfiguredProcs
+
Since our regular nodes have 1TB of available memory we do not recommend using bigmem nodes for jobs with memory requests lower than that.
 
 
The total number of configured processors and the total amount of configured memory vary slightly as machines are added to and removed from the batch system but, in general, each is fairly constant.  This means that jobs requesting only a single processor (core) but, say, 148 GB of RAM will be converted to an equivalent number or processors.  If the PE for your job exceeds your group's NCU allocation, your job will not run.
 
 
 
Once submitted, you may see the PE value assigned to your job via "checkjob <job id>".
 

Latest revision as of 15:33, 14 July 2022

Research Computing currently maintains the following resources for calculations requiring large amounts of physical memory.

Large Memory Servers
Nodes Architecture Vendor Processor Frequency Cores SLURM Memory(GB)
1 amd64 AMD EPYC 7702 2.0 128 4023
2 amd64 Intel E7-8850 3.2 192 1509

For more details see automatically generated data in Available Node Features

The big memory machines are available only via the HiPerGator "bigmem" partition. To submit jobs to this partition you will need to add the following directive to your job submission script.

 #SBATCH --partition=bigmem

Since our regular nodes have 1TB of available memory we do not recommend using bigmem nodes for jobs with memory requests lower than that.