Skip to content

Conversation

@ronlieb
Copy link
Collaborator

@ronlieb ronlieb commented Nov 30, 2025

supports flags:
-v:
-vv
-vvv
-taskset: use taskset
-numactl: use numactl(D)
-nobind: use nobind

lscpu numa info
NUMA node0 CPU(s): 0-7
NUMA node1 CPU(s): 8-15

rocm-smi --showtoponuma info
GPU[0] : (Topology) Numa Node: 6
GPU[0] : (Topology) Numa Affinity: 6
GPU[1] : (Topology) Numa Node: 6
GPU[1] : (Topology) Numa Affinity: 6

supports flags:
     -v:
     -vv
     -vvv
     -taskset: use taskset
     -numactl: use numactl(D)
     -nobind:  use nobind

 lscpu numa info
    NUMA node0 CPU(s):                       0-7
    NUMA node1 CPU(s):                       8-15

 rocm-smi --showtoponuma info
  GPU[0]          : (Topology) Numa Node: 6
  GPU[0]          : (Topology) Numa Affinity: 6
  GPU[1]          : (Topology) Numa Node: 6
  GPU[1]          : (Topology) Numa Affinity: 6
@z1-cciauto
Copy link
Collaborator

@ronlieb ronlieb requested a review from lfmeadow November 30, 2025 21:25
@z1-cciauto
Copy link
Collaborator

@ronlieb
Copy link
Collaborator Author

ronlieb commented Dec 2, 2025

next round will move to using amd-smi

  future options:
    #  -h   Print this help message and exit
    #  -md  Set number of desired devices for multi-device mode, default=1
    #  -s   suppress output, often useful in benchmarking
    #  -q   suppress output, quiet, alias of -s, same as GPURUN_VERBOSE=0
    #  -m   use numactl membind to CPUs in same NUMA domain. Note: Allocation
    #       fails when not enough memory available on these nodes.
    #  -l   use numactl localalloc to CPUs in same NUMA domain. Note: If
    #       memory cannot be allocated, alloc falls back to other nodes.
    #  -nr  use numactl ROCR_VISIBLE_DEVICES
    #  -nm  use numactl OMPI_COMM_WORLD_LOCAL_RANK
    #  --version Print version of gpurun and exit
@z1-cciauto
Copy link
Collaborator

@z1-cciauto
Copy link
Collaborator

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants