Is this page helpful?

MPS v3 Configuration File#

MPS v3 can load its full startup configuration from a TOML file with -a/--apply:

nvidia-cuda-mps-control -d -p 3 -a /path/to/config.toml

Configuration is applied once at startup; there is currently no persistence of subsequent CLI changes back to disk, so servers, namespaces, and partitions created after startup only exist for the lifetime of the daemon unless they are also added to the configuration file.

schema_version = "1.0"

[servers.training]
allowed_devices          = "GPU-abcd-1234,GPU-efgh-5678"
pinned_memory_limit      = "10G"
active_thread_percentage = "50"
client_priority          = "0"
uid                      = 1000
locality_domains         = false

[servers.training.namespaces.high_priority]
pinned_memory_limit      = "5G"
active_thread_percentage = "25"

[[servers.training.devices]]
uuid = "GPU-74d43ed3-cdf7-e667-3644-bf5b4f46ed65"

  [[servers.training.devices.sm_partitions]]
  name              = "large"
  chunks            = 4
  include_remainder = false

[features.memacct]
enabled   = true
audit_log = true

[features.context-share]
enabled        = true
default_socket = "on"

  • [servers.<name>] accepts allowed_devices, pinned_memory_limit, active_thread_percentage, client_priority, uid, and locality_domains (boolean only). Any other key is a configuration error.

  • [servers.<name>.namespaces.<name>] accepts only pinned_memory_limit, active_thread_percentage, and client_priority. A namespace named default refers to the server’s built-in default namespace rather than creating a new one. Refer to MPS v3 Namespaces for the equivalent CLI commands.

  • [[servers.<name>.devices]] requires uuid and optionally nests [[servers.<name>.devices.sm_partitions]] entries, each requiring name and chunks, with optional include_remainder (default false).

  • [features.<name>] tables are read per-feature; see the Features section in MPS v3 Interface for the keys each feature accepts.

  • TOML keys use underscores; the equivalent CLI flags use dashes (for example, pinned_memory_limit in the file corresponds to --pinned-memory-limit on the command line).