machine_namestring nullable
price_per_gpunumber nullable
pending_price_per_hournumber nullable
price_change_effective_datestring date-time nullable
executor_ip_addressstring required
validator_hotkeystring required
gpu_countinteger nullable
available_gpu_countinteger nullable
uptime_in_minutesinteger nullable
miner_hotkeystring nullable
collateral_depositedboolean nullable
effective_upload_speed_mbpsnumber nullable
effective_download_speed_mbpsnumber nullable
reliability_scorenumber nullable
gpu_power_limitedboolean nullable
true when a GPU on this node runs under 95 % of its default power limit, so a rental gets less than the card's stock power; false when every GPU that reports a default is at it. null = cannot be judged: no GPU reports both limits, or a Lium default-job container is on the node and every GPU that reports sits at Pearl's own temporary cap (70 % of the default) or at the card's driver minimum, which Pearl's cap clamps to on cards whose minimum is above 70 % (RTX 4090, RTX 5090) — so a host limit at that minimum is not judged while any Lium or miner default job runs; a host limit between the minimum and the default is.
gpu_power_limit_winteger nullable
Current power limit in watts of the most reduced GPU; null when gpu_power_limited is null.
gpu_power_limit_default_winteger nullable
That GPU's default power limit in watts; null when gpu_power_limited is null.
distance_milenumber nullable
min_gpu_count_for_rentalinteger nullable
max_cuda_versionnumber nullable
is_whole_host_freeboolean nullable
has_no_pending_rentalboolean nullable
infiniband_fabric_rentableboolean
roce_fabric_rentableboolean
provider_display_namestring nullable
ncu_profiling_enabledboolean required
disk_health_okboolean nullable required
Whether the node's docker root takes writes (DAH-2929). True: no read-only mount over the
docker root and the write probe did not fail. False: one of those; the readings behind it are
in specs.disk_health. null: the validator has not probed this node this cycle.
on_infiniband_fabricboolean required
on_roce_fabricboolean required
nvlinkboolean nullable required
true when the validator saw every GPU pair joined by NVLink and confirmed it: P2P reads OK on every pair, every GPU has its NVLinks up and the topology covers every GPU. false when a pair is not on NVLink or a check contradicts it; null when the topology is unreported or unconfirmed, or the node has one GPU.
gpu_classstring nullable required
The GPU type this node is listed under: the canonical model name of the type, a value machine_names accepts and that matches every variant of the type (NVIDIA B300 SXM6 AC for a NVIDIA B300 SXM6 PC node). Equal to machine_name for a type with one model name; null without one.
gpu_class_displaystring nullable required
The GPU type's name for a filter or a list (NVIDIA B300 SXM6); the exact model stays in machine_name. Equal to machine_name for a type with one model name; null without one.