vastai search offers

Search for instance types using custom query

Usage

vastai search offers [--help] [--api-key API_KEY] [--raw] <query>

Arguments

query · string · required

Query to search for. default: ‘external=false rentable=true verified=true’, pass -n to ignore default

Options

--type · string · default: "on-demand"

Show ‘on-demand’, ‘reserved’, or ‘bid’(interruptible) pricing. default: on-demand

--interruptible · string

Alias for —type=bid

--bid · string

Alias for —type=bid

--reserved · string

Alias for —type=reserved

--on-demand · string

Alias for —type=on-demand

--no-default · boolean

Disable default query

--new · boolean

New search exp

--limit · integer

--storage · number · default: "5.0"

Amount of storage to use for pricing, in GiB. default=5.0GiB

--order · string · default: "score-"

Comma-separated list of fields to sort on. postfix field with - to sort desc. ex: -o ‘num_gpus,total_flops-’. default=‘score-‘

Description

Query syntax:

query = comparison comparison… comparison = field op value field = <name of a field> op = one of: <, <=, ==, !=, >=, >, in, notin value = <bool, int, float, string> | ‘any’ | [value0, value1, …] bool: True, False

note: to pass ’>’ and ’<’ on the command line, make sure to use quotes note: to encode a string query value (ie for gpu_name), replace any spaces ’ ’ with underscore ’_‘

Examples

# search for somewhat reliable single RTX 3090 instances
    vastai search offers 'reliability > 0.98 num_gpus=1 gpu_name=RTX_3090'

    # search for datacenter gpus with minimal compute_cap and total_flops
    vastai search offers 'compute_cap > 610 total_flops > 5 datacenter=True'

    # search for reliable 4 gpu offers in Taiwan or Sweden
    vastai search offers 'reliability>0.99 num_gpus=4 geolocation in [TW,SE]'

    # search for reliable RTX 3090 or 4090 gpus NOT in China or Vietnam
    vastai search offers 'reliability>0.99 gpu_name in ["RTX 4090", "RTX 3090"] geolocation notin [CN,VN]'

    # search for machines with nvidia drivers 535.86.05 or greater (and various other options)
    vastai search offers 'disk_space>146 duration>24 gpu_ram>10 cuda_vers>=12.1 direct_port_count>=2 driver_version >= 535.86.05'

    # search for reliable machines with at least 4 gpus, unverified, order by num_gpus
    vastai search offers 'reliability > 0.99  num_gpus>=4 verified=False' -o 'num_gpus-'

    # search for arm64 cpu architecture
    vastai search offers 'cpu_arch=arm64'

Available fields:

      Name                  Type       Description

    bw_nvlink               float     bandwidth NVLink
    compute_cap:            int       cuda compute capability*100  (ie:  650 for 6.5, 700 for 7.0)
    cpu_arch                string    host machine cpu architecture (e.g. amd64, arm64)
    cpu_cores:              int       # virtual cpus
    cpu_ghz:                Float     # cpu clock speed GHZ
    cpu_cores_effective:    float     # virtual cpus you get
    cpu_ram:                float     system RAM in gigabytes
    cuda_vers:              float     machine max supported cuda version (based on driver version)
    datacenter:             bool      show only datacenter offers
    direct_port_count       int       open ports on host's router
    disk_bw:                float     disk read bandwidth, in MB/s
    disk_space:             float     disk storage space, in GB
    dlperf:                 float     DL-perf score  (see FAQ for explanation)
    dlperf_usd:             float     DL-perf/$
    dph:                    float     $/hour rental cost
    driver_version:         string    machine's nvidia/amd driver version as 3 digit string ex. "535.86.05,"
    duration:               float     max rental duration in days
    external:               bool      show external offers in addition to datacenter offers
    flops_usd:              float     TFLOPs/$
    geolocation:            string    Two letter country code. Works with operators =, !=, in, notin (e.g. geolocation not in ['XV','XZ'])
    gpu_arch                string    host machine gpu architecture (e.g. nvidia, amd)
    gpu_max_power           float     GPU power limit (watts)
    gpu_max_temp            float     GPU temp limit (C)
    gpu_mem_bw:             float     GPU memory bandwidth in GB/s
    gpu_name:               string    GPU model name (no quotes, replace spaces with underscores, ie: RTX_3090 rather than 'RTX 3090')
    gpu_ram:                float     per GPU RAM in GB
    gpu_total_ram:          float     total GPU RAM in GB
    gpu_frac:               float     Ratio of GPUs in the offer to gpus in the system
    gpu_display_active:     bool      True if the GPU has a display attached
    has_avx:                bool      CPU supports AVX instruction set.
    id:                     int       instance unique ID
    inet_down:              float     internet download speed in Mb/s
    inet_down_cost:         float     internet download bandwidth cost in $/GB
    inet_up:                float     internet upload speed in Mb/s
    inet_up_cost:           float     internet upload bandwidth cost in $/GB
    machine_id              int       machine id of instance
    min_bid:                float     current minimum bid price in $/hr for interruptible
    num_gpus:               int       # of GPUs
    pci_gen:                float     PCIE generation
    pcie_bw:                float     PCIE bandwidth (CPU to GPU)
    reliability:            float     machine reliability score (see FAQ for explanation)
    rentable:               bool      is the instance currently rentable
    storage_cost:           float     storage cost in $/GB/month
    static_ip:              bool      is the IP addr static/stable
    total_flops:            float     total TFLOPs from all GPUs
    ubuntu_version          string    host machine ubuntu OS version
    verified:               bool      is the machine verified
    vms_enabled:            bool      is the machine a VM instance

Global Options

The following options are available for all commands:

Option	Description
--url URL	Server REST API URL
--retry N	Retry limit
--raw	Output machine-readable JSON
--explain	Verbose explanation of API calls
--api-key KEY	API key (defaults to ~/.config/vastai/vast_api_key)