Skip to main content

Resources

The targon.Resources class centralizes available compute tier identifiers. Use these string constants when selecting resource_name for workloads and rentals.

from targon import Resources

resource_name = Resources.H200_SMALL  # "h200-small"

CPU Tiers​

ConstantValueTypical Workloads
Resources.CPU_SMALLcpu-smallBackground jobs, lightweight services.
Resources.CPU_MEDIUMcpu-mediumData processing, lightweight AI workloads.
Resources.CPU_LARGEcpu-largeBatch processing, API backends with steady load.
Resources.CPU_XLcpu-xlargeHeavy CPU-bound workloads.

GPU Tiers​

H200​

ConstantValueTypical Workloads
Resources.H200_SMALLh200-smallPrompt serving, small LLMs, diffusion warmups.
Resources.H200_MEDIUMh200-mediumMid-sized model inference, fine-tuning jobs.
Resources.H200_LARGEh200-largeHigh-throughput inference, multi-modal workloads.
Resources.H200_XLh200-xlargeTraining loops, multi-billion parameter models.

H100​

ConstantValue
Resources.H100_SMALLh100-small
Resources.H100_MEDIUMh100-medium
Resources.H100_LARGEh100-large
Resources.H100_XLh100-xlarge

B200​

ConstantValue
Resources.B200_SMALLb200-small
Resources.B200_MEDIUMb200-medium
Resources.B200_LARGEb200-large
Resources.B200_XLb200-xlarge

B300​

ConstantValue
Resources.B300_SMALLb300-small
Resources.B300_MEDIUMb300-medium
Resources.B300_LARGEb300-large
Resources.B300_XLb300-xlarge

RTX6000B and RTX4090​

ConstantValue
Resources.RTXP6000B_SMALLrtx6000b-small
Resources.RTX4090_SMALLrtx4090-small
Resources.RTX4090_MEDIUMrtx4090-medium
Resources.RTX4090_LARGErtx4090-large

All constants are defined in the SDK:

# targon-sdk/libs/python/src/targon/core/resources.py
class Resources:
    CPU_SMALL = "cpu-small"
    CPU_MEDIUM = "cpu-medium"
    CPU_LARGE = "cpu-large"
    CPU_XL = "cpu-xlarge"
    H200_SMALL = "h200-small"
    # ...

See the Compute API reference for the full list of tier identifiers.

Selecting Resources​

  • Choose a CPU tier for CPU-bound jobs, orchestration, or lightweight services.
  • Choose a GPU tier for accelerated workloads (PyTorch, TensorRT, vLLM, etc.).
  • Run targon inventory or client.inventory.capacity() to see which tiers are available and their current hourly cost.

See the Compute resources guide for in-depth recommendations.