CPU, GPU, or mixed. Containerized jobs. Long-running or batch. The network handles scheduling and fault tolerance.
Every job produces a cryptographic receipt. Output hashes are signed and published. Audit any result.
Spike to thousands of nodes when you need them. Pay only for what you use. No commitments.
H100, A100, MI300X. Multi-GPU jobs. Distributed training across nodes.
Mount distributed storage to your jobs. Stream inputs and outputs. Pay per GB.
No credit card. Stake to access compute at a discount. Burn on idle. Earn on usage.
If you can containerize it, you can run it on Aigarth.
Molecular dynamics, climate models, CFD, genomics.
Monte Carlo, risk sims, backtesting at scale.
Animation, VFX, architectural visualization.
Process terabytes nightly. Cheaper than reserved cloud.
Multi-node model training. Horovod, DeepSpeed, FSDP.
Transcoding, batch processing, watermark application.
Submit a job. The scheduler finds the lowest-cost, lowest-latency workers. Outputs are signed and replicated.
# Submit a job to the network
import aigarth
client = aigarth.Client(api_key="sk-...")
job = client.compute.submit(
image="docker.io/myorg/sim:latest",
command=["./run", "--scale", "1000"],
gpu="H100",
replicas=64,
timeout="6h",
)
# Poll for completion
result = job.wait()
print(f"Job complete: {result.output_hash}")