O narzędziu
RunPod — pay-as-you-go GPU cloud z 3 tierami. Community Cloud (Spot): od $0.22/h (RTX 3090) do $3.49/h (H100 SXM). Secure Cloud (On-Demand): H100 $2.39/h, H200 $3.59/h, gwarantowana availability. Serverless: pay-per-second (H100 ~$15/h, H200 ~$20/h). 40-60% taniej niż AWS/GCP on-demand. Pay-per-second billing.
Funkcje dodatkowe
▶Community Cloud (Spot)
Spot instances od $0.22/h (RTX 3090) do $3.49/h (H100 SXM) — najtansze GPU w branzy. Idealne dla non-critical workloads, ktore moga byc preempted bez utraty danych.
▶Secure Cloud (On-Demand)
Guaranteed availability bez ryzyka preemption — H100 $2.39/h, H200 $3.59/h, A100 $1.99/h. Dedykowana infrastruktura dla production workloads wymagajacych stabilno ranego.
▶Serverless GPU
Per-request workloads z pay-per-second billing — H100 $4.18/s (~$15/h), H200 $5.58/s (~$20/h). Idealny model dla inference APIs z nieregularnym ruchem.
▶Full GPU Range
Pelne spektrum NVIDIA GPU: H200 141GB, H100 80GB, A100 80GB, L40S, A6000, RTX 3090, RTX 4090. Wszystkie GPU dostepne w obu tier'ach (Spot i On-Demand).
▶Pay-per-second Billing
Rozliczanie tylko za rzeczywisty czas uzycia — no upfront, no idle charges, no minimum commitments. 40-60% taniej niz AWS/GCP on-demand pricing.
▶Network Volumes
Persistent storage attached do podow — dane przetrwaja restart instancji. Idealne do checkpointow modeli i datasetow podczas dlugoterminowych training runs.
▶Custom Docker Images
Mozliwosc uruchamiania wlasnych Docker images z pelnym CUDA stack i custom dependencies. Zero limit na container size i wbudowanymi toolami.
▶SSH Access
Pelny SSH access do uruchomionych pods — debug live, install dodatkowe biblioteki, edytuj kod w realtime. Eliminuje koszt rebuild'u containera dla drobnych zmian.
▶Templates dla frameworkow
Pre-built templates dla popularnych frameworkow ML: PyTorch, TensorFlow, JAX, Hugging Face, vLLM, ComfyUI. Setup w kilka sekund zamiast godzin manualnej konfiguracji.
▶API + CLI
REST API i CLI tool do automatyzacji deploymentu, scaling'u i monitorowania. Pozwala integrowac RunPod z CI/CD pipelines i orchestracja kubernetes.
✓ Zalety
Cennik
- •Community Cloud (Spot): od $0.22/h (RTX 3090) do $3.49/h (H100 SXM).
- •Secure Cloud (On-Demand): H100 $2.39/h, H200 $3.59/h, A100 $1.99/h.
- •Serverless: pay-per-second.
- •H100 $4.18/s (~$15/h), H200 $5.58/s (~$20/h).
- •Pay-per-second billing (no upfront, no idle charges).
- •40-60% taniej niż AWS/GCP.
Tiery compute
- •Community Cloud (Spot): od $0.22/h RTX 3090, do $3.49/h H100 SXM.
- •Secure Cloud (On-Demand): guaranteed availability, H100 $2.39/h, H200 $3.59/h.
- •Serverless: per-request workloads, billed by active compute time (per second).
Sprzęt GPU
- •NVIDIA H200 141GB: $3.59/h on-demand.
- •NVIDIA H100 80GB: $2.39/h on-demand, $3.49/h SXM.
- •NVIDIA A100 80GB: $1.99/h.
- •NVIDIA L40S, A6000, RTX 3090/4090.
- •Wszystkie GPU dostępne też w Spot tier (taniej).
Serverless
- •Pay-per-second pricing (no upfront).
- •H100 (80GB): $4.18/s.
- •H200 (141GB): $5.58/s.
- •Billed od start workera do full stop (rounded up to nearest second).