The Fractional GPU Cloud · Public Beta

Rent the GPU slice, not the card

Why rent a whole H100 to serve a 7B? Slice one into up to 7 hardware-isolated GPUs — NVIDIA or AMD — and pay for the fraction you use. Up to 8× cheaper, no CapEx.

Deploy your AI workloads, serverless

Containers, agents, and inference endpoints on one platform. Deploy from the dashboard and get a live endpoint that scales to zero when idle.

Serverless inference. Pay by the second.

OpenAI, Anthropic, and Ollama-compatible endpoints that scale to zero when idle. Cold starts under 3 seconds — you never pay for silence.

Deploy in 3 steps

Compute2× H100 sliceeu-west-1

EnginevLLMServing runtime

ModelDeepSeek R1671B params

llama-r1.istidlal.app

Live

H100 slice$0.40/hrFull H100$2.80/hrA100 80GB$1.90/hrL40S$0.95/hrA10G$0.38/hrT4$0.15/hrH100 slice$0.40/hrFull H100$2.80/hrA100 80GB$1.90/hrL40S$0.95/hrA10G$0.38/hrT4$0.15/hr

Serverless inference

pay / second

OpenAIcompatible API

Anthropiccompatible API

Ollamacompatible API

Point your existing SDK at us — no code changes.

Warm

Cold

<3s cold start

Fractional GPUs

up to 7 tenants / GPU

Qwen3 8B

FP8 · 8.7 GB

10 GB slice$0.40/hr

H100

NVIDIA

Gemma 4 12B

INT4 · 8.0 GB

20 GB slice$0.80/hr

MI355X

AMD

Phi-4 14B

FP8 · 14.7 GB

40 GB slice$1.60/hr

B300

NVIDIA

Hardware-isolated NVIDIA & AMD. Zero shared bandwidth.

10 GB slice from $0.40/hrvs $6.98/hr dedicated H100

Request accessSee pricing