DCS / Compute / VM Instances

Per-second billing · pre-configured

GPU instances in 30 seconds. Pay per second.

Launch H200, H100, A100, L40S, or RTX 4090 instances on-demand. Pre-configured templates (PyTorch, vLLM, Stable Diffusion, ComfyUI) boot instantly. SSH + Jupyter + Web terminal. Pay per second of actual GPU time. Receipt-signed session start + stop.

Launch an instance → Browse templates

GPU catalog

On-demand prices. Per-second billing.

Reserved (longer-term) and reverse-auction (cheapest) modes also available. Below shows on-demand rates.

GPU	VRAM	vCPU	RAM	Storage	On-demand
H200	141 GB	32	256 GB	500 GB NVMe	$24.00/hr
H100	80 GB	26	200 GB	400 GB NVMe	$22.00/hr
A100	80 GB	22	156 GB	300 GB NVMe	$14.00/hr
L40S	48 GB	16	96 GB	200 GB NVMe	$5.40/hr
RTX 5090	32 GB	12	64 GB	200 GB NVMe	$2.40/hr
RTX 4090	24 GB	10	48 GB	150 GB NVMe	$1.20/hr

Reverse-auction mode is ~12.75% cheaper on average. See ROI Calculator.

Templates

Pre-configured environments. Verified by DCS.

🔥

PyTorch 2.4

CUDA 12.4, common HF libraries, Jupyter. Most-used template.

⚡

vLLM

OpenAI-compatible inference server. Mount your weights, get an endpoint.

🎨

Stable Diffusion

ComfyUI + AUTOMATIC1111. Drop in checkpoints + LoRAs. Web UI ready.

🎬

Video render

Blender, FFmpeg, RunwayML-compatible. Headless render farm config.

🤗

HuggingFace

Transformers + Datasets + Accelerate. Trainer setup pre-wired.

🧪

Research base

Ubuntu 22.04 + CUDA + nothing else. Build your own from a clean slate.

🤖

DCS Agent host

Run a DCS Agent Studio composition on dedicated GPU. Auto-receipt signing.

📦

Custom image

Bring your own Docker image. Pull from public registry or DCS-hosted private.

Features

What's standard on every instance.

⚡

30-sec boot

Pre-warmed template images. Most launches complete in <30 seconds.

⏱

Per-second billing

Stop saves money instantly. No "rounded up to the hour" gotchas.

💾

Persistent volume

Storage survives stop/start. Don't lose your weights when you cycle the GPU.

🔐

SSH + Jupyter + Web

Three ways in. SSH key uploaded at launch. Jupyter token in dashboard.

📜

Session receipts

Start, stop, snapshot, destroy — all signed. Audit-grade billing trail.

🌍

Multi-region

US-East, US-West, EU-W, Asia-South. Per-region pricing identical; latency-based routing.

FAQ

Common questions.

How is this different from RunPod / Lambda Labs?

DCS adds receipt-signed sessions + the reverse-auction lane for ~13% cheaper average. Templates are about the same. If you need audit grade, this is the right place. If not, any provider works.

What's the boot time honestly?

28-45 seconds for most common templates. ~90 seconds for cold-start regions. We're working on a "warm pool" feature to get this below 10 sec for hot templates.

SSH key rotation?

Per-instance ephemeral keys by default. Re-launch = new keypair. Persistent keypair option in account settings.

Snapshots?

Full instance snapshot (OS + persistent volume) for $0.05/GB/mo storage. Restore launches new instance from snapshot in <60s.

What if I forget to stop my instance?

Idle-timeout auto-stop (configurable, default 30 min of no SSH/GPU activity). Plus weekly cost alerts via email.