Cycling through all six. Tap any point to stop.

Measured on six things

Rent cloud GPUs by the second at all-in prices, payable i...

Growth

Specific wedge — per-second all-in pricing, health-check-gated billing, EU regions — aimed at indie researchers and small startups. Crypto-only payment caps the addressable audience.

Screenshots

1 / 1

GPUBank

About this app

GPUBank is a cloud GPU rental platform that bills by the second, not the hour, and folds everything a serious AI workload needs into a single hourly price. You pick a card — B200, H200, H100 SXM/PCIe, RTX PRO 6000, A100, L40S, AMD MI300X, RTX 5090 or RTX 4090 — and the quoted rate already covers the vCPUs, RAM, local NVMe scratch disk, a public IPv4 address and unmetered traffic. The only possible extra line is a persistent disk at $0.03/GB-month, which keeps your data alive between runs. Cards are spread across four regions (Paris, Frankfurt, Ashburn and Singapore), with paid launches from H100 down to $1.85/h and RTX 4090 at $0.23/h. Around 25 ready-to-run templates — PyTorch, vLLM, SGLang, ComfyUI, Ollama, Axolotl, Unsloth, JupyterLab and more — come pinned to tested driver and CUDA versions and are re-verified on real GPUs every few days. Billing only starts after a health check passes, so failed launches cost nothing. Payment works off a prepaid balance topped up with crypto (B...

Key features

1Per-second billing with all-in pricing: vCPU, RAM, NVMe, IPv4 and egress included in the hourly rate.
2Billing starts only after the machine passes its health check — failed launches cost nothing.
325 ready-to-run templates (PyTorch, vLLM, SGLang, ComfyUI, Ollama, Axolotl, Unsloth, JupyterLab) re-verified on real GPUs every three days.

Target customer profile

Drawn from the product itself, not from a survey.

Demographic

Indie AI researchers and PhD students

Pain points

Tight grants, hourly minimums, GPU outages wasting budget

Primary needs

Cheap per-second billing, spill-proof disks, spend caps

Demographic

ML engineers at small startups

Pain points

Overpriced hyperscalers, slow provisioning, hidden fees

Primary needs

Fast H100/H200 access, API/CLI automation, transparent bills

Demographic

AI creators and image/video artists

Pain points

Local GPUs too weak for FLUX/ComfyUI, cloud setup is fiddly

Primary needs

Ready templates, 24–96GB cards, persistent model storage

Demographic

EU-based dev teams with data residency needs

Pain points

US-only clouds, GDPR risk, cross-region complexity

Primary needs

EU regions, readable DPA, portable disks across regions

Scout Review

Written by AI from measured evidence, scored out of 100.

GPUBank is a well-executed indie GPU cloud with a clear thesis: all-in per-second pricing, health-check-gated billing, EU regions, and a template library that's actually re-tested on real hardware. The site is fast (Lighthouse 82), the security headers are complete (12/12), and the docs quickstart is better than most funded competitors'. The pricing table is honest and the '45% cheaper than average' claim is at least sourced to a date and a provider count. The two real problems are structural, not cosmetic: crypto-only top-up from $25 shuts out the card-paying academic and startup audience the product explicitly targets, and the missing compliance layer (no SOC 2, ISO, or pen-test disclosure) will slow down any buyer whose procurement team asks. Accessibility is close but has three fixable failures. For an indie researcher who already holds crypto, this is a genuinely good option. For everyone else, it's a great product behind a payment wall that doesn't fit them.

Measure by measure

Usability

Six-step quickstart with copyable CLI, documented API and auto-filled templates make launch fast. Crypto-only top-up from $25 before any GPU is a real friction point.

Design

Distinctive receipt-style hero and live availability counter give the site real personality; dense pricing table is functional but heavy. Polished for an indie GPU cloud.

Speed

Lighthouse desktop 82/100 with LCP 1.3s and near-zero CLS; TBT 269ms is the only blemish. Advertised 48s median launch backs the speed pitch.

Security

Full security header suite (HSTS, CSP, frame protection, Permissions-Policy) and SPF/DMARC — 12/12 probe points. No SOC 2, ISO or pen-test disclosure found; crypto-only payments add trust friction.

Accessibility

Lighthouse a11y 94/100 with three real failures: prohibited ARIA, low contrast, and mismatched accessible names. No WCAG statement or a11y page found.

Growth

Specific wedge — per-second all-in pricing, health-check-gated billing, EU regions — aimed at indie researchers and small startups. Crypto-only payment caps the addressable audience.

Improvement areas

Crypto-only top-up from $25 excludes card-paying researchers and startups — the stated core audience.
No SOC 2, ISO 27001, pen-test disclosure or bug bounty page for procurement conversations.
Three Lighthouse accessibility failures (ARIA, contrast, accessible names) and no WCAG statement.

Design & Usability

GPUBank's site has a point of view — the receipt hero, the live GPU counter, the '45% cheaper than average' bar — that most indie infra tools never bother with. The pricing table is dense but honest: per-hour, per-second and per-month columns, spot and reserved tiers, and every card's region list. Usability is where it gets interesting. The docs quickstart is genuinely good: six numbered steps, copyable CLI, a real `gpubank launch` example with output. Templates auto-fill the launcher. But step two is 'top up from $25 in crypto,' and that single decision gates the entire funnel. For a PhD student on a grant card or a small startup with a corporate Amex, that's a hard stop. The product is easy to use once you're in; getting in is the work.

Speed & Security

Speed is a genuine strength. Lighthouse desktop 82/100, LCP 1.3s, CLS 0.001 — the site loads fast and the advertised 48s median launch (p95 71s) is the kind of number that wins converts from slower clouds. TBT of 269ms is the only soft spot. Security is strong on the wire: 12/12 header checklist points including HSTS, CSP, frame protection and Permissions-Policy, plus SPF and DMARC. Auth is email+password with a six-digit code, or GitHub/Google OAuth. What's missing is the compliance layer enterprise buyers ask for — no SOC 2, no ISO 27001, no pen-test disclosure, no bug bounty. For the indie/small-startup target that's probably fine; for the 'ML engineers at small startups' persona it's a question they'll ask eventually. Crypto-only payments are the bigger trust question.

Accessibility & Growth

Accessibility lands at 94/100 on Lighthouse with three real failures: prohibited ARIA attributes, insufficient color contrast, and visible text labels without matching accessible names. Those are the kinds of bugs that hit screen-reader users specifically, and there's no WCAG statement or a11y page to show the team is tracking them. Growth is the more interesting story. The wedge is real and specific: per-second billing, all-in pricing that folds vCPU/RAM/NVMe/IPv4/egress into one number, billing that only starts after a health check, a $0.03/GB-month persistent disk, and 25 templates re-verified on real GPUs every three days. That's a coherent pitch against Vast.ai (variable reliability), RunPod (per-second but template-light), Lambda (reserved-cluster focus) and CoreWeave (enterprise-only). The EU regions and readable DPA target a segment the US-only clouds underserve. The limit is the crypto-only top-up — it narrows the funnel to a crypto-native slice of an audience that mostly pays with cards and grants.

Conclusion

If you're crypto-native, EU-based, and tired of hyperscaler markups, GPUBank is worth a real look — the per-second billing and health-check-gated meter are the right primitives, and the template library is more than marketing. If you're paying with a grant card or a company Amex, wait for card payments before you build a workflow on it. The product is ready; the payment rail isn't.

Suggestions for improvement

Add card and invoice payment alongside crypto — the stated audience (PhD students, small startups) mostly can't pay in BTC.
Publish a SOC 2 Type I roadmap or a third-party pen-test summary; even a self-attestation page would help procurement conversations.
Fix the three Lighthouse a11y failures (ARIA, contrast, accessible names) and add a short WCAG conformance statement.
Offer a small free credit or trial tier so researchers can validate launch speed and template quality before committing $25.

How we score

Alternatives comparison

Named competitors, point by point. Nobody paid to appear here or to be left out.

	GPUBank	Vast.ai	RunPod	Lambda	CoreWeave
Billing granularity	Per-second, all-in (vCPU/RAM/NVMe/IPv4/egress included)	Per-second on most hosts; variable by provider	Per-second	Per-minute / per-hour on-demand	Per-hour, reserved-heavy
Payment methods	Crypto only (BTC, USDT, ETH, XMR), from $25	Card and crypto	Card and crypto	Card and invoice	Invoice / enterprise billing
Ready-to-run templates	25 templates, re-verified on real GPUs every 3 days	Docker images, community-provided	Large template library + serverless	Limited; mostly base images	Kubernetes-native, bring your own stack
EU regions / data residency	Paris, Frankfurt, Ashburn, Singapore	Global peer hosts, variable	Multiple regions incl. EU	US-centric	US and EU regions
Compliance certifications	None disclosed (strong headers, no SOC 2/ISO)	Limited public disclosure	SOC 2 Type II	SOC 2 Type II	SOC 2, ISO 27001, HIPAA