Original en anglais: https://argusic.com/servers

Clouds GPU et serveurs VPS, prix vérifiés

Vous préférez acheter une machine plutôt que la louer ? Consultez les laptops and desktops for programming, AI and data science.

Disclosure: When you buy through our links, we may earn a commission. Learn more›. Commissions never affect order or numbers. Prices are list rates with a verification state on every row: API-verified rows are refreshed daily from the provider's own pricing API and the response is stored as the source; Owner-verified rows were checked by hand against the provider's console or billing on the date shown; rows not yet verified say so plainly. Measured rows carry Argusic's own operational data from running its benchmark fleet, each number linked to its run evidence. Argusic runs on some of the providers listed (marked "Argusic runs here").

Suivi de 337 offres GPU de 114 fournisseurs dans 51 pays ; 226 avec un prix vérifié. Le moins cher (H100 vérifié, par heure GPU) : $1.33 chez Vast.ai (vérifié le 2026-11-04).

20 affichés sur 337 résultats

Vast.ai

GTX 1050 Ti (lowest of 1 listings)

MarketplaceSpainÀ la secondeTrafic sortant facturéCommunity

Peer-to-peer GPU marketplace: many GPU types, hosts in many countries, quality varies by host.

GTX 1050 Ti / 4 GB VRAM / 1x GPU / PCIE / 1 annonce

Rent or buyNo on-demand rental for this card was verified in the last 30 days. See it on Amazon detailsProduct prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.com at the time of purchase will apply to the purchase of this product.

Vérifié par API 2026-09-27 (il y a plus de 30 jours)

$0.023par heure GPU

Visiter le site

RunPodMesuréArgusic l'utilise

RTX A2000 (Community Cloud)

Spécialiste on-demandAustralia, Canada, Czechia +10À la secondeTrafic sortant gratuitCommunity

Per-second GPU pods on Community and Secure tiers, plus serverless endpoints. Argusic's benchmark fleet runs its CPU pods here.

RTX A2000 / 6 GB VRAM / 1x GPU / PCIE

Mesuré par ArgusicPods CPU (4 vCPU) : tarif horaire facturé à la création du pod, moyenne: $0.115/h (n=5295) preuvesPods CPU : de la création du pod au harness en cours d'exécution, médiane: 32 s (n=5295) preuves

Vérifié par API 2026-11-04

$0.12par heure GPU

Visiter le site

Thunder Compute

H100 (on-demand)

Spécialiste on-demandOn-demand

Low-cost A100 and H100, pay as you go.

H100 / 1x GPU

Vérifié par le propriétaire 2026-09-05 (il y a plus de 30 jours)

$3.20par heure GPU

Visiter le site

Vast.ai

Tesla T4 (lowest of 1 listings)

MarketplaceSouth KoreaÀ la secondeTrafic sortant facturéCommunity

Peer-to-peer GPU marketplace: many GPU types, hosts in many countries, quality varies by host.

T4 / 15 GB VRAM / 1x GPU / PCIE / 1 annonce

Vérifié par API 2026-11-03

$0.027par heure GPU

Visiter le site

RunPodMesuréArgusic l'utilise

RTX 3070 (Community Cloud)

Spécialiste on-demandAustralia, Canada, Czechia +10À la secondeTrafic sortant gratuitCommunity

Per-second GPU pods on Community and Secure tiers, plus serverless endpoints. Argusic's benchmark fleet runs its CPU pods here.

RTX 3070 / 8 GB VRAM / 1x GPU / PCIE

Mesuré par ArgusicPods CPU (4 vCPU) : tarif horaire facturé à la création du pod, moyenne: $0.115/h (n=5295) preuvesPods CPU : de la création du pod au harness en cours d'exécution, médiane: 32 s (n=5295) preuves

Rent or buyBuy at 473.04 USD, rent at 0.13 USD an hour: at 160 hours a month the rent total passes the buy price after 22.7 months.Before electricity, somewhere to put it, and resale.on Amazon, sold by New Cycle Electronics (as of 11/04/2026 06:09 PST) See it on Amazon detailsProduct prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.com at the time of purchase will apply to the purchase of this product.

Vérifié par API 2026-11-04

$0.13par heure GPU

Visiter le site

Vast.ai

Quadro P2000 (lowest of 1 listings)

MarketplaceUnited StatesÀ la secondeTrafic sortant facturéCommunity

Peer-to-peer GPU marketplace: many GPU types, hosts in many countries, quality varies by host.

Quadro P2000 / 5 GB VRAM / 1x GPU / PCIE / 1 annonce

Vérifié par API 2026-11-04

$0.030par heure GPU

Visiter le site

RunPodMesuréArgusic l'utilise

RTX A5000 (Community Cloud)

Spécialiste on-demandAustralia, Canada, Czechia +10À la secondeTrafic sortant gratuitCommunity

Per-second GPU pods on Community and Secure tiers, plus serverless endpoints. Argusic's benchmark fleet runs its CPU pods here.

RTX A5000 / 24 GB VRAM / 1x GPU / PCIE

Mesuré par ArgusicPods CPU (4 vCPU) : tarif horaire facturé à la création du pod, moyenne: $0.115/h (n=5295) preuvesPods CPU : de la création du pod au harness en cours d'exécution, médiane: 32 s (n=5295) preuves

Vérifié par API 2026-11-04

$0.16par heure GPU

Visiter le site

Vast.ai

RTX A2000 (lowest of 1 listings)

MarketplaceJapanÀ la secondeTrafic sortant facturéCommunity

Peer-to-peer GPU marketplace: many GPU types, hosts in many countries, quality varies by host.

RTX A2000 / 6 GB VRAM / 1x GPU / PCIE / 1 annonce

Vérifié par API 2026-11-04

$0.036par heure GPU

Visiter le site

RunPodMesuréArgusic l'utilise

RTX 3080 (Community Cloud)

Spécialiste on-demandAustralia, Canada, Czechia +10À la secondeTrafic sortant gratuitCommunity

Per-second GPU pods on Community and Secure tiers, plus serverless endpoints. Argusic's benchmark fleet runs its CPU pods here.

RTX 3080 / 10 GB VRAM / 1x GPU / PCIE

Mesuré par ArgusicPods CPU (4 vCPU) : tarif horaire facturé à la création du pod, moyenne: $0.115/h (n=5295) preuvesPods CPU : de la création du pod au harness en cours d'exécution, médiane: 32 s (n=5295) preuves

Rent or buyBuy at 608.82 USD, rent at 0.17 USD an hour: at 160 hours a month the rent total passes the buy price after 22.4 months.Before electricity, somewhere to put it, and resale.on Amazon, sold by IT Electronix (as of 11/04/2026 06:09 PST) See it on Amazon detailsProduct prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.com at the time of purchase will apply to the purchase of this product.

Vérifié par API 2026-11-04

$0.17par heure GPU

Visiter le site

Vast.ai

Tesla P4 (lowest of 1 listings)

MarketplacePolandÀ la secondeTrafic sortant facturéCommunity

Peer-to-peer GPU marketplace: many GPU types, hosts in many countries, quality varies by host.

Tesla P4 / 8 GB VRAM / 1x GPU / PCIE / 1 annonce

Vérifié par API 2026-10-26

$0.040par heure GPU

Visiter le site

RunPodMesuréArgusic l'utilise

RTX A4000 (Community Cloud)

Spécialiste on-demandAustralia, Canada, Czechia +10À la secondeTrafic sortant gratuitCommunity

Per-second GPU pods on Community and Secure tiers, plus serverless endpoints. Argusic's benchmark fleet runs its CPU pods here.

RTX A4000 / 16 GB VRAM / 1x GPU / PCIE

Mesuré par ArgusicPods CPU (4 vCPU) : tarif horaire facturé à la création du pod, moyenne: $0.115/h (n=5295) preuvesPods CPU : de la création du pod au harness en cours d'exécution, médiane: 32 s (n=5295) preuves

Vérifié par API 2026-11-04

$0.17par heure GPU

Visiter le site

Vast.ai

GTX 1060 (lowest of 1 listings)

MarketplaceFranceÀ la secondeTrafic sortant facturéCommunity

Peer-to-peer GPU marketplace: many GPU types, hosts in many countries, quality varies by host.

GTX 1060 / 6 GB VRAM / 1x GPU / PCIE / 1 annonce

Rent or buyThe only verified price for this card is a spot price, which can be taken away at any moment. The comparison uses on-demand prices, so it is not drawn here. See it on Amazon detailsProduct prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.com at the time of purchase will apply to the purchase of this product.

Vérifié par API 2026-11-04

$0.042par heure GPU

Visiter le site

RunPodMesuréArgusic l'utilise

RTX 3080 Ti (Community Cloud)

Spécialiste on-demandAustralia, Canada, Czechia +10À la secondeTrafic sortant gratuitCommunity

Per-second GPU pods on Community and Secure tiers, plus serverless endpoints. Argusic's benchmark fleet runs its CPU pods here.

RTX 3080 Ti / 12 GB VRAM / 1x GPU / PCIE

Mesuré par ArgusicPods CPU (4 vCPU) : tarif horaire facturé à la création du pod, moyenne: $0.115/h (n=5295) preuvesPods CPU : de la création du pod au harness en cours d'exécution, médiane: 32 s (n=5295) preuves

Rent or buyBuy at 1,195.00 USD, rent at 0.18 USD an hour: at 160 hours a month the rent total passes the buy price after 41.5 months.Before electricity, somewhere to put it, and resale.on Amazon, sold by Everlast Computers (as of 11/04/2026 06:09 PST) See it on Amazon detailsProduct prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.com at the time of purchase will apply to the purchase of this product.

Vérifié par API 2026-11-04

$0.18par heure GPU

Visiter le site

Vast.ai

GTX TITAN X (lowest of 3 listings)

MarketplaceSouth KoreaÀ la secondeTrafic sortant facturéCommunity

Peer-to-peer GPU marketplace: many GPU types, hosts in many countries, quality varies by host.

GTX TITAN X / 12 GB VRAM / 1x GPU / PCIE / 3 annonces

Rent or buyThe only verified price for this card is a spot price, which can be taken away at any moment. The comparison uses on-demand prices, so it is not drawn here. See it on Amazon detailsProduct prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.com at the time of purchase will apply to the purchase of this product.

Vérifié par API 2026-11-04

$0.042par heure GPU

Visiter le site

RunPodMesuréArgusic l'utilise

RTX 4000 Ada SFF (Community Cloud)

Spécialiste on-demandAustralia, Canada, Czechia +10À la secondeTrafic sortant gratuitCommunity

Per-second GPU pods on Community and Secure tiers, plus serverless endpoints. Argusic's benchmark fleet runs its CPU pods here.

RTX 4000 Ada SFF / 20 GB VRAM / 1x GPU / PCIE

Mesuré par ArgusicPods CPU (4 vCPU) : tarif horaire facturé à la création du pod, moyenne: $0.115/h (n=5295) preuvesPods CPU : de la création du pod au harness en cours d'exécution, médiane: 32 s (n=5295) preuves

Vérifié par API 2026-11-04

$0.18par heure GPU

Visiter le site

Vast.ai

GTX 1660 Ti (lowest of 1 listings)

MarketplaceNorwayÀ la secondeTrafic sortant facturéCommunity

Peer-to-peer GPU marketplace: many GPU types, hosts in many countries, quality varies by host.

GTX 1660 Ti / 6 GB VRAM / 1x GPU / PCIE / 1 annonce

Vérifié par API 2026-11-04

$0.043par heure GPU

Visiter le site

RunPodMesuréArgusic l'utilise

RTX 4070 Ti (Community Cloud)

Spécialiste on-demandAustralia, Canada, Czechia +10À la secondeTrafic sortant gratuitCommunity

Per-second GPU pods on Community and Secure tiers, plus serverless endpoints. Argusic's benchmark fleet runs its CPU pods here.

RTX 4070 Ti / 12 GB VRAM / 1x GPU / PCIE

Mesuré par ArgusicPods CPU (4 vCPU) : tarif horaire facturé à la création du pod, moyenne: $0.115/h (n=5295) preuvesPods CPU : de la création du pod au harness en cours d'exécution, médiane: 32 s (n=5295) preuves

Rent or buyBuy at 1,299.00 USD, rent at 0.19 USD an hour: at 160 hours a month the rent total passes the buy price after 42.7 months.Before electricity, somewhere to put it, and resale.on Amazon, sold by Techno Sphere (as of 11/04/2026 07:05 PST) See it on Amazon detailsProduct prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.com at the time of purchase will apply to the purchase of this product.

Vérifié par API 2026-11-04

$0.19par heure GPU

Visiter le site

Vast.ai

GTX 1080 (lowest of 2 listings)

MarketplaceAustria, South KoreaÀ la secondeTrafic sortant facturéCommunity

Peer-to-peer GPU marketplace: many GPU types, hosts in many countries, quality varies by host.

GTX 1080 / 8 GB VRAM / 1x GPU / PCIE / 2 annonces

Rent or buyThe only verified price for this card is a spot price, which can be taken away at any moment. The comparison uses on-demand prices, so it is not drawn here. See it on Amazon detailsProduct prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.com at the time of purchase will apply to the purchase of this product.

Vérifié par API 2026-11-04

$0.044par heure GPU

Visiter le site

RunPodMesuréArgusic l'utilise

RTX A4500 (Community Cloud)

Spécialiste on-demandAustralia, Canada, Czechia +10À la secondeTrafic sortant gratuitCommunity

Per-second GPU pods on Community and Secure tiers, plus serverless endpoints. Argusic's benchmark fleet runs its CPU pods here.

RTX A4500 / 20 GB VRAM / 1x GPU / PCIE

Mesuré par ArgusicPods CPU (4 vCPU) : tarif horaire facturé à la création du pod, moyenne: $0.115/h (n=5295) preuvesPods CPU : de la création du pod au harness en cours d'exécution, médiane: 32 s (n=5295) preuves

Vérifié par API 2026-11-04

$0.19par heure GPU

Visiter le site

Vast.ai

GTX 1660 S (lowest of 1 listings)

MarketplaceUnited StatesÀ la secondeTrafic sortant facturéCommunity

Peer-to-peer GPU marketplace: many GPU types, hosts in many countries, quality varies by host.

GTX 1660 S / 6 GB VRAM / 1x GPU / PCIE / 1 annonce

Rent or buyThe only verified price for this card is a spot price, which can be taken away at any moment. The comparison uses on-demand prices, so it is not drawn here. See it on Amazon detailsProduct prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon.com at the time of purchase will apply to the purchase of this product.

Vérifié par API 2026-11-04

$0.046par heure GPU

Visiter le site

Ce hub regroupe les clouds GPU et les fournisseurs VPS/CPU au même endroit. Les prix sont les tarifs catalogue avec un état de vérification; les lignes marquées Mesuré comportent les données opérationnelles d'Argusic (temps de démarrage, coût réel facturé, taux d'échec) issues de l'exécution de notre flotte de benchmarks, chaque nombre étant lié aux preuves de son exécution. Nous ne vendons pas d'hébergement et n'appartenons à aucun fournisseur.

Comment nous mesurons et classons

Default order is price, ascending, with verified prices first. Argusic-measured columns come from our production workloads, not synthetic tests: every pod we rent logs its spin-up seconds, preemptions, and the rate it was billed at. Community signals never affect order, and neither do affiliate commissions. Reviews are intentionally absent: this page publishes what we measured and verified, not opinions.

GPU Cloud vs. CPU Cloud: Which One Do You Need?

Les deux onglets ci-dessus répondent à deux questions différentes, et choisir le mauvais est le moyen le plus simple de payer trop cher pour du calcul. Un cloud GPU loue l'accès à une carte graphique, une H100, une A100, une RTX 4090, facturé à l'heure ou à la seconde. Un cloud CPU, généralement vendu comme VPS, loue une partie d'un serveur ordinaire facturé au mois. Aucun n'est meilleur que l'autre; chacun est conçu pour un type de travail différent.

Optez pour un cloud GPU lorsque la charge de travail elle-même a besoin d'un processeur graphique: l'entraînement ou le réglage fin d'un modèle, l'inférence locale sur un LLM à poids ouverts, la génération en masse d'images ou de vidéos, ou tout ce qui appelle CUDA ou un tensor core directement. Un VPS uniquement CPU ne peut pas accélérer ce travail; soit il s'exécute sur un GPU, soit il ne vaut pas la peine de payer pour l'exécuter à grande échelle. La propre liste d'Argusic d'applications AI auto-hébergées testées est un moyen rapide de voir quels outils open source s'attendent réellement à un GPU en dessous avant d'en louer un.

Optez pour un cloud CPU lorsque la charge de travail est limitée par le réseau, la mémoire ou la logique pure: un site web, une base de données, un runner CI, la plupart des déploiements Docker, et la plupart des agents AI, qui passent leur temps à appeler l'API d'un modèle plutôt qu'à exécuter les poids eux-mêmes. Un VPS avec suffisamment de RAM pour le processus et un disque assez rapide pour la base de données coûte généralement une fraction même de l'heure GPU la moins chère; parcourez tous les sujets testés par Argusic et la plupart fonctionnent de cette façon.

En règle générale: si le travail consiste à exécuter ce modèle, louez un GPU à l'heure uniquement pendant son exécution. Si le travail consiste à maintenir ce service en ligne, louez un VPS au mois. Argusic teste les deux côtés de ce choix de la même manière qu'il teste tout sur la page d'accueil: installez-le, exécutez-le, et enregistrez ce qui s'est réellement passé.

Questions fréquentes

Which provider has the cheapest verified H100 right now?

The summary line at the top of the GPU tab computes this live from the catalog: the lowest verified on-demand H100 rate among tracked providers, with its verification date.

What does the Measured badge mean?

Argusic runs its own benchmark fleet on some of these providers. Measured rows carry operational data from that usage, each number linked to run logs. Where a metric is not recorded yet, the row says so instead of showing a figure.

Do affiliate links change the ranking?

No. Order derives only from prices, stated criteria, or measurements. Providers without any affiliate program list on equal footing, and the disclosure sits at the top of the page.

Why is a row marked not yet verified?

It is listed for coverage: the provider exists and the plan is real, but Argusic has not yet confirmed the price against the provider's own pricing page or API. Such rows show no figure until they are verified.

What is the difference between API-verified and Owner-verified?

API-verified rows are refreshed daily straight from the provider's own pricing API, and the response is stored as the source. Owner-verified rows were checked by hand against the provider's own console or billing on the date shown. Both carry a verification date; neither is a guess.

What is the Featured partners tab?

It shows only the providers Argusic has an affiliate partnership with: their Visit button carries a tracked link, and the partnership is the only reason they are grouped there. Prices, verification states, and the default sort are identical to the other two tabs; being a partner changes no number and no rank anywhere else on the page.

How do I narrow the list to only the plans I care about?

Use the filters on the left: datacenter location, price range, GPU model or RAM and vCPU minimums, billing interval, provider type, and stock. Every option shows how many offers still match, options with zero matches grey out, and active filters render as removable chips above the list with a Clear all button.

Does the Argusic measured facet change the default sort order?

No. Sorting stays price, ascending, unless a measured sort is picked from the dropdown. The Measured facet and badge exist to separate what Argusic itself recorded from what a provider states on its own pricing page, so the two are never confused.

Does a low price mean the plan is available right now?

Only if its stock state says so. Each offer carries an in-stock, out-of-stock, or unknown state from the same source as its price, and the Availability filter can hide anything not confirmed in stock.

How often are the prices on this page refreshed?

Providers with a public pricing API are re-fetched daily, and every attempt, successful or not, is recorded in an operations log. Providers without an API are checked by hand, and the row carries the date of that check instead.

Les utilisateurs GitHub connectés peuvent discuter des fournisseurs et signaler des changements de prix. Les commentaires sont hébergés sur GitHub Discussions, y sont modérés, et n'affectent jamais les classements.

Les commentaires n'influencent jamais le classement ni aucun chiffre de cette page.

Échange

CERTAIN CONTENT THAT APPEARS ON THIS SITE COMES FROM AMAZON. THIS CONTENT IS PROVIDED 'AS IS' AND IS SUBJECT TO CHANGE OR REMOVAL AT ANY TIME.