--- title: Where to Rent an RTX 4090 at the Cheapest Price (October 2026) url: https://gpubank.io/rent-rtx-4090 description: RTX 4090 rental prices checked 16 Oct 2026: GPUBank $0.23 all-in ($0.14 spot), SaladCloud from $0.16, Vast.ai from $0.36, used cards near $2,200. generated: 2026-11-02T00:20Z publisher: GPUBank (GPU Compute OÜ, Tallinn, Estonia) summary_for_ai: https://gpubank.io/llms.txt full_text_for_ai: https://gpubank.io/llms-full.txt --- > Markdown version of https://gpubank.io/rent-rtx-4090, generated from the live page. GPUBank is a GPU cloud billed by the second at one all-in price per GPU (vCPU, RAM, NVMe, a public IPv4 and traffic included), in Paris, Frankfurt, Ashburn and Singapore. Prices are in US dollars, excluding VAT. Updated 16 Oct 2026 · 29 offers checked # Where to rent an RTX 4090 at the cheapest price We read every RTX 4090 offer we could find on the provider's own page or public API, priced what each hour actually includes, and worked out the monthly bill, the speed per dollar and the point where a used card would cost less than renting one. **GPUBank Research** 16 min read · Prices read on 16 Oct 2026 **24 GB**GDDR6X **1,008 GB/s**memory bandwidth **450 W**board power **$1,599**launch, Oct 2022 The short answer, 16 Oct 2026 1. $0.14 /hr **Cheapest hour you can launch and plan on.** GPUBank's interruptible RTX 4090 costs $0.14 with the same 8 vCPU, 41 GB of RAM, 500 GB NVMe disk, IPv4 address and traffic as on demand, five minutes of notice before a reclaim and the disk kept. Clore.ai's own snapshot of 21 September found one free spot server at $0.10, against a $0.365 spot median, and a higher bid can take it mid-run. 2. $0.16 /hr **Cheapest published tier elsewhere.** SaladCloud's lowest priority costs $0.16 an hour with vCPU and RAM included, on consumer PCs that can be reclaimed at any time and with no persistent disk. 3. $0.23 /hr **Cheapest whole machine on demand.** GPUBank's $0.23 covers 8 vCPU, 41 GB of RAM, a 500 GB NVMe disk, an IPv4 address and all traffic, billed per second, ahead of Runpod Community at $0.34 and Vast.ai's cheapest listing at $0.36 before storage; a top-up bonus brings the hour to about $0.14 to $0.19. [RTX 4090 $0.23 /hr all-in Launch an RTX 4090 Spot $0.14 · reserved $0.18 Billed per second](https://console.gpubank.io/launch?gpu=rtx-4090) The RTX 4090 is the card people rent when they need 24 GB of fast memory and refuse to pay datacenter prices, and it is also the card whose price is hardest to compare, because one model number covers several products that only share the silicon. One is a managed machine with its CPU, memory, disk and network folded into a single hourly figure. Another is a listing on a peer-to-peer marketplace where an independent host sets the GPU rate and bills storage and bandwidth on top. A third is a slot on somebody's gaming PC in a distributed network, cheap because it can disappear when the owner sits down to play, and a fourth is a five-card bare-metal block reserved for two weeks. Headline rates for these run from ten cents to almost ninety cents an hour, and most of that spread comes from what each price includes and what it allows the provider to do to your job. We read each provider's pricing page, documentation or public offer API on 16 October 2026 and recorded what every price covers, after a first survey on 15 October. Several figures from our first survey had moved or were wrong: Vast.ai's advertised floor of $0.14 is no longer in its live offers, where the cheapest single RTX 4090 now costs $0.36 on demand; Novita AI's own page now shows $0.33 rather than $0.67; packet.ai bills egress at $0.04 per GB; and the GeForce licence quoted in older articles has been rewritten since 2017. This page is published by GPUBank, a GPU cloud run by GPU Compute OÜ in Tallinn, Estonia, so read its entry with that in mind. It is measured on the same terms as everyone else, and the text says plainly where a Clore.ai bid, a SaladCloud priority tier or a used card bought for a full-time workload beats it, and what the buyer gives up in each case. The figures come from the providers' own pages unless a row is marked as an aggregator figure or as a provider's dated snapshot, and every price is per GPU-hour in US dollars. 01 / Where cheap hours come from ## Why a consumer card rents for ten cents The cheapest RTX 4090 hours come from consumer machines because the people who own them have already paid for the card. A host on Vast.ai or Clore.ai, or a gamer whose PC runs SaladCloud's agent, bought the 4090 for play or as a side business, pays household electricity and no rent, and is happy with anything above the cost of power, which is how a card that resells for about $2,200 ends up rented for ten to thirty-six cents an hour. Because the licence keeps those cards out of the big clouds, a large share of the supply sits in homes, small server rooms and colocated racks rather than in the facilities a renter would picture. The floors show it: Vast's cheapest on-demand single card was $0.36 and its median $0.53, Clore's spot floor was $0.10 against a $0.375 on-demand median, and Salad's lowest tier was $0.16. For the renter that has three consequences. The floor is real but thin: Clore's $0.10 was one free server among 243, Vast's listings change by the hour, and the cheapest slot is often the one with the fewest CPU cores, the slowest disk or the weakest uplink, so the price that matters is the cost per finished job after a short benchmark on the host you rent. Interruption is part of the deal, since the owner can take the machine back or a higher bid can take it from you, and on most of these networks the renter pays for the lost time. And the hardware is an end-of-life consumer card without ECC that has often run hot for years, which is fine for image generation and experiments and a poor fit for customer data, unreleased weights or anything a user waits on. A managed all-in rate costs a few cents more an hour and buys a known machine, a notice period and somebody who answers for the host. Marketplace floors and medians, USD per GPU-hour Lowest listing Median GPUBank all-in $0.23 Clore.ai **$0.10** 378 cards, 21 Sep SaladCloud **$0.16** Published rate Lium **$0.30** Nothing rentable Runpod Community **$0.34** Published rate TensorDock **$0.35** Published rate Vast.ai **$0.36** Published rate Floors and medians read on 16 Oct 2026 from each marketplace's own listing or price page; Clore.ai figures are its own dated snapshot. The blue line is GPUBank's on-demand price for the whole machine. AWS, Google Cloud, Microsoft Azure and Oracle Cloud list no RTX 4090 instance, so the usual exercise of dividing an eight-GPU instance price by eight has nothing to divide for this card. Their Ada-generation offers are the datacenter L4 and L40S, which carry ECC memory, run on datacenter drivers and fit server airflow, and which cost more per hour than any RTX 4090 offer on this page. For a team that must stay inside one of those clouds for compliance or data-gravity reasons, the L40S with 48 GB is the nearest equivalent, and the comparison with the 4090 is then a question of whether the extra memory and the platform are worth the price, which the section on other cards answers per job. For everyone else, the absence of the large clouds, which the licence explains, is the reason this market exists at all. NVIDIA GeForce driver licence Current version **2.8**You agree that GeForce or Titan SOFTWARE: (i) is licensed for use only on GeForce or Titan hardware products you own, and (ii) is not licensed for datacenter deployment. The clause as published by NVIDIA, read on 16 Oct 2026. 02 / The card and its price ## The card, and why its price varies so much The GeForce RTX 4090 is NVIDIA's Ada Lovelace flagship on the AD102 chip, with 16,384 CUDA cores, 512 fourth-generation Tensor Cores, 24 GB of GDDR6X on a 384-bit bus and a 450 W total graphics power that NVIDIA pairs with an 850 W minimum supply. Its 1,008 GB/s of memory bandwidth is what sets the speed at which it writes tokens in a language model, Clore.ai gives its dense FP16 and BF16 tensor throughput as 165 TFLOPS, noting that the 330 TFLOPS sometimes quoted assumes 2:4 sparsity, and its Tensor Cores run FP8 natively, the precision at which FLUX.1 dev fits in its memory and runs well. It went on sale on 12 October 2022 at $1,599, was replaced at the top of the range by the RTX 5090, on sale from 30 January 2025 at $1,999, and is no longer in production, so the supply of new cards has dried up and the used market sets the price of the hardware that hosts rent out. It has no NVLink, so two cards in one machine are two separate 24 GB devices joined over PCIe. The large clouds do not rent it, and the reason is a licence clause rather than a missing price list. Clause 2.8 of the current NVIDIA driver licence says that GeForce software is licensed only for GeForce hardware you own and is not licensed for datacenter deployment, a restriction NVIDIA first introduced at the end of 2017 to keep consumer cards out of large enterprise installations. The supply therefore comes from specialised GPU clouds, peer-to-peer marketplaces and networks of consumer PCs, where the 4090 is one of the most listed cards of all: Clore.ai alone counted 378 of them on 243 servers in its snapshot of 21 September. People rent it for ComfyUI, Stable Diffusion WebUI Forge and InvokeAI, where SDXL fits easily and FLUX.1 dev runs at FP8, for LoRA and QLoRA fine-tuning of models up to about 8B to 14B parameters, and for single-card inference of language models up to about 14B at 8-bit or about 32B at 4-bit with a short context; Clore's rule of thumb of half a gigabyte per billion parameters at 4-bit puts a 32B model near 16 GB of weights and a 70B model near 40 GB, beyond one card. GPUBank positions its RTX 4090 for ComfyUI, SDXL and LoRA training on 8B models. The price range follows from where the cards sit: a card in a spare room costs its owner little more than electricity, a card in a rack carries the cost of the building, the network and the staff, and the rate reflects which of the two you are renting. What one RTX 4090 cost to buy, October 2022 to October 2026 New Used, sold or valued Used, asking Oct 2022 **$1,599** Launch price, on sale 12 October 2022 Sep 2023 **$1,603** eBay average sold price, September 2023 Oct 2023 **$1,704** eBay average over 304 sales before the China export rules Dec 2024 **$2,500** Remaining Amazon stock started near $2,500 as production ended Jan 2026 **$2,352** Used market value, 17 January 2026 May 2026 **$2,328** Used market value, 30 May 2026 Sep 2026 **$3,010** Used market value spike, 12 September 2026 Oct 2026 **$2,214** Used market value, 6 October 2026 Oct 2026 **$2,999** Median asking price, 54 used listings, 15 October 2026 03 / The price ladder ## How to read the RTX 4090 market The fair way to compare RTX 4090 offers is to group them by what you are actually buying before looking at the number. A managed on-demand offer sells a defined machine at a fixed hourly rate that stays in force until you stop it, and the useful question is what that rate leaves out: GPUBank, CloudRift and Beam include the local disk and charge nothing for traffic, while Runpod bills volumes separately and packet.ai bills egress at $0.04 per GB. A marketplace listing sells a slot on a host whose CPU share, memory, disk speed, PCIe generation and reliability vary from one listing to the next, at a GPU rate the host sets, with storage and bandwidth as separate lines. Putting the two side by side without saying so is how most cheapest-GPU lists end up comparing things that cannot be compared. The other three kinds of offer each trade something for the price. Spot and priority tiers can be taken back, either by a higher bid on Vast.ai and Clore.ai or by a higher-priority workload on SaladCloud, and they differ less in their discount than in what happens at the moment of interruption: some delete the container, some leave a disk you keep paying for, and only a few give notice. Reserved capacity lowers the rate in exchange for paying for a week, a month or longer whether you use it or not, and for this card it ranges from GPUBank's $0.18 for a single card to Clore.ai's five-GPU and eight-GPU bare-metal blocks at $0.40 to $0.94. Serverless workers bill only the seconds spent on requests, at an hourly equivalent of $0.69 on Beam for the GPU alone and $1.10 for a Runpod flex worker, which only pays when traffic is rare and bursty: once a $0.69 worker is busy for a third of the month, a dedicated $0.23 machine already costs the same. *GetDeploying median, 16 Oct $0.47* **Managed on demand** A fixed machine at one hourly price. *GPUBank**$0.23*** *Novita AI**$0.33*** *CloudRift**$0.39*** *packet.ai**$0.39*** *Beam**$0.44*** *Spheron**$0.58*** *Oblivus**$0.64*** *Runpod Secure Cloud**$0.89*** **Marketplace** Hosts set their own price; machines vary. *Clore.ai**$0.28*** *Lium**$0.30*** *Runpod Community**$0.34*** *TensorDock**$0.35*** **Spot and interruptible** Cheaper hours that can be taken back. *Clore.ai**$0.10*** *GPUBank**$0.14*** *SaladCloud**$0.16*** *Novita AI**$0.17*** *SaladCloud**$0.217*** *Vast.ai**$0.27*** *SaladCloud**$0.33*** **Reserved** Committed for a term, paid either way. *GPUBank**$0.18*** *Novita AI**$0.27*** *CloudRift**$0.33*** *CloudRift**$0.35*** *Clore.ai**$0.40*** *Oblivus**$0.44*** **Serverless** Billed by active seconds. *Beam**$0.69*** *Runpod**$1.10*** USD per GPU-hour, read on 16 Oct 2026. Hover a dot for the offer; crowded dots keep their label in the table below. The large clouds are listed separately. 04 / Provider by provider ## What each price buys, and what it leaves out. The captures below were taken on 16 Oct 2026 from each provider's own RTX 4090 or pricing page, so every figure can be read next to the conditions printed beside it. gpubank.io/pricing **GPUBank** All-in $0.23/hr Managed on demand · Frankfurt, Ashburn, Paris queue 8 vCPU 41 GB RAM 500 GB NVMe IPv4 All traffic Disk $0.03/GB-mo Interruptible $0.14 Reserved $0.18 GPUBank lists the RTX 4090 at $0.23 an hour on demand, $0.000064 a second, and the rate covers the vCPUs, the memory, the local NVMe scratch disk, one public IPv4 address and all traffic, in machines of one to four GPUs. The only other line on the bill is a persistent disk at $0.03 per GB-month. Billing is per second with no minimum and starts only once the machine has passed its health check, a failed launch is not billed, and a host failure during a job restarts the machine on a healthy host with the same disk and credits the session in full. The interruptible tier costs $0.14 with five minutes of notice and a stop rather than a deletion, and a week or a month reserved costs $0.18. Payment is a prepaid balance topped up in crypto from $25, with a bonus that brings the effective hour to between $0.14 and $0.19; cards are not accepted for now. The card is listed in Frankfurt and Ashburn, with Paris on a waiting queue. Checked 16 Oct 2026 vast.ai/pricing/gpu/RTX-4090 **Vast.ai** from $0.36/hr Marketplace · host-set price per listing Host's CPU and RAM share Storage at host rate Bandwidth at host rate Bids from $0.27 Vast.ai is the reference RTX 4090 marketplace, and its public offer search, which we queried on 16 October, is where the floor and median below come from. The cheapest single-card on-demand listing was $0.36 for the GPU, on a US host with a 99.6 percent reliability score, the median per-GPU rate across all listings was $0.53, and interruptible bids started at $0.27. The $0.14 floor still quoted on some Vast pages and in older articles was not in the live offers. Each host also sets a storage rate, about $0.20 per GB-month on the cheap listings, billed for as long as the instance exists even while it is stopped, and a bandwidth rate per gigabyte. Listings show the CPU share, the PCIe link and the measured memory bandwidth, and the cheapest hosts are not always the fastest, so a short benchmark before a long job is worth its few cents. Checked 16 Oct 2026 salad.com/pricing **SaladCloud** $0.16 to $0.33/hr Distributed consumer PCs · four priority tiers vCPU RAM No persistent disk Reclaimable at every tier SaladCloud runs containers on consumer PCs and sells the RTX 4090 at four priorities: $0.330 an hour at high, $0.273 at medium, $0.217 at low and $0.160 at the lowest batch tier, billed per second while the container runs, with allocation, image download and start-up not charged and vCPU and RAM included. Those rates have applied since a price change on 12 September 2026. The price is low because the machines belong to people who use them, so a container can be moved off a PC at any time and a higher-priority workload displaces a lower one. There is no persistent disk, so model weights travel in the image or come from object storage you pay for elsewhere, and a production endpoint needs at least two replicas. For a large batch of images that tolerates restarts, the lowest tier is one of the cheapest published 4090 hours anywhere. Checked 16 Oct 2026 runpod.io/pricing **Runpod** $0.34/hr Community Cloud · Secure Cloud at $0.89 6 vCPU 41 GB RAM Volume $0.10/GB-mo, $0.20 idle Network volume $0.07/GB-mo Runpod's pricing page, updated 8 October 2026, lists the RTX 4090 with 6 vCPU and 41 GB of RAM at $0.89 an hour on Secure Cloud, the tab it shows by default, and its structured data gives $0.34 for Community Cloud, which runs on third-party hosts. GetDeploying showed the Secure tier in stock in 1, 2 and 4-card sizes and the Community tier out of stock on 16 October. Billing is per second, and storage is a separate line: a volume disk costs $0.10 per GB-month while the pod runs and $0.20 while it is stopped, and a network volume costs $0.07 per GB-month under 1 TB, which is how a 100 GB model library adds $7 a month. The serverless RTX 4090 worker is $1.10 an hour. Runpod's own guide, updated on 15 October, describes the card as end of life and still selling above its launch price. Checked 16 Oct 2026 cloudrift.ai/pricing **CloudRift** $0.39/hr Managed on demand · reservations from $0.33 800 GB local storage Networking No egress fees 1 month $0.35 3 months $0.33 CloudRift lists the RTX 4090 at $0.39 an hour on demand, billed by the second while the instance runs, with up to ten GPUs per instance. Its FAQ says the rate covers compute, the attached local storage shown in the pricing table, 800 GB per card for this model, networking and standard support, and that there are no egress, ingress or API-call charges. A one-month reservation costs $0.35 and a three-month reservation $0.33, and the page warns that prices fluctuate and availability varies. The pricing table does not state the vCPU and RAM that come with each card, so check the configuration in the console before you launch. CloudRift also publishes its own vLLM benchmarks of the 4090 against the RTX 5090 and the RTX PRO 6000, which we use in the performance section. Checked 16 Oct 2026 clore.ai/rent-4090.html **Clore.ai** spot from $0.10/hr Marketplace · snapshot of 21 Sep 2026 Host-set CPU, RAM, disk Renter fee 1.25% spot, 5% on demand Per-minute billing Clore.ai publishes the most useful dated snapshot of any marketplace. On 21 September 2026 it counted 243 servers carrying 378 RTX 4090s, 130 of them unrented; the median on-demand quote was $0.375 per GPU-hour across 230 quotes and the median spot quote $0.365, the cheapest free server on spot was $0.100, and the median of the unrented servers was $0.583, because idle cards tend to be the overpriced ones. The page title advertised from $0.28 on the day we read it. Clore bills by the minute and takes a 2.5 percent fee on spot and 10 percent on on-demand rentals, split equally between renter and host. Spot rentals are won by bid and a higher bid can take the machine mid-run. Bare-metal blocks of five cards for 14 days in the UK and eight cards for 30 days in Japan and Hong Kong were quoted at $0.40 to $0.94 per GPU-hour. Checked 16 Oct 2026 ### Where GPUBank wins, and where it does not GPUBank is the cheapest managed on-demand RTX 4090 we found: $0.23 an hour, or $0.000064 a second, for 8 vCPU, 41 GB of RAM, a 500 GB local NVMe disk, one public IPv4 address and all inbound and outbound traffic, in machines of one to four GPUs. Billing starts only after the machine passes its health check, a failed launch is not billed, a host failure during a job restarts the machine on a healthy host with the same disk and credits the session in full, and a stopped machine costs only its persistent disk at $0.03 per GB-month. The interruptible tier at $0.14 gives five minutes of notice, runs a checkpoint hook and stops the machine instead of deleting it, and a reservation for a week or a month costs $0.18. GPUBank lists the card in Frankfurt and Ashburn, with Paris on a waiting queue, and the median time from launch to a machine that has passed its health check is 48 seconds. It loses in three places. A lucky Clore.ai spot bid near $0.10 is cheaper than any GPUBank tier, and SaladCloud's lowest priority at $0.16 is close to its interruptible price for batch work that tolerates reclaims. A used card bought for a workload that runs around the clock beats GPUBank's list on-demand rate past about 690 hours a month, as the section on buying shows. And payment is a prepaid balance topped up in crypto from $25 to $5,000, with cards not accepted for now; the balance never expires and carries a bonus of 20 to 60 percent, which brings the effective on-demand hour to about $0.19 with a $25 top-up, $0.17 with $100 and $0.14 with $1,000, but the bonus only pays if the credit is actually used. ### The rest of the market Several managed offers sit just above the providers shown in the cards. Novita AI's own RTX 4090 page lists $0.33 an hour on demand and warns that the price follows the computing power market, while GetDeploying showed its single-card configuration with 16 vCPU and 125 GB of RAM out of stock on 16 October, along with a $0.17 spot rate and a 12-month four-card reservation at $0.27 per GPU, all sold out. packet.ai rents a dedicated card at $0.39 with hourly billing, local NVMe and $0.04 per GB of egress, and its own page names both US Central and EU West as the region. Beam's on-demand machine with 8 vCPU, 64 GB of RAM and 500 GB of NVMe starts at $0.44 with no egress fees. Spheron lists 8 vCPU, 24 GB of RAM and 500 GB of storage at a last known $0.58, billed per minute after a 20-minute minimum, but its page said no RTX 4090 capacity was live when we read it. The marketplace tail is thinner than its reputation. Lium's own page shows a reference price of $0.30 with no pod available, while GetDeploying listed one Lium node in stock at $0.62, and TensorDock's page quotes a typical $0.35 that varies by host, with vCPU at $0.003 an hour, RAM at $0.002 per GB-hour and NVMe at $0.00005 per GB-hour billed on top, on a pricing page last updated in July 2024. Further up, GetDeploying tracked Oblivus at $0.64 on demand and out of stock, a 36-month Oblivus reservation at $0.44, GPUhub at $0.44, SwissGPU at $0.48, Database Mart reservations at $0.51 to $0.76 and Runpod's Secure Cloud at $0.89 in stock, for a median on-demand rate of $0.47 across twelve providers on 16 October. io.net's explorer, which our first survey quoted at $0.25, would not load for us, so it is left out of the comparison. 05 / Every offer ## The whole comparison, sorted by kind and price. | Provider | USD / GPU-hr | Kind | Unit | What the price includes | Billing | Stock | Basis | | --- | --- | --- | --- | --- | --- | --- | --- | | GPUBank | $0.23 | Managed on demand | 1 GPU | 8 vCPU, 41 GB RAM, 500 GB NVMe, IPv4, traffic | Per second | In stock | Provider page | | Novita AI | $0.33 | Managed on demand | 1 GPU | Price follows the market; 16 vCPU per aggregator | Not stated | Sold out | Provider page | | CloudRift | $0.39 | Managed on demand | 1 GPU | 800 GB local storage, networking, no egress | Per second | Not stated | Provider page | | ppacket.ai | $0.39 | Managed on demand | 1 GPU | Dedicated card, local NVMe; egress $0.04/GB | Hourly | In stock | Provider page | | Beam | $0.44 | Managed on demand | 1 GPU | 8 vCPU, 64 GB RAM, 500 GB NVMe, no egress | Not stated | Not stated | Provider page | | Spheron | $0.58 | Managed on demand | 1 GPU | 8 vCPU, 24 GB RAM, 500 GB NVMe; 20 min minimum | Per minute | Sold out | Provider page | | Oblivus | $0.64 | Managed on demand | 1 GPU | 12 vCPU, 70 GB RAM | Not stated | Sold out | Aggregator | | Runpod Secure Cloud | $0.89 | Managed on demand | 1 GPU | 6 vCPU, 41 GB RAM; disks extra | Per second | In stock | Provider page | | Clore.ai | $0.28 | Marketplace | 1 GPU | Headline 'from' price; host-set machine; 5% renter fee | Per minute | In stock | Provider page | | Lium | $0.30 | Marketplace | 1 GPU | Reference price, varies by node | Per second | Sold out | Provider page | | Runpod Community | $0.34 | Marketplace | 1 GPU | 6 vCPU, 41 GB RAM; volumes extra | Per second | Sold out | Provider page | | TensorDock | $0.35 | Marketplace | 1 GPU | GPU only; vCPU, RAM, NVMe billed on top | Per second | Not stated | Provider page | | Vast.ai | $0.36 | Marketplace | 1 GPU | Host CPU and RAM share; storage, bandwidth extra | Per second | In stock | Provider page | | Clore.ai | $0.10 | Spot and interruptible | 1 GPU | Cheapest free spot server; median $0.365 | Per minute | Not stated | Provider snapshot | | GPUBank | $0.14 | Spot and interruptible | 1 GPU | Same machine, 5 min notice, disk kept | Per second | In stock | Provider page | | SaladCloud | $0.16 | Spot and interruptible | 1 GPU | Lowest priority; vCPU, RAM; no persistent disk | Per second | Not stated | Provider page | | Novita AI | $0.17 | Spot and interruptible | 1 GPU | Spot, 16 vCPU | Not stated | Sold out | Aggregator | | SaladCloud | $0.217 | Spot and interruptible | 1 GPU | Low priority | Per second | Not stated | Provider page | | Vast.ai | $0.27 | Spot and interruptible | 1 GPU | Lowest interruptible bid; storage, bandwidth extra | Per second | In stock | Provider page | | SaladCloud | $0.273 | Spot and interruptible | 1 GPU | Medium priority | Per second | In stock | Provider page | | SaladCloud | $0.33 | Spot and interruptible | 1 GPU | High priority, still on consumer PCs | Per second | In stock | Provider page | | GPUBank | $0.18 | Reserved | 1 GPU | Week or month, prepaid, same machine | Per term | In stock | Provider page | | Novita AI | $0.27 | Reserved | cluster | Four cards for 12 months | Per term | Not stated | Aggregator | | CloudRift | $0.33 | Reserved | 1 GPU | Three months, local storage included | Per term | Not stated | Provider page | | CloudRift | $0.35 | Reserved | 1 GPU | One month, local storage included | Per term | Not stated | Provider page | | Clore.ai | $0.40 | Reserved | cluster | Bare metal, 5 GPUs for 14 days (UK) and up; to $0.94 | Per term | On request | Provider snapshot | | Oblivus | $0.44 | Reserved | 1 GPU | 36 months | Per term | Not stated | Aggregator | | Beam | $0.69 | Serverless | 1 GPU | GPU only; CPU and RAM billed on top | Per millisecond | Not stated | Provider page | | Runpod | $1.10 | Serverless | 1 GPU | Serverless worker; storage separate | Per second | Not stated | Provider page | "Aggregator" rows could not be read on the provider's own page that day; "Provider snapshot" rows come from a page dated by the provider. 06 / The real monthly bill ## What an RTX 4090 hour really costs Storage is the line that most often changes the ranking for this card, because image and video work keeps large model libraries between sessions. On GPUBank the local NVMe is part of the rate and a persistent disk costs $0.03 per GB-month, so 100 GB of checkpoints and LoRAs costs $3 a month. Runpod charges $0.07 per GB-month for a network volume under 1 TB and $0.20 for an idle volume disk. Vast.ai bills each host's storage rate for as long as the instance exists, stopped or not, and the cheap listings we queried charged about $0.20 per GB-month, so the same 100 GB costs about $20. SaladCloud has no persistent disk at all, so weights are pulled again whenever the container lands on a new PC. Billing granularity and lost time are the other two hidden lines. A 61-minute ComfyUI session at $0.23 costs $0.23 when billed per second, as on GPUBank, Runpod, CloudRift, SaladCloud and Vast.ai, and $0.46 when the meter rounds up to the started hour, while Spheron's 20-minute minimum makes a five-minute test cost twenty minutes. Lost time matters more on the cheapest offers, because on a marketplace the renter usually pays for it: an outbid spot instance is gone and billed to the minute, a host that runs the card on PCIe 3.0 or gives it two CPU cores makes the job longer at your expense, and the cheapest Vast listing reported a measured memory bandwidth of 876 GB/s for a card rated at 1,008. Clore's own guide says renters are not charged for the downtime when an instance goes down unexpectedly, but a slow host is still paid in full. GPUBank does not bill a failed launch and credits a session interrupted by a host failure, which is why the scenarios below add an explicit allowance for lost time to every marketplace and spot row. ### Three workloads, priced for a month **GPUBank** 40.67 h GPU $9.35 Storage $3.00 Other None Month $12.35 On demand, per second, traffic and IPv4 included All-in **SaladCloud** 40.67 h GPU $13.42 Storage $0.00 Other Model storage kept elsewhere Month $13.42 High priority, no persistent disk **CloudRift** 40.67 h GPU $15.86 Storage $0.00 Other Persistent storage not priced Month $15.86 On demand, per second, local disk only while running **Runpod** 40.67 h GPU $13.83 Storage $7.00 Other None Month $20.83 Community Cloud, 100 GB network volume at $0.07 **Vast.ai** 40.67 h GPU $14.64 Storage $20.00 Other Bandwidth at host rate Month $34.64 Cheapest on-demand listing; 100 GB at $0.20 kept all month Forty sessions of 61 minutes are 40.67 billed hours on any per-second meter. On GPUBank that is $9.35 of GPU time plus $3.00 for a 100 GB persistent disk, $12.35 in all, or about $9.15 of real money from a $100 top-up credited as $135. SaladCloud's high tier comes to $13.42 but keeps no disk, so the model library must live elsewhere and download at each start. Runpod Community is $13.83 plus $7.00 of network volume, $20.83. CloudRift is $15.86 with nothing kept between sessions. Vast's cheapest listing costs only $14.64 of GPU time, but keeping 100 GB on the instance all month at $0.20 per GB-month adds $20, for $34.64. GPUBank wins here, and the answer changes only for a user who keeps nothing between sessions and accepts Salad's re-downloads. 07 / Speed per dollar ## What it does well, and where it stops Measured results put the RTX 4090 where its memory says it should be. In ComfyUI's community benchmark of the standard FLUX.1 dev FP8 workflow, a 4090 on PyTorch 2.9.1 and ComfyUI 0.6.0 took 11.28 seconds an image, against 8.78 seconds for an RTX 5090 and 26 seconds for an RTX 3090, so 1,000 images take 3.13 GPU-hours; NVIDIA's own figure is about 10 seconds at FP8 and 15 at FP16. On SDXL at 1024 by 1024 and 20 steps, Formula Mod reports about 3.2 seconds on the 4090 and 2.2 on the 5090, and Spheron about 2.1 seconds on the 4090 against 2.5 on the L40S, so 1,000 SDXL images cost about $0.20 on GPUBank's 4090 against $0.29 on its 5090 and about $0.38 on its L40S. In llama.cpp's CUDA scoreboard, which runs Llama 2 7B at Q4_0 with flash attention, the 4090 generates 188.96 tokens per second against 300.40 for the 5090 and 200.90 for an 80 GB A100. Batched serving changes the scale: CloudRift measured 2,259 tokens per second on one 4090 serving Qwen3-Coder-30B-A3B in AWQ through vLLM under heavy concurrency. The card stops being the right tool in three cases. The first is any model that does not fit in 24 GB: FLUX.1 dev in BF16, which Clore puts at about 24 GB with no headroom and Spheron at 30 to 33 GB with its text encoders loaded, a 70B language model, which needs about 40 GB at 4-bit, or a 32B model served with a long context. CloudRift's own test needed two 4090s to serve Llama 3.3 70B at 4-bit and reached 467 tokens per second, against 1,031 on a single 96 GB RTX PRO 6000. The second is work that splits one model across cards, because without NVLink every layer pays a PCIe toll under tensor parallelism. The third is unattended production serving where ECC memory, a datacenter driver and a service agreement matter more than the hourly rate, which is the ground of the L40S and the datacenter cards. **FLUX.1 dev, seconds per image** ComfyUI standard FLUX dev FP8 template, community runs (4090: PyTorch 2.9.1, ComfyUI 0.6.0, Dec 2025) RTX 3090 26 s RTX 4090 11.3 s RTX 5090 8.78 s ComfyUI GitHub discussion #9002; each result posted by a user, second or later render **llama.cpp, tokens per second generated** Llama 2 7B Q4_0, llama-bench tg128, flash attention on, CUDA backend RTX 4090 189 tok/s A100 80 GB 201 tok/s RTX 5090 300 tok/s llama.cpp CUDA scoreboard, ggml-org discussion #15013 **vLLM, batched throughput** Qwen3-Coder-30B-A3B-Instruct AWQ, one GPU, 1,200 requests of 1,000 in and 1,000 out tokens, high concurrency RTX 4090 2,259 tok/s RTX 5090 4,570 tok/s RTX PRO 6000 8,425 tok/s CloudRift, RTX 4090 vs 5090 vs PRO 6000 LLM inference benchmark, 9 Oct 2025 CHEAPEST ALL-IN $0.23 GPUBank on demand, whole machine and traffic MARKETPLACE MEDIAN $0.53 Vast.ai per-GPU median across listings, 16 Oct USED CARD $2,214 GetPCParts used value, 6 Oct 2026 08 / Against other cards ## The RTX 4090 against the other cards, per job Per job rather than per hour, the RTX 4090 is the cheapest card GPUBank rents for anything that fits in 24 GB. On FLUX.1 dev at FP8, 1,000 images cost $0.72 on the 4090 at 11.28 seconds each and $1.17 on the RTX 5090 at 8.78 seconds and $0.48 an hour, so the 5090 buys speed rather than savings. For a single user in llama.cpp, a million tokens cost $0.34 on the 4090, $0.44 on the 5090 and $1.31 on an 80 GB A100 at GPUBank's $0.95 A100 rate, because the A100 generates only 6 percent faster here. Under batched vLLM serving the 4090 and the 5090 converge at about $0.03 per million tokens each, since the 5090's 4,570 tokens per second are almost exactly twice the 4090's 2,259 at a little more than twice the price. The other cards earn their price through memory rather than speed. The RTX 5090 at $0.48 adds 8 GB and FP4 support, which is enough for FLUX at higher precision, more LoRAs loaded at once or a 32B model with a usable context. The L40S at $0.55 brings 48 GB of ECC memory, which suits BF16 FLUX, stacked ControlNets and high-resolution upscaling, although we found no measurement on the same tests to price it per job. The A100 SXM at $0.95 has 80 GB of HBM2e at 2.0 TB/s and is the card for models and batches that need it, not a faster 4090: to match the 4090 per job it would have to finish 4.1 times sooner, and on the workloads measured here it is barely faster at all. RTX 4090 **24 GB** GDDR6X · 1.0 TB/s $0.23 /hr ComfyUI, SDXL and LoRA training on 8B models RTX 5090 **32 GB** GDDR7 · 1.8 TB/s $0.48 /hr FLUX, SDXL, AI video and small language models L40S **48 GB** GDDR6 ECC · 864 GB/s $0.55 /hr Stable Diffusion, video, Whisper transcription and embeddings A100 SXM **80 GB** HBM2e · 2.0 TB/s $0.95 /hr LoRA and QLoRA fine-tuning, batch inference GPUBank on-demand prices, all-in. Memory bar scaled to 80 GB. 09 / Rent or buy ## Rent or buy a used RTX 4090 The realistic purchase is one used card in a workstation, because new stock has gone and an eight-card 4090 server is neither sold nor licensed for a datacenter. Trackers disagree on what that card costs. GetPCParts put the used value at $2,214 on 6 October 2026 after a $3,010 spike in September, PCPrice.watch reported an eBay sold median of €2,185 over 111 sales in 90 days, and DealHunter's median asking price across 54 used listings was $2,999 on 15 October. We use $2,200 for the card, $1,200 for the machine around it, $1,400 of combined resale value after 24 months, a draw of 0.6 kW under load and US residential electricity at 18.19 cents per kWh. Owning then costs $83.33 a month in capital plus $0.11 in power for every hour the card works. Against GPUBank's $0.23 list rate, each rented hour costs $0.12 more than the electricity of an owned card, so the $83.33 of monthly capital is recovered after about 690 hours, which is a card that works 23 hours a day. At a full 730 hours owning costs $163.01 against $167.90 on demand, a narrow win for the buyer, but a month reserved at $0.18 costs $131.40 and the on-demand rate paid from a $100 top-up works out at $124.37, and the owned card never beats either, because the margin they leave above its electricity is too small to recover the capital in a 730-hour month. Buy if the card will run almost continuously, if your data must stay on your premises, if you already own one or if it also serves as your desktop card. Rent for anything less, for bursty work, and to try the RTX 5090 or the L40S when a job needs more memory. Monthly cost of One used RTX 4090 in a workstation (1 GPU) against hours of use a month, 0 to 730. Owning: price + $1,200 host − $1,400 resale over 24 months, plus 0.6 kW at the wall under load. Rent: per GPU-hour × 1. | Hours a month | Rent at $0.23 | Rent at $0.14 effective | Own | Cheaper | | --- | --- | --- | --- | --- | | 50 | $11.50 | $7.19 | $88.79 | Rent | | 100 | $23.00 | $14.38 | $94.25 | Rent | | 200 | $46.00 | $28.75 | $105.16 | Rent | | 400 | $92.00 | $57.50 | $126.99 | Rent | | 730 | $167.90 | $104.94 | $163.01 | Rent | Used value $2,214 on 6 Oct 2026 (GetPCParts); eBay sold median €2,185 (PCPrice.watch, 15 Oct); asking median $2,999 (DealHunter, 15 Oct) 10 / What to choose ## The cheapest RTX 4090 depends on the job. Creators running ComfyUI, SDXL or FLUX GPUBank RTX 4090 on demand Per-second billing, a cheap persistent disk for the model library and included traffic make short sessions cheapest at $0.23 an hour. Large batch jobs that checkpoint GPUBank interruptible, or a Clore.ai or SaladCloud floor At $0.14 with notice and the disk kept, or $0.10 to $0.16 if you accept reclaims without notice and re-downloads. LoRA fine-tuning up to 8B RTX 4090, interruptible for long runs QLoRA on 8B fits in 24 GB, and a trainer that checkpoints loses minutes, not hours, when the card is reclaimed. Models above 24 GB or long contexts RTX 5090, L40S or RTX PRO 6000 A 70B model needs about 40 GB at 4-bit; paying for memory beats splitting a model across cards with no NVLink. 11 / Questions ## What people ask before renting one. The cheapest RTX 4090 hours are on capacity that can be taken back, and which one is cheapest depends on how much risk you accept. Clore.ai's snapshot of 21 September 2026 found a free spot server at $0.10 per GPU-hour, but that was one server among 243 and a higher bid can take it. GPUBank's interruptible tier is $0.14 with five minutes of notice and the disk kept, and SaladCloud's lowest priority is $0.16 on consumer PCs with no persistent disk. For capacity that cannot be interrupted, GPUBank's $0.23 is the lowest all-in price we found, ahead of Novita AI at $0.33, Runpod Community at $0.34 and Vast.ai's cheapest listing at $0.36 before storage and bandwidth. Sources ## Sources 1. 01 [GPUBank pricing](https://gpubank.io/pricing) RTX 4090 rates, inclusions, billing, top-up bonuses, other cards · read 16 Oct 2026 2. 02 [Clore.ai, Rent RTX 4090](https://clore.ai/rent-4090.html) 21 Sep 2026 snapshot, fees, billing, bare-metal quotes · read 16 Oct 2026 3. 03 [SaladCloud pricing](https://salad.com/pricing) RTX 4090 at four priorities, inclusions, billing · read 16 Oct 2026 4. 04 [Runpod pricing (updated 8 October 2026)](https://www.runpod.io/pricing) Secure and Community rates, storage, serverless · read 16 Oct 2026 5. 05 [Vast.ai public offer search, RTX 4090](https://console.vast.ai/api/v0/bundles/) Cheapest listing, median rate, bids, host storage rates, measured bandwidth · read 16 Oct 2026 6. 06 [Novita AI, RTX 4090 GPU instance](https://novita.ai/gpu-instance/gpu/rtx-4090) On-demand rate · read 16 Oct 2026 7. 07 [packet.ai, Rent RTX 4090](https://packet.ai/articles/rent/rtx-4090) Dedicated rate, billing, egress · read 16 Oct 2026 8. 08 [NVIDIA Driver License Agreement (v. 25 February 2025)](https://www.nvidia.com/en-us/drivers/geforce-license/) Clause 2.8, no datacenter deployment · read 16 Oct 2026 9. 09 [NVIDIA GeForce RTX 4090 product page](https://www.nvidia.com/en-us/geforce/graphics-cards/40-series/rtx-4090/) CUDA cores, memory, bus, power · read 16 Oct 2026 10. 10 [Engadget, NVIDIA's $1,599 GeForce RTX 4090 arrives on October 12th](https://www.engadget.com/nvidia-rtx-4090-announced-152529456.html) Launch date and price · read 16 Oct 2026 11. 11 [The Register, Nvidia: Using cheap GeForce, Titan GPUs in servers? Haha, nope!](https://www.theregister.com/2018/01/03/nvidia_server_gpus/) Origin of the clause and NVIDIA's statement · read 15 Oct 2026 12. 12 [CloudRift pricing](https://www.cloudrift.ai/pricing) On-demand and reserved rates, inclusions, no egress · read 16 Oct 2026 13. 13 [Beam pricing](https://www.beam.cloud/pricing) On-demand machine, serverless rate, no egress · read 16 Oct 2026 14. 14 [GetDeploying, RTX 4090 rental prices](https://getdeploying.com/gpus/nvidia-rtx-4090) Aggregator rates and stock, $0.47 median across twelve providers on 16 Oct · read 16 Oct 2026 15. 15 [Spheron, RTX 4090 rental](https://www.spheron.network/gpu-rental/rtx-4090/) Last known rate, minimum runtime, no capacity · read 16 Oct 2026 16. 16 [Lium, RTX 4090](https://lium.io/gpu/rtx-4090) Reference price, no pod available · read 16 Oct 2026 17. 17 [TensorDock cloud GPUs](https://www.tensordock.com/cloud-gpus.html) Typical rate and per-resource prices · read 16 Oct 2026 18. 18 [GPUBank, GPU board and regions](https://gpubank.io/) RTX 4090 regions, median launch time · read 16 Oct 2026 19. 19 [Clore.ai blog, How to rent a GPU: complete beginner's guide](https://blog.clore.ai/how-to-rent-a-gpu-complete-beginners-guide/) No charge for unexpected instance downtime · read 15 Oct 2026 20. 20 [ComfyUI discussion #9002, FLUX dev FP8 benchmark](https://github.com/Comfy-Org/ComfyUI/discussions/9002) Seconds per image on RTX 3090, 4090, 5090 · read 16 Oct 2026 21. 21 [NVIDIA blog, GeForce RTX 50 Series launch: FLUX.1 dev timings](https://blogs.nvidia.com/?p=77533) FLUX.1 dev on the RTX 4090, about 15 s at FP16 and 10 s at FP8 · read 15 Oct 2026 22. 22 [Formula Mod, Best GPU for Stable Diffusion and Flux in 2026](https://www.formulamod.net/blogs/new/best-gpu-for-stable-diffusion-and-flux-in-2026-rtx-4060-ti-vs-3090-vs-4090-vs-5090-benchmarks) SDXL 1024 px, 20 steps: about 3.2 s on RTX 4090, 2.2 s on RTX 5090 · read 15 Oct 2026 23. 23 [Spheron, Best GPU for AI image generation 2026](https://www.spheron.network/blog/best-gpu-ai-image-generation-2026/) SDXL 1024 px, 20 steps: about 2.1 s on RTX 4090, 2.5 s on L40S · read 15 Oct 2026 24. 24 [llama.cpp discussion #15013, performance on NVIDIA CUDA](https://github.com/ggml-org/llama.cpp/discussions/15013) Tokens per second, Llama 2 7B Q4_0 · read 16 Oct 2026 25. 25 [CloudRift, RTX 4090 vs 5090 vs PRO 6000: LLM inference benchmark](https://www.cloudrift.ai/blog/benchmarking-rtx-gpus-for-llm-inference) vLLM throughput, settings · read 16 Oct 2026 26. 26 [GetPCParts, used RTX 4090 market price](https://www.getpcparts.com/market-prices/gpu-graphics-cards/rtx-4090) Used values January to October 2026 · read 16 Oct 2026 27. 27 [PCPrice.watch, RTX 4090 used price](https://www.pcprice.watch/gpu-buying-guide/rtx-4090-price-used-and-specs) eBay sold median over 90 days · read 16 Oct 2026 28. 28 [DealHunter, used RTX 4090 price](https://dealhunter.io/prices/rtx-4090) Median asking price, 15 Oct 2026 · read 16 Oct 2026 29. 29 [EIA, Electric Power Monthly, table 5.3](https://www.eia.gov/electricity/monthly/epm_table_grapher.php?t=epmt_5_3) US residential electricity price · read 15 Oct 2026 30. 30 [SaladCloud blog, price changes of 12 September 2026](https://blog.salad.com/saladcloud-price-changes-september-2026/) Date of the current rates · read 15 Oct 2026 31. 31 [Runpod, Is the RTX 4090 still worth it for AI in 2026?](https://www.runpod.io/articles/guides/nvidia-rtx-4090) End of life, street prices in 2026 · read 16 Oct 2026 32. 32 [Tom's Hardware, eBay historical GPU prices 2023](https://www.tomshardware.com/news/gpus-historical-ebay-pricing) eBay average sold prices, September and October 2023 · read 16 Oct 2026 33. 33 [Wccftech, RTX 4090 sees massive price hike prior to RTX 5090 launch (8 Dec 2024)](https://wccftech.com/nvidia-geforce-rtx-4090-massive-price-hike-prior-to-rtx-5090-launch-over-2500-usd/) Retail prices in December 2024 · read 16 Oct 2026 What we could not confirm Several figures could not be confirmed at source on 16 October 2026. Clore.ai's $0.10 spot floor, its $0.365 and $0.375 medians and its $0.40 to $0.94 bare-metal quotes come from Clore's own snapshot of 21 September, not from live listings. Vast.ai's figures come from its public offer search at one moment and move by the hour; its pricing page itself rendered no figure for us, and the $0.14 floor in our first survey and on Vast's CLI page was not in the live offers. Runpod's $0.34 Community rate is in the pricing page's structured data, while the visible card showed only the $0.89 Secure rate. Novita AI's stock, its spot and reserved rates, all Oblivus figures and Runpod Community's stock come from GetDeploying, which on 16 October also listed a Lium node at $0.62 that Lium's own page did not show. TensorDock's typical $0.35 sits on a page last updated in July 2024. io.net's explorer would not load, so its $0.25 rate was dropped, as were Hostinger, Shadeform, Sesterce, Akash and LeaderGPU, which we could not reread. packet.ai's page names two different regions. Whether GPUBank's bonus credit can pay for reservations is not stated. We found no checkable L40S result on the three benchmarks used here. Our first survey quoted the 2017 licence wording with a blockchain exception as current; the current clause 2.8 has none. The rent-or-buy arithmetic rests on stated assumptions for the machine cost, resale and power draw. The SDXL timings from Formula Mod and Spheron were read on 15 October and are vendor blog figures without full settings; Formula Mod's site did not answer on 16 October, so we could not reread it. RTX 4090 on GPUBank ## The cheapest whole machine with a 4090 inside. - 8 vCPU, 41 GB RAM, 500 GB NVMe, IPv4 and traffic in $0.23/hr - Billed by the second after the health check - Interruptible $0.14, reserved $0.18 [Launch an RTX 4090 · $0.23/hr](https://console.gpubank.io/launch?gpu=rtx-4090) [Compare 63 GPU clouds](https://gpubank.io/compare) More GPU price guides ## Where to rent the other cards at the cheapest price. [B200 SXM 180 GB HBM3e From $2.49 spot · $4.15 all-in](https://gpubank.io/rent-b200) [H200 SXM 141 GB HBM3e From $1.49 spot · $2.49 all-in](https://gpubank.io/rent-h200) [H100 SXM 80 GB HBM3 From $1.11 spot · $1.85 all-in](https://gpubank.io/rent-h100) [H100 PCIe 80 GB HBM2e From $0.81 spot · $1.35 all-in](https://gpubank.io/rent-h100-pcie) [MI300X 192 GB HBM3 From $0.99 spot · $1.65 all-in](https://gpubank.io/rent-mi300x) [RTX PRO 6000 96 GB GDDR7 ECC From $0.69 spot · $1.15 all-in](https://gpubank.io/rent-rtx-pro-6000) [A100 SXM 80 GB HBM2e From $0.57 spot · $0.95 all-in](https://gpubank.io/rent-a100) [L40S 48 GB GDDR6 ECC From $0.33 spot · $0.55 all-in](https://gpubank.io/rent-l40s) [RTX 5090 32 GB GDDR7 From $0.29 spot · $0.48 all-in](https://gpubank.io/rent-rtx-5090)