Image and video generation
Pick the card by VRAM: RTX 4090 (24 GB) at $88 for 7 days, RTX 5090 (32 GB) at $124, or 48 GB on the RTX A6000 ($66), RTX 6000 Ada ($104) or L40S ($118). The card is yours alone, the price is fixed for the term, and you pay on-chain from a username-only account.
Create your account, deposit from $20, credited in about one minute.
VRAM guide
Weights set the floor; resolution, batch, frames and training state set the peak.
Weight memory is parameters times bytes per parameter. FLUX.1 [dev] has 12 billion parameters: 24 GB in 16-bit, about 12 GB in 8-bit, before text encoders, VAE, LoRAs and activations. Our guide to sizing GPU memory explains the method.
| Workload | Memory driver | Start with | Price |
|---|---|---|---|
| SDXL in ComfyUI | Resolution, batch, ControlNets | RTX 4090, 24 GB | $88 / 7 days |
| Flux.1 [dev], 8-bit | About 12 GB of weights | RTX 4090, 24 GB | $88 / 7 days |
| Flux.1 [dev], 16-bit | About 24 GB of weights | RTX 5090, 32 GB | $124 / 7 days |
| SDXL LoRA training | Weights, gradients, optimizer state | RTX 4090 or RTX A6000 | $88 or $66 / 7 days |
| Flux LoRA training | 24 GB of weights plus training state | RTX 6000 Ada, 48 GB | $104 / 7 days |
| Video, Wan 2.1 14B class | 28 GB of 16-bit weights; frames processed together | L40S or RTX 6000 Ada, 48 GB | $118 or $104 / 7 days |
Small video models fit 24 GB: Wan's model card lists 8.19 GB for its 1.3B model. Validate one real job on a 7-day term first, as our AI video GPU guide recommends.
The RTX 5090 (Blackwell, 32 GB GDDR7) adds 8 GB over the RTX 4090 (Ada, 24 GB GDDR6X) on the same 16 vCPU, 128 GB RAM and 2 TB NVMe host, for $36 more per week. Blackwell needs a CUDA 12.8-class stack, which the PyTorch 2.8 template targets. Compare the RTX 5090 and RTX 4090.
Prices
| GPU | VRAM | Host per card | 7 days | 30 days | Per hour, 30 days | 4 cards, 30 days |
|---|---|---|---|---|---|---|
| RTX A6000 | 48 GB GDDR6 ECC | 14 vCPU · 96 GB · 2 TB | $66 | $232 | $0.32 | $854 |
| RTX 4090 | 24 GB GDDR6X | 16 vCPU · 128 GB · 2 TB | $88 | $312 | $0.43 | $1,148 |
| RTX 6000 Ada | 48 GB GDDR6 ECC | 16 vCPU · 128 GB · 2 TB | $104 | $368 | $0.51 | $1,354 |
| L40S | 48 GB GDDR6 ECC | 16 vCPU · 128 GB · 2 TB | $118 | $416 | $0.58 | $1,531 |
| RTX 5090 | 32 GB GDDR7 | 16 vCPU · 128 GB · 2 TB | $124 | $432 | $0.60 | $1,590 |
For SD 1.5, the 24 GB RTX 3090 is $52 a week. Options, per week or month: IPv4 $3 or $10, 10 TB bandwidth $12 or $39, 1 TB NVMe $9 or $29.
Price it
Pick the card, node size, region and term; the quote shows the full USD total before any charge.
Why choose AnchorGPU
The RTX A6000 ($66 a week) and RTX 6000 Ada ($104) undercut the $124 RTX 5090 and leave 24 GB of headroom beyond 16-bit Flux weights. BTCServers, the no-KYC host we compared, jumps from its 32 GB RTX 5090 to an 80 GB A100 at $690 a month. See the RTX 6000 Ada.
Each RTX 4090, RTX 5090, RTX 6000 Ada and L40S comes with 16 vCPU, 128 GB of RAM and 2 TB of NVMe: room for ComfyUI model offload and a large LoRA library. RunPod lists its RTX 4090 Pod with 6 vCPU and 41 GB of RAM. See the RTX 4090 configuration.
The term is charged once and its price cannot change; renewal reuses the accepted price. Overnight queues and idle hours cost nothing extra. See how billing works.
Seven assets, Monero included, fund a username-only account from $20, credited within about a minute once the network confirmations are reached. RunPod asks for any required KYC before a first crypto payment; BTCServers requires an email. Rent a GPU with Monero.
Nodes of 2, 4 or 8 identical cards take 5, 8 or 10% off each. Run one ComfyUI worker per GPU: four queues on 4× RTX 4090 cost $324 a week. Price an 8× RTX 4090 node.
Next step
Create the account, then price the exact card and term; availability is confirmed before any payment.
Facts
| Item | What you get |
|---|---|
| Price unit | One dedicated card, in USD, for exactly 7 or 30 days; no hourly meter |
| Minimum | One card for 7 days: $44 (RTX A5000); $66 for 48 GB (RTX A6000) |
| Minimum deposit | $20 (USD 20 equivalent in crypto) |
| Node discount | 2 cards 5%, 4 cards 8%, 8 cards 10%, applied to every card; options are added once per node and term |
| Availability | Available — deposit from $20, credited in about one minute |
| Regions | Dallas (US-TX-1), Ashburn (US-VA-1), Portland (US-OR-1) |
| Crypto assets and networks | BTC on Bitcoin (1 confirmation), ETH on Ethereum (12), USDT on TRC-20 (19), USDC on ERC-20 (12), XMR on Monero (10), SOL on Solana (32), LTC on Litecoin (6); credited within about a minute once the network confirmations are reached |
| KYC scope | Never asked at signup: name, address, email, phone, selfie, ID. Kept: username, SSH public keys, ledger and deposit records. The Acceptable Use Policy and valid legal process apply. |
| Charge timing | Full term debited once from your USD balance when the deployment is accepted or starts |
| Fees | No setup fee; $0 deposit fee on the quote; the sender pays network fees |
| Provisioning time | 20 to 90 minutes depending on configuration |
| Key conditions | Stopping does not pause the term; no prorated refund for early release; 72-hour grace after expiry, then local data is deleted |
Detail: billing ledger, crypto FAQ, terms.
Compared
Hourly rates multiplied by 720 hours, our 30-day term.
| GPU | AnchorGPU, 30 days | RunPod Secure, 720 h | RunPod Community, 720 h | Vast.ai on-demand, per hour | BTCServers, per month |
|---|---|---|---|---|---|
| RTX 4090 | $312 | $532.80 ($0.74/h) | $244.80 ($0.34/h) | From $0.14 (64+ offers) | $189 |
| RTX 5090 | $432 | $712.80 ($0.99/h) | $496.80 ($0.69/h) | From $0.34 (64+ offers) | $279 |
| L40S | $416 | $784.80 ($1.09/h) | $568.80 ($0.79/h) | $0.63 to $0.87 (8 offers) | Not listed |
| RTX 6000 Ada | $368 | $604.80 ($0.84/h) | $532.80 ($0.74/h) | $0.46 to $0.80 (9 offers) | Not listed |
| RTX A6000 | $232 | $381.60 ($0.53/h) | $237.60 ($0.33/h) | $0.29 to $0.60 (8 offers) | Not listed |
Against RunPod Secure, break-even is 12.7 to 14.6 hours of use a day, depending on the card; above that, the fixed term costs less on all five. At 24/7 it also beats RunPod Community on every card except the RTX 4090; BTCServers and the lowest Vast.ai offers cost less on the RTX 4090 and 5090.
| Criterion | AnchorGPU | RunPod | Vast.ai | BTCServers |
|---|---|---|---|---|
| Crypto route | On-chain, 7 assets | Payment processors; coins not published | BitPay and Crypto.com | BTC, XMR, LTC, ETH, USDT and 3 others |
| Monero | Yes | Not published | Not published | Yes |
| Identity | Username and password | “Complete any required KYC verification before your first crypto payment” | Not published | Email and password; burner email accepted |
| Billing | Fixed 7 or 30 days | Per second | Per second; reserved 1, 3 or 6 months | Hourly or monthly |
| Host per RTX 4090 | 16 vCPU, 128 GB RAM, 2 TB NVMe | 6 vCPU, 41 GB RAM | Varies by host | 16 vCPU, 128 GB RAM, 2 TB |
Setup
A 3 to 32 character username and a 12 to 128 character password. Store the eight recovery codes offline.
Run ssh-keygen -t ed25519 -a 64 and paste the .pub line. SSH key guide.
Card, 1 to 8 GPUs, region, 7 or 30 days, options.
ComfyUI template, PyTorch 2.8 with CUDA 12.8, or Ubuntu 24.04; administrative access covers custom nodes and containers.
Availability is confirmed in your account first; the term is debited when the deployment is accepted or starts.
Run ssh -L 8188:localhost:8188 user@host, open localhost:8188, and sync outputs off the node.
Walkthrough: prepare your first node request and choose a machine image.
Before you rent
Card plus options: an RTX 4090 for 30 days is $312, or $322 with a dedicated IPv4. No setup fee; the sender pays network fees; deposits start at $20.
Availability and provisioning time are confirmed in your account for each order, before any payment. A quote on its own does not reserve capacity. Crypto deposits are credited within about a minute once the network confirmations are reached.
You can check everything first: per-card prices, node discounts and options are public on the pricing page, your account shows the full USD total and confirmed availability before any debit, and that price holds for the whole term. No uptime SLA is published; the 7-day term keeps your first commitment small.
Early release is not refunded pro rata, so prove the workflow on 7 days first. RunPod credits and Vast.ai credits bought with crypto are not refunded either.
The intended design does not inspect workload content by default; the Acceptable Use Policy and valid legal process still apply.
Our recommendation
Right for you if the card runs most of the week and you pay on-chain without an email. Not the right fit for a few hours of work, non-US regions or card payment.
Get started
Create a username account, then price your card for 7 or 30 days; deposit from $20, credited in about one minute.
FAQ
About 24 GB for FLUX.1 [dev] weights in 16-bit, 12 GB in 8-bit, plus encoders, VAE and LoRAs. 8-bit Flux fits a 24 GB RTX 4090; for 16-bit, rent 32 GB (RTX 5090) or 48 GB (RTX 6000 Ada, RTX A6000, L40S).
Link to this answer ↗
It has 32 GB of VRAM against 24 GB, on the newer Blackwell generation, for $124 instead of $88 per week. Take it when 24 GB is the limit; otherwise the RTX 4090 saves $120 per 30 days.
Link to this answer ↗
Yes for SDXL generation, and usually for SDXL LoRA training: 24 GB, 16 vCPU, 128 GB RAM and 2 TB NVMe for $88 a week. If your peak exceeds 24 GB, the 48 GB RTX A6000 costs less, at $66.
Link to this answer ↗
No. The shortest term is 7 days: $88 for an RTX 4090, or $0.52 per hour of the term, and stopping the node does not pause it. For a single short job, an hourly cloud costs less.
Link to this answer ↗
BTC, ETH, USDT on TRC-20, USDC on ERC-20, XMR, SOL and LTC. Prices stay in USD, the conversion is indicative and time-stamped, and the sender pays network fees. Deposits start at $20 and are credited within about a minute once the network confirmations are reached. See the crypto payment guide.
Link to this answer ↗
No. Signup takes a username and a password, with no name, email, phone or ID. No KYC is not anonymity: the ledger, deposit records and SSH keys are kept, and valid legal process applies. Read the no-KYC scope.
Link to this answer ↗
Auto-renew is off by default. After expiry comes a 72-hour grace period, then the instance is released and local data permanently deleted. Backups are not part of the published offer, so sync outputs off the node.
Link to this answer ↗