Generate video with H3
Check which GPUs the image supports, then review instance, image and disk charges.
Compshare · View H3 image options ↗Read setup steps & cost breakdown →RTX 5090
32 GB VRAMNVIDIA
Last quote collected:
Starting from
$0.4411
per GPU / hour
vast.ai · 2026-10-26
Rentable when checked
Per GPU on a 2-GPU node
$0.88/node/hour
Price Comparison Across Platforms
Latest collected quote per provider. Valid records from the last 48 hours appear first, then sorted by price. Check collection dates and confirm current rates and inventory with the provider.
| Platform | Price/Hour (USD) | Price/Hour (CNY) | Availability | Quote collected | Action |
|---|---|---|---|---|---|
vast.aiLowest recent quote 2-GPU node · prices per GPU $0.88/node/hour Host verification revoked Japan, JP | $0.4411 | ¥3.20 | Rentable when checked | Rent → | |
Compshare 1 GPU · 16 CPU cores / 96GB host RAM cn-wlcb · cn-wlcb-01 Hourly whole-instance quote · native CNY Official configurations & billing ↗ | $0.5503 | ¥3.99 | Rentable when checked | Rent → | |
Daytona | $1.2900 | ¥9.35 | Confirm availability | Check availability → |
Same RTX 5090, different instance configurations
CPU, host RAM, minimum GPU count and extra charges can differ. An hourly price alone is not a like-for-like configuration comparison.
Compare configurations and 10 / 100-hour instance costs
Swipe to see whole-instance costs and extra charges →
| Provider / observed | Recorded configuration | Whole instance / hour | 10 hours / instance | 100 hours / instance | Check before ordering |
|---|---|---|---|---|---|
| vast.aiRentable when checked | 2-GPU nodeCPU and host RAM not recordedHost verification revokedJapan, JP | $0.88Original whole-node quote | $8.82 | $88.22 | Storage, transfer, host status and minimum rental conditions |
| CompshareRentable when checked | 1 GPU · 16 CPU cores / 96GB host RAMcn-wlcb / cn-wlcb-01 | $0.55Native CNY 3.99 / hourWhole-instance quote; do not multiply by GPU count | $5.50Native CNY 39.90 | $55.03Native CNY 399.00 | Image, additional storage, networking and billing plan |
| DaytonaConfirm availability | Full configuration not recorded | —Cannot infer from a per-GPU rate | — | — | GPU count, CPU, RAM and included charges |
Estimates use the hourly rate with no monthly or volume discount. Storage, networking, images and taxes are excluded. Converted amounts use a fixed display rate of USD 1 = CNY 7.25, not a dated market rate; check the native-currency total at checkout. Budgets marked “Confirm availability” are quote references, not confirmed stock.
Trying an RTX 5090 on Compshare?
After signing in, select RTX 5090 and check CPU, host RAM and region. Compare hourly billing for a trial and monthly billing for a persistent service. Review the complete order breakdown before creating an instance.
Choose GPU configuration ↗See hourly and monthly configuration examples →Compared prices? Choose your next step
Host a language model
4090 has 24GB; 5090 has 32GB. Model size, quantization and concurrency determine the memory you need.
Choose memory & concurrency capacity →Running daily? Compare monthly rental
Compare whole-instance monthly fees, capacity reservation and renewal terms alongside hourly rates.
Compare hourly & monthly terms →7-Day Price History
Latest available daily quote per platform over the past 7 days.
4090 or 5090: compare a completed job
Keep the model, precision, image and output settings constant. Record startup time, runtime, peak memory and retries. A higher hourly rate does not necessarily mean a higher job cost, but hardware generation alone does not establish performance.
Price check
Confirm per-GPU versus whole-server billing, minimum duration, disk and network charges. Check collection dates and provider inventory; include downloads and initialization for short jobs.
Workload fit
Check framework, CUDA, driver and kernel support in the provider image. Successful startup does not prove your batch, context or output resolution can finish reliably; test the intended settings.
Compare next
If the task fits a 4090, compare completed-job costs on both cards. If memory remains insufficient, evaluate H200, B300 or a model API before repeatedly switching instances.
FAQ
How much does RTX 5090 rental cost?
RTX 5090 is quoted from $0.4411/GPU/hour on vast.ai. This is a per-GPU hourly quote on a 2-GPU node. The whole node costs $0.88/hour. The whole node must be rented; it is not a single-GPU rental offer. Prices can change with inventory; check the collection date and provider billing terms.
What is RTX 5090 best for?
RTX 5090 has 32GB VRAM, making it ideal for LLM inference, AI training, and other high-performance computing workloads.
Which platform has the cheapest RTX 5090?
Among currently tracked quotes, vast.ai has the lowest RTX 5090 quote at $0.4411/GPU/hour. This is a per-GPU hourly quote on a 2-GPU node. The whole node costs $0.88/hour. The whole node must be rented; it is not a single-GPU rental offer. Check the collection date, availability and provider billing terms.
Is renting an RTX 5090 always better value than a 4090?
No. Compare runtime, hourly cost, startup overhead, retries and extra charges for the same workload and output requirements. Neither hardware name nor hourly price alone establishes value.