B300
Blackwell
- Vast reference
- $9.94
- nafnaf estimate
- $9.17
USD per GPU per hour
$0.77 less per GPU hour in this model
Vast #148737 · 1 GPU
Last checked 10 Oct 2026, 00:54 UTC
A future vision of distributed compute
On-demand GPUs. A smaller platform cut.
More from
every GPU hour.
Rent the compute you need. Put your own GPUs to work. A smaller platform cut means more value for renters and suppliers.
On-demand compute · Ubuntu + NVIDIA hosts
Watch the one-minute demos Compare B300, H200 & RTX PRO 6000The GPUs you want. A smaller platform cut.
Serious GPUs.
Less per hour.
See what a verified Vast GPU hour could cost under nafnaf's shared-savings model. Same reference price. More of the value for you.
H200
Hopper
- Vast reference
- $5.00
- nafnaf estimate
- $4.62
USD per GPU per hour
$0.38 less per GPU hour in this model
Vast #152822 · 1 GPU
Last checked 10 Oct 2026, 00:54 UTC
RTX PRO 6000
Blackwell Max-Q Workstation
- Vast reference
- $1.00
- nafnaf estimate
- $0.92
USD per GPU per hour
$0.08 less per GPU hour in this model
Vast #140962 · 1 GPU
Last checked 10 Oct 2026, 00:54 UTC
Illustrative prices, not bookable quotes. References refresh in the background; if a lookup fails, we keep its last checked price and date. We use the lowest per-GPU rate among returned verified offers for each exact model, assume an 80% supplier share on Vast, and apply nafnaf's 5% fee with equal dollar savings for renter and supplier. Bundle sizes vary; Vast requires the full bundle. Storage and transfer are separate. See the calculation ↓
The proposed split
Put the fee back
on both sides.
Keep a small fee. Split the rest of the marketplace spread equally: every dollar saved by the renter adds a dollar to the supplier's revenue.
| Compute | Vast example | nafnaf model |
|---|---|---|
| Renter pays | $100.00 | $92.31 |
| Supplier receives | $80.00 | $87.69 |
| Platform keeps | $20.00 | $4.62 |
The renter saves $7.69. The supplier earns $7.69 more. nafnaf keeps 5% of its own charge. Supplier revenue is before operating costs.
Vast describes prices as typically about 25% above host earnings, which implies a 20% share of renter spend in this example. Source: Vast, May 2026. Actual splits vary; use matched renter prices and host payouts. Compare the same rental terms. Storage and transfer are separate.
Your hardware. Ready for more.
Connect. Set your price. Get to work.
Connect your host
Install the nafnaf agent on your Ubuntu NVIDIA machine. It checks the hardware and makes eligible, free GPUs available to rent.
Choose your price
Use autopilot to follow market benchmarks, or set your own rates and rental terms. Review everything in your dashboard.
Run the workload
Renters launch a Docker workspace on an available GPU. Each booking reserves its GPU until the rental ends and its resources are released.
Respect every booking.
Existing workloads keep their allocated GPUs. New rentals use available capacity, and each confirmed booking keeps its agreed compute rate for the rental.
What if my machine has multiple GPUs?
Each free GPU can accept a separate nafnaf rental, one GPU per booking. GPUs already allocated to a workload remain reserved until that workload releases them.
Do I need another marketplace account?
No. Rent directly with your nafnaf account, or install our agent to offer your own GPU host. No other marketplace account or installation is required.
Can I keep my existing workloads?
Yes. The agent checks GPU allocations before accepting a booking. If you also offer the host on another marketplace, configure its supported integration to coordinate new bookings. See the optional co-hosting instructions.
Is an idle rented GPU available?
No. An existing renter's allocation remains theirs even when it is not doing GPU work. Shared CPU, memory, disk, and network capacity must also remain available to existing rentals.