RTX PRO 6000 Blackwell Max-Q
watson
$0.42/ GPU hour on nafnaf
Vast $1.0058% less than VastSame GPU model
95.6 GiBusable VRAM
Zug, Switzerland
Configure GPU01 / For consumers
GPU rental.Inference.
More compute.
Less between you and it.
02 / For providers
Your GPUs.Your upside.
You bring the hardware.
You keep more of the revenue.
Compute, on your terms
Less spend.
More possibility.
Rent a GPU for your workload, or call a model directly. Two ways to build. One nafnaf account.
Available GPUs
GPUs you can rent now.
Our lowest available rates. No long-term commitment.
USD, platform fee included. Storage and transfer are separate.
About the Vast comparison
The lowest verified single-GPU offer returned for the same model and memory class. Different hosts may have different CPU, RAM, network, and rental terms. Compute prices only; no assumed fee discount. Missing or expired references are not used.
OpenAI-compatible inference
Qwen3.8 Flash-Next
50%below OpenRouter's Qwen3.8 Flash rates
- Median decode
- 250 tok/s
| Tokens | nafnaf | OpenRouter |
|---|---|---|
| Input | $0.075 | $0.15 |
| Output | $0.235 | $0.47 |
| Cached input | $0.008 | $0.016 |
Add credit, connect your API key, and pay only for the tokens you use.
- Chat Completions
- Streaming
OpenRouter Qwen3.8 Flash benchmark, checked 22 October 2026. Flash-Next is not an exact OpenRouter listing; quality, context and capacity can differ. Throughput varies with the request and server load.
DeepSeek Flash remains free. GLM and MiMo have their own rates. Explore all models
Your hardware, working for you
Less overhead.
More yours.
A smaller fee, sensible hardware requirements, and pricing that follows the market.
The fee makes a difference
on host earnings
5%nafnaf consumer fee
on your rate
on your rate
Your rate. Yours to keep.
You set the provider rate for compute, storage, and metered transfer. Renters pay 5% on top, already included in their displayed prices. Ask $1.00, receive $1.00; the renter pays $1.05. Operating costs and applicable taxes remain yours.
Hardware that fits
No 95% RAM-to-VRAM rule.
Your GPUs don't need almost matching system RAM. Standard hosts start at 48 GiB RAM, with memory reserved per workload.
Less price watching
Pricing on autopilot.
Track GPU market benchmarks at the discount you choose. Prefer a fixed rate? Set it yourself. Existing bookings keep their agreed prices.
Market benchmark− your discountRenter price, fee included
Start providing
Connect. Set your rate. You're ready.
Install nafnafd on your Ubuntu NVIDIA host. Eligible, unallocated GPUs become available after the host checks pass. Existing workloads keep their GPUs.
Connect your host Host requirementsA shared foundation
Neutral ground.
Global ambition.
nafnaf is a neutral Swiss platform connecting people who need compute with people who provide it.
Swiss platform. Global hardware.
Inference hosted and managed by nafnaf. GPU rentals run on independent providers' hardware, with host locations shown in the marketplace.
GPU rental or inference?
Rent a GPU when you need a Docker workspace, your own model, or a custom workload. Qwen3.8 Flash-Next inference uses prepaid credit, with no GPU reservation. DeepSeek Flash remains free; check each model's published token rates and limits.
Can I rent out spare GPUs while other jobs run?
Yes. A GPU already assigned to your own job or an existing rental is not offered to new renters, even when its utilization is zero. Only separate, unallocated GPUs can accept new rentals after the host checks pass. Review host setup.
Are all GPUs hosted in Switzerland?
No. nafnaf is a Swiss platform; independent rental hosts can be elsewhere. Listings show an approximate location based on the host's public IP address when available, not a verified physical address. The Earth scene illustrates distributed compute, not live host locations.