Rent an RTX PRO 6000 Blackwell Server Edition
NVIDIA RTX PRO 6000 Blackwell Server Edition
Specifications
Renting a RTX PRO 6000 Blackwell Server Edition
RTX PRO 6000 Blackwell: what 96 GB on one card does
The RTX PRO 6000 Blackwell Server Edition puts 96 GB of GDDR7 on a single card. The usual way to get that much memory is to split a model across several smaller cards, and the split has a cost: every token crosses the link between them. One card pays none of it.
We benchmarked it against a pair of RTX 4090s with the same model, the same vLLM version and the same arguments. On a single request the two are within 12% of each other. Under load, and on long prompts, they are not close.
On ParalonCloud a machine is rented whole and billed by the minute, with Jupyter Lab in the browser and no contract. This card is rare on the network; the top of this page shows whether one is free right now, and the list price when none is.
What fits
What fits in 96 GB
With a 27B model at 4-bit loaded, almost all of the card is left for context. That changes what the model can be asked to do.
Measured
Measured: RTX PRO 6000 against two RTX 4090s
Qwen3.8-27B W4A16 on vLLM 0.27.1, same arguments on both machines. Two runs at each point agreed within 4%.
Steps
How to rent one
- 01
Check availability
The top of this page says whether a machine with this card is free. If none is, the rent page lists what is, including pairs of RTX 4090s.
- 02
Start the rental
Choose Jupyter Lab, accept the terms and start. You need credits for one hour to open a session and are charged only for the minutes you use.
- 03
Work, then stop
Use the terminal for pip, conda or apt, run your job, download what you want to keep, and stop.
Limits
Know before you rent
- This card is rare on the network and is often not available. The price shown when none is free is the list price, not an offer.
- A rental is an unprivileged container: Docker inside it does not work. Install what you need with pip, conda or apt.
- Access is through Jupyter Lab and its terminal. Other ports are not exposed to the internet.
- Everything on the machine is wiped when the rental ends. Download results before you stop.
FAQ
Questions about the RTX PRO 6000 Blackwell Server Edition
Is one RTX PRO 6000 faster than two RTX 4090s?
On a single request, barely: 106 against 94 tokens per second on a 27B model. At 32 concurrent requests it delivered 1,342 against 435, about three times as much, because nothing has to cross between cards. On long prompts it prefills about three times as fast.
How long a prompt can it take?
With a 27B model at 4-bit loaded, the card holds 1,843,794 tokens of context cache, enough for seven requests at the model's full 262,144-token context at once. A 256,361-token prompt took 99.3 seconds to its first token.
When is the big card worth its price?
When you serve many users at once, or when prompts run past what a 24 GB or 32 GB card can hold. For one user and short prompts a single RTX 4090 or RTX 5090 costs much less for almost the same speed per request.
Is it available to rent right now?
The badge at the top of this page says so, from the live network. Machines with this card come and go; when none is free the page shows the list price and the rent page shows what is available instead.
How do I know the card is real?
Before a machine is listed, a verification job allocates its memory and measures memory bandwidth and compute against the card's specification. The result, and when it was last checked, is on the machine's page.