What is FuriosaAI?
An efficient accelerator for LLM inference at 180W.
Who is FuriosaAI best for?
FuriosaAI is best for data centres that run language models on their own hardware and are limited by rack power.
Power-limited data centres
An RNGD card draws 180W and an eight-card NXT RNGD Server about 3 kW, so more inference fits into racks capped at around 15 kW.
On-premises LLM inference
The server can run on site, in a managed environment or in a colocation facility, for teams that keep their models in-house.
ML engineers moving PyTorch models
The Furiosa SDK works with PyTorch 2.x and supports containers, Kubernetes and SR-IOV for sharing cards across workloads.
Computer vision workloads
A separate Vision NPU is sold for computer vision models, alongside the RNGD cards for language and multimodal models.
How much does FuriosaAI cost?
FuriosaAI does not publish prices: its RNGD cards and NXT RNGD Server are sold by quote, and companies can apply to test them through the Furiosa Access Program.
FuriosaAI has no pricing page. The RNGD page offers Request a quote, and the Furiosa Access sign-up form lets a company ask for bare metal, virtual machine or model API access to evaluate the hardware in Seoul, the Bay Area, Lisbon or Johor Bahru. Plans and current prices are on furiosa.ai.
How do readers rate FuriosaAI?
No one has rated FuriosaAI yet. Be the first.
Sign in to rate. One rating per account.
What are the best FuriosaAI alternatives?
One tool on Recaply does FuriosaAI's job.
Questions about FuriosaAI
What is FuriosaAI?
An efficient accelerator for LLM inference at 180W.
Who is FuriosaAI for?
FuriosaAI is best for data centres that run language models on their own hardware and are limited by rack power.
Is FuriosaAI free?
No. FuriosaAI is a paid tool.
What are the best FuriosaAI alternatives?
The closest alternatives on Recaply are Rebellions.
Where can I try FuriosaAI?
On its official website, furiosa.ai.
Can I test FuriosaAI chips before buying?
Yes. Companies can apply to the Furiosa Access Program, which offers bare metal, virtual machine or model API access to evaluate RNGD, online or at sites in Seoul, the Bay Area, Lisbon and Johor Bahru. The form asks for your company and the models you want to run.
Mythic
Modal
Together AI
Salad
Prime Intellect
Modular