Amsterdam-based Juvoly, a company developing a proprietary speech model, announced that it has officially begun using Nvidia DGX B200 supercomputers to power its new AI model for speech recognition in the healthcare sector.
The Nvidia DGX B200 systems are now up and running at NorthC’s data centre in Rotterdam.
In addition to hosting Juvoly’s models, the facility also provides necessary storage and processing services.
Constantijn Van Oranje-Nassau unveiled the first NVIDIA DGX B200 chips in the Netherlands at the touch of a button.
Unlike other data centers that provide Tesla V-GPUs, Juvoly’s supercomputer is a fully integrated system designed for AI applications.
It not only features powerful GPUs but also optimised software, resulting in efficient performance, fast data processing, and low latency, reports TechZine.
What makes NVIDIA DGX B200 an ideal choice?
NVIDIA DGX B200 is a unified AI platform designed for businesses.
It features eight NVIDIA Blackwell GPUs connected via fifth-generation NVIDIA NVLink, delivering three times the training speed and fifteen times the inference speed of previous models.
This makes it ideal for handling diverse workloads like large language models, recommender systems, and chatbots.
NVIDIA DGX B200 provides the computing power needed for complex AI tasks throughout the entire pipeline, from training to inference, helping developers streamline their workflows.
- High Performance: Offers 3X training and 15X inference speed compared to DGX H100.
- Versatile Infrastructure: Fully optimized hardware and software platform, including the NVIDIA AI software stack and expert support.
This platform is perfect for businesses looking to accelerate their AI capabilities, claims Nvidia.
Juvoly V2: OpenAI Whisper’s rival
Juvoly has created a speech recognition model tailored specifically for medical conversations in Dutch.
Their latest model, Juvoly V2, claims to be 10 per cent more accurate than OpenAI’s Whisper and operates 40 times faster.
The model also adapts to different dialects and accents, ensuring patients’ speech is accurately understood.
The Dutch company’s solution has received support requests from various regions, including Friesland, and has adapted its model based on regional needs.
In addition to its accuracy, the Juvoly V2 model is significantly more energy-efficient.
While traditional systems use around 11 kWh, Juvoly V2 only requires 350 Wh for every 100 users, says the report.
Importantly, Juvoly ensures that all data remains within Europe, supporting the call for sovereignty in AI development.
By collaborating with NorthC Datacenters and using Nvidia’s latest Blackwell hardware, Juvoly aims to deliver this promise to its users.
01
Upstream Festival is back for its sixth edition! 4 solid reasons to mark your calendar