Medical device algorithm research and development has always required a specialized PC with high-end compute power to explore the most advanced algorithms.
The invention of the parallel computing platform and programming model CUDA has driven the integration of NVIDIA GPUs in many medical devices, medical instruments, and models. This has enabled software development cycles for high-performance algorithm workloads. Today, AI in medical imaging workflows has been adopted and is naturally progressing into other areas where medical imaging is used, such as in medical devices with video or image captures. The desire to integrate AI into medical devices has created an even greater need for computing resources to accelerate the development of new features.
We are introducing the NVIDIA Clara AGX Development Kit to address this requirement, enabling research of advanced algorithms and AI skills that target medical instruments. The Clara AGX Development Kit includes a Jetson AGX Xavier, Mellanox Connect X-6 NIC, RTX 6000 GPU, and HDMI 2.0 input. The Clara AGX SDK provides the necessary OS, drivers and libraries, along with a containerization runtime and example AI skills. This combination of hardware and software creates a unique platform that allows medical device manufacturers, software developers, and the medical research community to bring AI advancements to medical instruments more easily.
Scalable Compute for AI Skill Development
Built on top of Jetson AGX Xavier, the Clara AGX Development Kit includes an incredibly rich and diverse ecosystem of libraries and tools supported by the Jetson product family. The Jetson AGX Xavier module makes AI-powered autonomous machines possible, running in as little as 10W and delivering up to 32 TOPs. As part of the world’s leading AI computing platform, it benefits from NVIDIA’s rich set of AI tools and workflows, enabling developers to quickly train and deploy neural networks. The Jetpack SDK, a comprehensive solution for building AI applications, supports the Clara AGX Development kit with a standard Linux operating system, a BSP supporting the board’s components, and CUDA support, which brings with it the entire ecosystem of CUDA-enabled acceleration. Additionally, the Clara AGX SDK includes Rivermax and DeepStream for building streaming data processing pipelines, a containerization runtime to support deployment via EGX, and reference applications modeling specific use cases.
While the Jetson platform is a great technical fit for productization, developing next-generation algorithms requires the increased computing resources provided by an NVIDIA discrete GPU. The RTX GPU’s high performance reduces the amount of time spent optimizing code and enables algorithm experimentation without compromising performance.
The Clara AGX Development Kit includes the NVIDIA RTX 6000 GPU, delivering 200+ INT8 AI TOPs of peak performance. The RTX 6000 can run multiple simultaneous inference pipelines of the most complex and demanding AI applications, facilitating the research and development of your latest AI models.
When the R&D phase is complete and the final AI models are ready, developers can optimize and deploy them on a fully scalable architecture. Those looking for lower power use cases can utilize the Jetson product family that meets a smaller form factor’s power and size budget. High-performance use cases can be deployed on a productized system that targets the right mix of the Jetson product and RTX GPU. We are delivering a truly complete and scalable product family from one set of application software.
Scalable Connectivity for High Bandwidth Sensors
Medical instruments, along with many other embedded systems, require high bandwidth connectivity between a variety of upstream sensors and GPUs, via GPU Direct or RDMA streaming. The Clara AGX Development Kit allows customers to make these specialized connections over two different media.
First, there is support for both 100G Ethernet and 10G Ethernet by integrating the NVIDIA Mellanox ConnectX-6 network interface card (NIC). This advanced device supports GPU Direct to the RTX 6000 GDDR DRAM at a full line rate. The Clara AGX Development Kit enables offloading the networking stack allowing the streaming sensor data packets to achieve the highest possible throughput and performance with minimal impact on CPU utilization.
Second, two PCIe Gen4x8 slots are available to users, supporting custom PCIe cards or additional network interface cards. The first PCIe port is connected to the integrated switch in the NVIDIA Mellanox ConnectX-6 NIC. This port supports RDMA data transfer directly into the RTX6000 GDDR DRAM. Example uses for this port include custom video capture cards with multiple 4k60 12G SDI inputs. The second PCIe port is connected directly to the Jetson AGX Xavier module. This port can connect to an external host or enable specialized output via a custom card.
Partner Developer Program
We’re also introducing a Partner Developer Program for the Clara AGX Development Kit. The Partner Developer Program includes NVIDIA hardware and collaborative networking. Apply through the link above or contact your NVIDIA representative to learn more about getting a Clara AGX Development Kit.
If you’re looking to learn more about Clara AGX, check out the in-depth presentation on our Fall GTC page called Architecting AI-Powered Medical Instruments by Mathias Blake, Juha Tukkinen, and Dr. Brian Feeley. They discuss how the AI revolution will create massive value for medical devices and how Clara AGX will play a role in that revolution.