The medical computing market is one in which NVIDIA has grown the most in recent years and where its advances in artificial intelligence have allowed it to become one of the largest technology providers in this regard, with Clara AGX being the bet of those of Jensen Huang in said market. Thus, it is a workstation designed for scientific computing, but oriented especially to the most lucrative and most interesting market in the scientific world, such as medicine.
What is Clara AGX?
Clara AGX is a PC that combines NVIDIA Tegra SoCs with graphics cards from the same brand in a single machine. So from a certain point of view they are an adaptation to medical computing of the Drive PX for autonomous vehicles.
After the Tegra fiasco in the tablet and mobile market, everyone knows that they decided to orient them to other markets, especially to take advantage of the turn of the creators of the GeForce towards the artificial intelligence market. Which allowed them to set foot in two different industries of great importance. The first of these is the growing market for automated driving. The second, on the other hand, is related to the enormous health market, which includes several applications: from the capture of medical information through tomography to the synthesis of proteins for the creation of new drugs.
The most important change that NVIDIA made in its Tegra from that moment was the inclusion of a PCI Express interface, this allowed them to connect a graphics card and optimize the joint launch of both parts. Thus, they make sure to give their platforms like Clara AGX a capacity for interaction between both parties that is not possible on the PC, the most relevant being the use of a totally unified memory space in terms of addressing. That is, 100% consistent.
What is inside a Clara AGX unit?
Each Clara AGX workstation is made up of the following components:
- A state-of-the-art NVIDIA Tegra SoC. In the first version of the Clara AGX it was a Xavier, but it has been updated to the most recent architecture which is Orin.
- An NVIDIA graphics card aimed at the professional market. Currently, they are using NVIDIA Quadro RTX 6000 GPUs with Turing architecture, equivalent to an RTX 2080 Ti. They currently offer an A6000 with the same capabilities as an RTX 3090.
- An NVIDIA ConnectX network card with a SmartNIC, allowing it to communicate with other units. It has two network ports, one at 100 Gbps of the QSFP28 type and another RJ45 or Ethernet at 10 Gbps.
- Two PCIe Gen 4 interfaces with 2 lines each.
- 250 GB of storage in NVMe SSD format
Built for AI and real-time computing
On the other hand, as many of you may have deduced, the operating system used in Clara AGX is not Windows, but a GNU/Linux distribution for ARM optimized for its hardware and with libraries and applications widely used in medical computing. That is why the Clara AGX is not a conventional PC, but rather a workstation for creating medical applications in real time.
But what does real time mean? Refers to applications where interrupt requests to the CPU are executed at the time the interrupt is generated. So the way they are managed by the operating system is different from how it is traditionally done. They share this change with the Drive PX platform, where timely response to information captured by sensors and user interaction is crucial for safety.
Although its main application is to take advantage of the artificial intelligence-optimized hardware of the current Tegra SoCs and GPUs, especially units such as the NVDLA or Tensor Cores, for the development of artificial intelligence models and algorithms. The system also provides a series of previously trained applications and models and with versions of applications from the scientific world optimized for the hardware that Clara AGX integrates by NVIDIA itself.
Why is AI important in medicine?
The diagnosis of diseases is made by doctors from information based on whether certain symptoms are present in the patient. In some cases the symptoms can be read through the information that has been obtained visually. For example, it may be that something that seems innocuous to the patient, such as a small spot on the skin, indicates something more serious.
One of the things we are training AI systems with is learning to draw conclusions from visual information. This allows them to make predictions from training with medical images and inference algorithms about the health status of different tissues, which will have been obtained from different devices used for diagnosis. Specifically, NVIDIA has oriented its Clara AGX towards diagnosis via radiology, but has found its use in other medical applications, such as the study of the development of different tumors.
This does not mean that the Clara AGX ends up replacing health professionals, since NVIDIA itself is the one that makes it very clear that it is not a device for diagnosing diseases and a great tool for classifying information in form of millions of pieces of data that reach a hospital or any medical center every year. In the same way, it can also be used for other fields of scientific research that can benefit from the CUDA ecosystem.
