Bench Talk

(Source: Iokanan Pro - stock.adobe.com)

High-end graphics play a critical role in both industrial and medical applications. From creating 3D models for industrial design and prototyping to using visualization and simulation for surgical planning, graphics are used to visualize and analyze data in new and innovative ways. With the integration of artificial intelligence (AI) and machine learning (ML), these graphics are becoming even more powerful and enabling more accurate decisions and improved outcomes.

When faced with complex design challenges that come along with medical and industrial computing, design engineers and managers have choices to make. They can bring all the expertise and know-how in-house to perform the entire design, prototype, and test, or they can use an original equipment manufacturer (OEM) and integrate technology from market leaders. Both approaches have merit depending on the cost and time-to-market constraints.

If manufacturing volumes are very high and time to market is not a critical constraint, then every penny counts, and in-house design and expertise will ultimately be the best choice. But high-end, low-volume applications benefit most by integrating leading-edge technology into products or designs.

The performance of high-end applications, such as mission-critical medical machines, industrial manufacturing, and harsh environments, can be impacted by real-time demands and the limitations of having a single central processor. To overcome this, partitioning a design and using pipelining techniques is often necessary.

The market offers a range of general-purpose and application-specific high-end processors and modules, many of which boast a mix of powerful peripherals and performance. Single and multicore reduced instruction set computing (RISC) and complex instruction set computing (CISC) processors provide high GHz speeds at lower power consumption. Additionally, mature development systems allow hardware and software engineers to work seamlessly together.

OEM’ed modules can offer smaller form factors due to refined designs. Size constraints are important, and the ability to fit high-performance functional blocks in small spaces is desirable. Form factors like MMX cards can typically be half the size of other standard process-intensive cards and fit in smaller and closed boxes. This is when functioning without a fan is essential.

Harsh environments like factory floors need fully enclosed form factors to protect from dust, moisture, combustible gasses, and other corrosives. A unit cannot be completely sealed if a fan is required since external airflow is needed to prevent everything inside the box from equilibrium heating. Fanless sealed enclosures can drastically increase reliability. Another reason industrial and medical designs benefit from modular integrations is that they carry higher price tags and are typically not ultra-high-volume products. This means they can shoulder the higher cost burden of integrating cutting-edge technologies like AI and machine learning. More important than cost is time to market. Coming out first with a working solution can make a difference when penetrating a new market.

Another factor is that OEM’ing blocks of technology means that documentation is in place. This saves time in creating quality documentation for end customers. In addition, application software may also be readily available. This can streamline the development and introduction of a new machine since much work can be cut and pasted instead of generated from scratch.

Any time you can offload a complex and real-time demand from your primary central processing unit (CPU), you have made your system more effective. And applying AI to Graphics Processor Units (GPUs) can allow a more reliable and faster image data analysis.

As such, GPU subsystems contain their own large memory pools and accelerated hardware that can quickly move object data, perform zooming, scrolling, and rotation without CPU intervention, and keep track of 3D image data for layered applications. For example, a medical scanner like a Computerized Tomography (CT) takes hundreds or thousands of 2D images that are digitally stitched together in 3D (Figure 1). A doctor can inspect, rotate, and pass through the top layers to reveal data inside. This task needs to be fast and in real-time. Dedicating high-end GPUs to this task unburdens system processors and allows higher throughput and performance overall.

Figure 1: Hundreds to thousands of 2D images are digitally stitched together in 3D for CT scans. (Source: LeArchitecto- stock.adobe.com)

Another high-end application is air traffic control. With thousands of planes in the air, all moving at different rates in different directions at different altitudes, safety and life are in balance. The graphics systems underlying a display terminal for air traffic controllers are critical in maintaining security in the air and on the ground. This is one place AI is making inroads.

AI combined with GPUs enables even faster response and accelerated learning. The parallel processing capabilities of GPUs permit multiple analyses in real-time. For example, an automotive driver-assist technology could analyze video data to detect the speed and direction of approaching vehicles and objects as part of a collision avoidance system. It could also perform bit plane separation of video frames to enhance unseen objects, potholes, and other typically invisible dangers. Large memory pools and extensive data sets aid deep learning algorithms in ‘remembering’ what is learned. Inscribing the learned behaviors into neural nets puts this new knowledge in the ‘consciousness’ of the AI processor, making it available more quickly in the future.

The combination of AI with high-end graphics performance has already demonstrated the ability of a machine to learn and discern patterns. And, as doctors get overwhelmed with new information, new drugs, and new data for drug interactions, AI can handle massive amounts of data to maintain patient safety and health.

Industrial manufacturing, too, can benefit from AI and GPU marriages. The ability to inspect manufactured goods with speed and precision that humans can't match means better quality in real-time because the AI can tweak manufacturing parameters and tighten the tolerances in statistic process controls.

Advantech leads the pack when it comes to high-end graphics and AI subsystems. As a leader in the IoT space, machine vision, industrial computing, and medical machines, Advantech provides the AIMB-768 9th generation ATX motherboards and SKY-MXM NVIDIA Quadro Modules to deliver exceptional performance and features. 

The SKY-MXM NVIDIA Quadro Modules can use T1000, A1000, or A2000 graphics cards with MXM 3.1. (Figure 2). Amazing amounts of parallel processing can be used thanks to up to 2560 CUDA cores, 30 RT cores, and 240 Tensor cores for up to 8.25 TFLOPS processing power.

These data crunchers and movers have 192-bit interfaces supporting 336GB/sec transfer speeds to their internal 6GBytes of GDDR6 memory. Scaled performance in the same form factor can also be realized. The SKY-MXM-T1000-4SDB System on a Module (SoM) features the T1000 processors with 4Gbytes of memory, while the SKY-MXM-A2000-4SDA modules use the A2000 processors and 4Gbytes of memory

Figure 2: Featuring the NVIDIA® Quadro® A2000 with MXM 3.1 TYPE A form factor (82 x 70 mm), the SKY-MXM-A2000-4SDA module is an ideal drop-in solution for graphic and computational-intensive designs. (Source: Mouser Electronics).

Guaranteeing a long life and support cycle, Advantech's proven and mature industrial computer (IPC) and AI edge technologies make ideal drop-in solutions for graphic and computational-intensive designs. In addition to graphics and processing systems, Advantech provides audio, display, communications, sensor management systems, power management, and more to become a one-stop shop for systems designers that understand that integrating high end proven and reliable subsystems allow companies to deliver cutting-edge solutions with rapid development times. Integrating fully functional and debugged modules is much easier than designing and debugging everything all at once.

The NVIDIA AI technology using DGX and A100 tensor cores allows engineering teams to take advantage of the hybrid architectures without bringing in-house high-priced experts. Many industry experts agree that the NVIDIA RTX4090 is the best GPU for deep learning and AI. Its superior performance makes it a perfect fit for next-generation neural networks.

Leveraging this, Advantech products can lead the way with graphic-intensive designs, speech recognition, security breach protections, pattern and image processing, and medical, robotic, and industrial automation. The more sophisticated building blocks mean more sophisticated end systems, and the quicker a design is ready to shine.

The integration of AI and high-end graphics plays a critical role in industrial and medical applications, providing advanced visualization and analysis capabilities for complex design challenges. AI combined with GPUs provides faster response times and improved performance, enabling real-time image data analysis and pattern recognition that can improve patient safety and health, as well as manufacturing quality.

Advantech is a leading provider of high-end graphics subsystems, offering a range of products including the AIMB-768 9th generation ATX motherboards and the Advantech SKY-MXM NVIDIA^® Quadro^® Modules give proven and mature IPC and AI edge technologies, Advantech provides a one-stop shop for systems designers, allowing companies to deliver cutting-edge solutions with rapid development times. Design engineers and managers are not alone when tasked with quickly incorporating state-of-the-art and cutting-edge technology. The integration of Advantech's products and Nvidia AI technology allows engineering teams to take advantage of hybrid architectures, streamlining the development process and reducing costs.