From gaming to life-saving: how cloud GPUs are improving the medical industry

Who would’ve thought that GPUs, once exclusive to gaming, could one day be saving lives? Back in 2007-2008, teenage me certainly had no clue about it; I was convincing my parents to get me an NVIDIA GeForce 8800 GT so I could play the newest games.

But as time passed, these chips found their way into dozens of industries beyond gaming. One of their most recent applications is in healthcare, where cloud GPUs specifically are providing computational power that’s reshaping medicine.

Naturally curious as to how a GPU, once the beating heart of a gaming rig, could be used in healthcare, I started to dig deeper into the matter. Backed by Cybernews researchers, I investigated the most common use cases of cloud GPUs in the medical industry. What follows is that story, alongside our predictions about what the future of health tech could look like.

What is a GPU?

GPUs (Graphics Processing Units) are specialized pieces of hardware that accelerate image and animation rendering. Unlike CPUs, which have a few larger cores that excel at sequential workloads, GPUs have thousands of smaller cores that can handle multiple tasks simultaneously.

As such, these parallel processing beasts can handle everything from matrix math to complex simulations. And yet, they’re still most commonly used in gaming and 3D graphics to render pixels and create life-like reflections in the newest games.

Brief history of GPUs

The road to becoming the all-powerful processors we know today was a rough one. In the 1960s, these chips were first used in computers with cathode tube displays. In those early days, graphics were a mere afterthought.

It wouldn’t be until the 1970s and 1980s that dedicated hardware began to appear. Back then, arcade systems started introducing these components into their 2D gaming machines. And while they were simple at the time, these cards laid the groundwork for future advancements.

When the PC revolution of the 1990s happened, video cards evolved from basic framebuffers into full-blown 3D accelerator chips. In fact, before the end of the 20th century, NVIDIA and ATI would release cards that put rendering, transformation, clipping, and lighting onto a single chip.

During the 2000s, their capabilities would steadily increase year after year. Yet, the real breakthrough came in 2007 with NVIDIA’s CUDA platform. This essentially made GPUs capable of general-purpose computing for the very first time.

From there, these robust chips would find their way into scientific simulations and the analysis of massive datasets. They would also be used to power multiple machine learning and artificial intelligence algorithms.

Transitioning to the renting model and cloud GPUs

As the capabilities of these powerful pieces of hardware grew with the rise of ML and AI applications, their accessibility plummeted. GPUs that were once a couple of hundred dollars for top-shelf products quickly turned into massive investments worth thousands of dollars.

Suddenly, these chips were out of reach not just for individuals, but also for startups and smaller companies. Fortunately, the bright minds of our generation had a great idea: move from owning GPUs to renting them.

With this shift, hosting providers entered the scene and brought modern, high-end GPUs back within reach. They made it possible for anyone to easily rent top-tier GPUs, such as NVIDIA L4, L40S, and H100 NVL, which you’ll find in offerings like Liquid Web’s GPU servers.

Some hosting providers went a different route. Instead of letting you rent an entire GPU server for yourself, they place multiple users on a single GPU server to share resources. This made the rental model even more affordable, albeit at the cost of reduced performance and security.

How do cloud GPUs work?

To make a single GPU accessible to multiple users simultaneously, hosting providers rely on virtualization. This technology, otherwise known as hypervisors, allows the host to deploy scheduling and load balancing to assign GPU cores to different tasks.

When a user spins up a GPU-based virtual machine via APIs, they gain access to one or several physical GPUs housed by the provider. The cloud then allocates resources to the user and executes the requested tasks, such as processing medical scans. As a result, most cloud GPUs (and single-tenant GPU servers) are rented by the hour.

Cloud GPUs in the medical industry: key applications

GPUs have evolved far beyond displaying data and rendering visuals for gaming purposes. Today, these chips are in everything from AI, cybersecurity, and finance to medicine, where they’re literally saving lives through the following applications:

• Medical imaging and diagnostics. Modern-day scans, be it CT, MRI, or ultrasound, and their accompanying software, provide vast amounts of data. They create a detailed picture of what’s going on inside you. Not only do GPUs make processing said data much more efficient, but they also improve the image quality of these machines.
• Human genome sequencing. Despite consisting of just As, Gs, Cs, and Ts, our genetic code has billions of possible combinations, and sequencing it produces an enormous amount of data. GPUs reduce the processing time required for these workloads, allowing researchers to find out which parts of the genome are associated with which disorders.
• Drug discovery. Graphics processors can simulate molecular interactions and protein folding on a massive scale. This enables pharmaceutical companies to perform complex computations to create and optimize new compounds more efficiently. As such, GPUs can significantly improve the speed and success rate of drug development.
• Epidemiology. In 2020, NVIDIA put out a call for users to join their GPUs and help COVID-19 researchers. They would utilize their cards to simulate protein dynamics, thereby gaining a better understanding of the coronavirus. Today, these chips can also be used to simulate the spread of a disease, making them indispensable to epidemiologists.
• Personalized treatment. Physicians are also getting their work cut out for them, as GPUs help predict how a patient might react to a specific therapy. But beyond supporting tailored treatment plans, these powerful chips also offer incredibly accurate diagnoses. As such, they empower healthcare experts to spot even the earliest signs of disease.

Benefits of using cloud-enabled GPUs in healthcare

While traditional computer hardware can certainly handle medical imaging or drug and epidemiology simulations, it’s not particularly efficient at these tasks. GPUs, on the other hand, can handle these workloads much more easily, so they offer the following benefits over CPU-based machines:

• Faster medical image processing. By providing access to multiple graphics chips with thousands of parallel processing cores, GPUs drastically reduce the time needed to create a complex 3D scan of your body. In fact, going from CPU to GPU power often means handling these reconstructions in seconds rather than hours.
• Improved image resolution and clarity. Besides speeding up medical imaging, GPUs can also enhance the output quality of these machines. That’s because they can simultaneously run advanced noise-reduction algorithms to increase a CT or an MRI scan’s resolution and make these images much clearer.
• Seamless load scalability. With cloud GPUs, hospitals and medical experts can enjoy incredible flexibility. In fact, these healthcare providers can easily scale GPU resources up to handle higher demand during emergencies or down when the need for medical imaging or protein structure predictions decreases.

Challenges of relying on cloud GPUs in the medical industry

It should be pretty clear how powerful cloud GPUs are and how advantageous they can be for both hospitals and medical researchers. Yet, as beneficial as they are, cloud GPUs are, unfortunately, not without drawbacks, with the most common ones being:

• Compliance considerations. Uploading sensitive private info to a public or poorly secured cloud is a quick way to violate HIPAA. Hosting providers need to be audited and certified for HITECH and HIPAA compliance readiness, and many are not.
• Increased latency and inconsistent performance. While latency will always be an issue with GPU servers, cloud environments exacerbate the problem due to shared infrastructure and the overhead of virtualization. Plus, you can lose up to 25% of GPU performance with a cloud setup, which will slow down your medical imaging.
• Data privacy and security risks. Healthcare professionals handle extremely sensitive data. On a cloud server, each tenant poses a security risk that could lead to sensitive data falling into the wrong hands. While that’s not the case with reputable providers, it’s still something to have in mind.

Single-tenant GPU servers: a powerful alternative to cloud GPUs

Although the cloud environment can be somewhat problematic, there is, however, an alternative to cloud GPUs – single-tenant GPU servers. These hosting options, such as those offered by Liquid Web, provide all the benefits of GPU-powered medical imaging, diagnostics, and simulation, without the disadvantages of the cloud.

They offer significantly lower and more predictable latency compared to cloud GPUs. Their performance is far greater and more consistent as well, since there are no hypervisors to worry about here. As a result, you get access to 100% of the GPU’s power.

Single-tenant GPU servers also offer better security than cloud GPUs. As there are no other users on the server besides you, the likelihood of sensitive data being leaked – or of server neighbors introducing security risks – is minimal. Plus, reputable providers like Liquid Web offer HIPAA-compliant hosting, which makes these servers an ideal choice for all things healthcare-related.

Our take: what does the future hold for GPUs in healthcare?

Due to their sheer parallel processing capabilities, GPUs have already become indispensable in the healthcare industry. They’re used to accelerate medical imaging and genome sequencing, as well as to simulate various systems, such as molecular interactions and the spread of viruses. Yet, we’re still very far away from seeing their full potential.

Looking ahead, healthcare providers may also start combining GPU power with AI. This could lead to more precise diagnostics and personalized therapies. In any case, the future of GPUs in healthcare is looking bright. I, for one, can’t wait to see what becomes possible soon.

FAQ