Adopting GPU Cloud to Improve Medical AI Performance

Promedius

Promedius is a medical imaging AI startup that developed the world’s first AI solution for diagnosing osteoporosis based on Chest X-rays. Recently, Promedius adopted Elice Cloud to improve product performance through foundation model training. In this interview, we spoke with Jinhoon Jeong, Head of the Research Team, to learn more about the background and experiences of using Elice Cloud.

Q. Hello. Please introduce yourself and tell us a little bit about Promedius.

Hello. I’m Jinhoon Jung, Head of the Research Team at Promedius. Promedius is a medical imaging AI startup that spun off from Asan Medical Center’s research lab in 2019 and has been continuously innovating with AI technology that supports fast and accurate diagnosis in the medical field. As the Head of the Research Team, I am responsible for all research activities in the company, determining research directions and allocating resources. Our solutions are particularly strong in processing and analyzing medical data efficiently, which helps medical professionals make faster and more accurate decisions.

Q. What was the background for adopting Elice Cloud?

Just as OpenAI’s ChatGPT garnered significant attention, the paper on RayDINO has also become a hot topic in the field of medical imaging. This paper built a foundation model using Chest X-ray data, inspiring us to establish a plan to train our own foundation model in a similar manner. Since training a foundation model requires a large amount of GPU resources, we reviewed several GPU cloud service options and chose Elice Cloud for its performance and efficiency.

► Excerpt from RayDINO paper (Promedius successfully replicated this experiment with Elice Cloud)

Q. How did you hear about Elice Cloud? What was your first impression of it?

We came to know Elice Cloud while forming a consortium to apply for the semicondutor sector of the AI Voucher program. I didn’t have any particular expectations at first, but I was impressed to see that there is such a high-performance GPU cloud service in Korea. In particular, I felt that Elice Cloud well reflects the specialized needs of domestic companies.

Q. Why did you choose Elice over other GPU providers?

I was naturally interested in Elice Cloud while working together on an AI voucher project. I was also impressed by how friendly and responsive they were during the quotation process. We asked a variety of questions as we were new to large-scale GPU cloud services, and they answered all of our inquiries with sincerity and care. They also offered flexible conditions in terms of cost and supported us in preparing the PoC (Proof of Concept) for a sufficient amount of time, which led us to choose Elice Cloud.

Q. What are you using Elice Cloud for, and what do you hope to accomplish with GPUs?

As mentioned earlier, we aim to train a foundation model using Chest X-ray data to further improve the performance of our products. Since most of our major AI products are based on Chest X-rays, we expect the foundation model we are building to have strong synergies with our existing models. This will improve the accuracy of diagnosis and further maximize its use in the medical field.

Q. Did you feel comfortable with the process of adopting Elice Cloud? Please tell us about any inconveniences you encountered.

Since it was our first time renting and using bare-metal servers, the initial setup took some time. However, the Elice Cloud team provided us with a long enough PoC (Proof of Concept period to smooth this out. This gave us the time to adapt to the new environment, and we’ve been using the service without any issues since.

Q. What benefits have you experienced with Elice Cloud?

The biggest advantage was being able to rent the latest GPUs such as the H100 at an afforadable price. In addition, the representatives provided us with quick and continuous feedback whenever we encountered issues during use, which allowed us to achieve optimal performance even with limited budget and resources. Thanks to this attentive support, we were able to efficiently utilize GPU resources.

Q. How has Elice Cloud helped Promedius?

The foundation model we trained using the GPU resources of Elice Cloud is now playing an important role in analyzing our results and further research. This model is being utilized in conjunction with our various Chest X-ray-based studies, significantly improving the speed and efficiency of our research. We expect this model to serve as an important foundation for the development of various AI solutions in the future.

Q. Can you share any achievements you’ve had with Elice Cloud?

Through Elice Cloud, we were able to successfully build our own foundation model and conduct various research based on it, which has greatly contributed to strengthening our research capabilities and increasing the value of our company. We also acquired valuable know-how through experience in utilizing large-scale GPUs, which laid the foundation for achieving results more efficiently and quickly in similar research endeavors in the future.

Q. Tell us what you particularly liked about Elice Cloud (e.g. features, pricing, etc.).

First of all, I was very satisfied with the ability to utilize as many as 32 of the latest H100 GPUs. It was also very cost-effective due to the lowest pricing policy in Korea. Most importantly, I appreciated the pre- and post-sales support, which was very kind and sincere.

Q. How satisfied are you with Elice Cloud overall?

Overall, I am very satisfied. If I had to point out a downside, I found the technical support for setting up Slurm to be somewhat lacking, especially since Slrurm is mainly used in academia for resource management and scheduling in HPC environments. Other than that, I am pleased with the results, which have exceeded my expectations.

Q. Would you recommend Elice Cloud to other organizations? If so, why?

Although the GPU performance offered by each cloud service is actually similar, I think Elice Cloud stands out for its competitive pricing, allowing users to access GPU resources at a reasonable cost. In addition, I highly recommend Elice Cloud for its generous PoC period and the friendly and prompt pre- and post-sales support.

Q. Can you tell us a bit about the future direction of Promedius?

Promedius is currently conducting domestic clinical trials for its osteoporosis classification product using Chest X-rays (CXR), and we aim to enter the global market next year through FDA and CE certification. Along with this, we plan to develop various CXR-based AI products and to grow as a leading company in the medical AI field.

Elice Cloud has provided meticulous pre- and post-sales support to help Promedius quickly adapt to the initial environment while actively assisting us strengthen our research capabilities. We hope that Elice Cloud will continue to serve as a solid foundation for Promedius’s research and development efforts, as well as its global market entry.

Elice Cloud provides the latest GPU resources at a reasonable cost to support your research and development. If you need GPUs optimized for research, consider adopting Elice Cloud.

👉 Learn more about Elice Cloud On-Demand

👉 Get an Elice Cloud GPU service consultation

*Elice owns the copyright to this content, which is protected under copyright law.

*Without prior consent, secondary processing and commercial use of the content are prohibited.