Loading...

Scalable and Expandable AI stack

4U Rackmount GPU server and Expandable Storage System

About Our Customer

Alfred medical imaging department service two hospitals; Alfred health and Sandringham hospital. One of their visions is to develop and use technology to improve diagnostic imaging and reporting with Artificial Intelligence and Deep Learning.

Being the best imaging department in the country, they have received multiple accreditation awards including Full DIAS (Diagnostic Imaging Accreditation Scheme) compliance. These awards are a reflection of their community of diligent and dynamic staff, who are devoted to caring for patients’ needs. The services they provide include XRAY, CT scan, Dexa Scan, Mammography, Ultrasound, MRI, Fluoroscopy, and Angiograph.

In May 2019, Alfred medical imaging department reached out to DiGiCOR. The situation required a server that could store both large amounts of up to 100tb image data from Alfred Health and Sandringham Hospital as well as compute. Ultimately, they needed this system so that it could be used for Artificial Intelligence and Deep learning.



Highlights

  • The medical imaging department for Alfred Health and Sandringham Hospital wanted an off the cost reliable hardware that wouldn’t require lots of maintenance.
  • They required large-scale storage of up to 100TB to store hundred and thousands of CT scans for Artificial Intelligence and Deep Learning.
  • Their other requirements consisted of having a system that was expandable of up to eight GPUs.
  • Key Benefits

  • DiGiCOR provided a scalable and expandable GPU optimized system to accommodate to future growth
  • Provided a system that had both storage and compute located on the same system to enable Artificial Intelligence and Deep Learning use case
  • Connecting two physically distant imaging departments in two hospitals via scalable AI stack
  • NVIDIA Quadro RTX 8000 GPU’s provided significant memory for three-dimensional CT scansy
  • Challenges

    The medical imaging department for Alfred Health and Sandringham Hospital wanted an off the cost reliable hardware that wouldn’t require lots of maintenance. Previously, they were extracting and storing data on clinical system. They found that this was expensive, limited the amount of data they could store, and made it difficult to share information with other researchers. To create flexibility to share data with other researchers, they needed a separate hardware partitioned from clinical system.

    Furthermore, because one of their focal points was creating storage for research unit, they required large-scale storage of up to 100TB to store hundred and thousands of CT scans for Artificial Intelligence and Deep Learning. This was crucial as they needed Artificial Intelligence and Deep Learning to learn from these CT scans.

    This included Ubuntu stack and XNAT; which is a research PACS which stores images and organizes them in a library.

    Their other requirements consisted of having a system that was expandable of up to eight GPUs. Due to the current requirements, two or four slots would be immediately occupied and the remaining slots would allow simple expansion. In addition, because GPU technology is rapidly growing, they wanted to be able to upgrade when a new NVIDIA was released either having to be stuck with legacy hardware or, having to buy a new system. This meant that scalability was another important factor.



    How We Helped

    DiGiCOR provided a Supermicro server that was GPU optimized with up to eight GPU slots. This gave them the scalability and expandability requirements they needed. It is also a system, designed for High performance computing for 3D imagery which made it suitable for the medical imaging. The entire solution stack is enterprise grade with best in breed componentry.

    DiGiCOR also provided NVDIA Quadro RTX 8000 graphics card because their use case required significant amounts of GPU memory as CT scans are three-dimensional. Training time was not a critical requirement so the NVIDIA Quadro RTX 8000 was selected due to its large memory capacity and cost effectiveness.

    Handle workloads that would enable them to develop solutions to optimize lottery performance.




    Results

    The overall solution consisted of a scalable GPU server with support for up to 8x GPUs and direct attached high-performance storage via 44 bay disc expansion shelf.

    As the GPU server and storage was segmented into separate elements, greater flexibility and expandability was achieved. We also provided a system that was scalable to enable them to upgrade to newer GPUs, and expandable to allow them to use to up to eight graphics cards.

    In addition, we also provided the medical imaging department significant space of up to 100TB of storage via direct attached storage for medical imagery to be used for Artificial Intelligence and Deep learning for CNN, Image recognition, detection and classification.

    Currently, they have populated half the GPU unit, and 80% drive base. They were also able to meet their objectives and goals with this solution because they were able to store legacy CT and medical imaging on the system.

    In addition, they were also able to do server reach projects with this server to complete research projects. The medical department for Alfred Health and Sandringham hospital continue to be the best imaging department in the country as a result of they using technology to improve diagnostic imaging and reporting.



    Contact Us