Fujifilm to Adopt World's Fastest Deep Learning Supercomputer

Fujifilm develops unique parallel processing method
for rapid distributed processing for problem-solving across multiple GPUs
- further accelerating AI development

September 13, 2018

FUJIFILM Corporation (President: Kenji Sukeno) will adopt, in October 2018, the NVIDIA® DGX-2™ AI supercomputer to power deep learning with world-leading performance*. The systems from NVIDIA will be deployed by Fujifilm to accelerate the development of artificial intelligence (AI) for use in various fields including healthcare and highly functional materials. The adoption of these new systems will greatly reduce the time required for complex deep learning training, accelerating the speed of the development of AI technology, and supporting the development of various new products and services.

Fujifilm's operations actively use AI in order to meet various needs and resolve a diverse range of issues in markets worldwide in the healthcare business field including the medical systems business - which provides various medical devices and services; the pharmaceuticals, bio CDMO**, and regenerative medicine businesses; as well as highly functional materials business, including materials for displays and fine chemicals. For example, Fujifilm photobook service makes use of its proprietary AI-enabled image analysis technology to automatically select the best images from a user's large stockpile of digital images and generate a photo book. Fujifilm also provides the AI-based image diagnosis service of inspecting infrastructure for cracks to greatly improve the efficiency of the process, under the increasing need to confirm the safety of aging infrastructure such as bridges and tunnels.

Fujifilm has developed a proprietary method of high-speed parallel processing which distributes the processing required for single calculation among a multiple number of NVIDIA graphics processing units (GPUs). In recent years, advancements in medical image diagnosis technology, including the greater use of multiple CT scans, have led to huge growth in the number of diagnostic images, which have in turn increased the workload of physicians.

To address this issue, Fujifilm is working hard to harness high-speed processing to greatly reduce the time required for learning three-dimensional image diagnosis, processes that were previously extremely time-consuming, as well as for direct learning from large volumes of three-dimensional image data. Through these processing developments, Fujifilm is investing heavily in the development of AI technology that will support image diagnosis systems capable of identifying early stage medical issues that could not be detected with previous technology as well as high precision measurement and quantification of a patient's current state.

The NVIDIA DGX-2 is the world's most powerful AI system in a single computer and capable of carrying out 2,000 trillion calculations every second, greatly reducing the calculation time required for deep learning***. The system is also equipped with a total memory capacity of 512GB, enabling rapid learning from a large volume of medical image and video data. Fujifilm's proprietary high-speed parallel processing technology allows the effective use of the computing power composed of several interconnected NVIDIA DGX-2 systems, facilitating the development of high-performance AI applications.

Fujifilm is committed to the development of AI applications in a wide variety of fields and will harnesses the technology to develop innovative products and solutions to resolve social issues.

  • * As of September 12, 2018
  • ** CDMO is an abbreviation of Contract Development & Manufacturing Organization. CDMOs are responsible for developing and manufacturing pharmaceuticals under contract for third parties. They provide a wide range of services to pharmaceutical companies at all stages of the drug development process including the transition, in the initial stages of pharmaceutical development, from development work on cellular strains to process development, as well as the implementation of safety trials, drug development and manufacture of commercial products.
  • *** The computational speed of the computer is 12 times higher than that of NVIDIA's previous generation of computers.
<Specifications for NVIDIA DGX-2>
GPU 16X NVIDIA® Tesla V100 GPU
Memory Total: 512GB
Performance 2 petaFLOPS
CPU Dual Intel Xeon Platinum 8168, 2.7GHz, 24 core
System memory 1.5TB
Storage OS 2X 960GB NVME SSD
Internal storage 30TB (8x 3.84TB) NVME SSD

 

For inquiries on information in this media release, contact:

Media Contact:
FUJIFILM Corporation
Corporate Communications Division
TEL: +81-3-6271-2000