With 90 percent of the world’s data generated in the past two years, researchers have access to more raw material than ever. Along with it comes an insatiable demand for enough computational muscle to actually do something with that data — analyze it, visualize it, build predictive models with it.
And not every scientist can afford to wait for free time on a shared supercomputer. For critical applications like medical research, it’s essential to shrink the time it takes to get research findings.
That’s how the University of Queensland came to build a supercomputer of its own, harnessing the power of NVIDIA GPUs to enable imaging-intensive scientific research in the fields of cancer biology, neuroscience and quantum physics.
Analysis: The Big Bottleneck for Big Data
If you give researchers terabytes of data, they’ll want a supercomputer to go with it.
Researchers at the University of Queensland installed last year a world-class lattice light-sheet microscope, which can generate a colossal 7TB of imaging data per day. This powerful machine allows for 4D imaging of live biological specimens, ranging from individual molecules to small organisms.
But analyzing the microscopy data was a slow, cumbersome process. The data comes with image noise, which must be extracted algorithmically and computationally by deconvolution, a process to clean up image distortions and sharpen blurry images.
The university researchers often had to rely on external high performance computing systems to speed up this analysis. And this required going through a competitive process to allocate computational resources.
To more efficiently analyze the big data coming from the microscopy facilities, the university’s Research Computing Centre decided to build Wiener, a supercomputer to reduce scientists’ time to discovery.
Named after Norbert Wiener, the mathematician who created a model for denoising images, the supercomputer was designed to expedite research across diverse subject areas.
Built in late 2017, the supercomputer uses cutting-edge Dell servers and harnesses the capabilities of powerful NVIDIA GPUs — three NVIDIA Tesla V100 accelerators for its visualisation nodes and another two for its compute and analysis nodes. That’s a whopping total of 204,3800 CUDA cores and 25,600 Tensor Cores, delivering more than four petaflops of processing power.
Wiener provides processing power for scientific instruments across the campus, while the two large visualisation nodes allow researchers to view their datasets in real time.
Using a mixture of deconvolution algorithms, machine learning and pattern recognition techniques, the supercomputer rapidly provides outputs of deconvolved, tagged and appropriately characterised data, giving researchers immediate feedback on the quality of data being collected and enabling faster interpretation of microscopy data.
Super-Powering Scientists Towards Results
So far, Wiener is being used by University of Queensland scientists focusing on cancer biology, infection and brain connectome analysis. Other early adopters include quantum chemistry, quantum physics and computational chemistry researchers.
“One bioinformatics researcher can now get results within minutes rather than weeks, thanks to Wiener,” said Professor David Abramson, who directs the university’s computing center. “This particular researcher is aiming to automate the analysis of histology slides of skin cancer biopsies as an assistive technology to pathologists.”
Since its installation, word of the supercomputer’s accessibility and speedy output has spread among the university’s research community. Though the original motivation for the supercomputer was deconvolution and microscopy data analysis, other university researchers interested in deep learning have signed on, using CUDA and OpenCL frameworks to leverage the power of GPUs.
The number of postgrads and professor-led research groups using Wiener is expected to grow from about 120 users currently to around 400 in a few months.