GPU-accelerated computing is one of the hottest trends among technology companies here in Silicon Valley. Companies including Google and Facebook are building out huge data centers leveraging parallel compute for machine learning, deep learning, and accelerated compute. Being able to share knowledge, ideas, and experiences is one of the great things about being in a tech hub, and to encourage this, we’ve created a GPU-Accelerated Computing Meetup group, which will run events to educate, engage, and inspire our community to push the boundaries of what is possible.
The inaugural Meetup was hosted by NVIDIA and attracted a packed room with more than 200 attendees. The presentations covered how GPU databases are enabling new opportunities across industries, helping to accelerate analytical processing, enhancing location-based analytics, and democratizing data science with advanced in-database analytics on the GPU.
For those who were unable to attend, here are some of the highlights from the presentation by Mark Brooks, Principal Systems Engineer for Kinetica:
Why Another Database?
Mark discussed the evolution of data processing and the need now to have to real-time processing. This has resulted in GPU-accelerated compute, where GPU cores bulk process tasks in parallel, which is a far more efficient approach for compute-intensive tasks than CPUs. Mark emphasized that Kinetica is not a “bolt-on” GPU solution, but rather a purpose-built GPU solution.
Democratizing Data Science
Mark demonstrated how you can write custom logic directly in Kinetica. UDFs can leverage CUDA libraries and the parallel processing power of the GPU, and BI/AI workloads can be brought together onto the same platform. This type of in-database processing opens the way for machine learning and data science workloads to be performed in-database and to take advantage of the GPU.
GPU-Accelerated In-Database Analytics Architecture.
Data science teams can develop and test gold-standard simulations and algorithms while making them directly available on the systems used by end users. In addition to querying data with traditional relational functionality, users could also call a Monte Carlo simulation or other custom algorithm straight from their BI dashboard. More on In-Database Analytics
VRAM Boost Mode
VRAM Boost Mode allows users to prioritize their data tables. Data sets can be forced to always sit in very fast, cluster-wide GPU Video RAM (VRAM) for lightning-fast, improved query performance. It also enables users to leverage cluster-wide system RAM to both scale up and scale out to multi-terabyte, in-memory processing.
Interactive Location-Based Analytics / Geospatial Capabilities
Kinetica combines native geospatial object types, specialized geospatial functions, and advanced visualizations, all while leveraging the power of the GPU.
Comparing Spark SQL to Kinetica
Mark discusses specific use cases that best fit each solution.
Our next Meetup will focus on advanced in-database analytics on the GPU and the democratization of data science. Join us on Tuesday, February 28th from 6:00pm-8:00pm at The Square Bar & Kitchen in San Francisco. Plenty of free food and drink will be available! Here is a link to the full description and registration: Advanced In-Database Analytics [GPU-Accelerated UDFs].
Join the Bay Area GPU-Accelerated Computing Meetup to stay up to speed with all of our upcoming meetups and webinars. Meet fellow technologists, exchange cool ideas, start new ventures, and achieve other goals you may be pursuing.
Want to give a lightning talk at a Meetup? A 10-minute presentation? Curious about new technology? Join the group today and reach out to me through the Meetup!