GPUs usher in a new era in government analytics
- By Ray Falcione
- Apr 26, 2019
Decades ago, graphics processing units were used mostly for rendering ninja fighters and Formula One racecars. Since their days in game systems, however, GPUs have experienced an amazing evolution in processing power. Today they sit at the very center of enterprise computing -- and in doing so, are ushering in a new era of capability and insight for government agencies.
The unique ability of these more advanced GPUs to handle artificial intelligence, machine learning and other high-performance tasks have made them a vital component of digital transformation. Many agency IT departments are discovering the power of GPUs to make better use of massive amounts of data. With the volume of data amassing at ever-increasing rates -- 40% annually by many accounts -- processing power is becoming essential to better public service and improved national security.
Currently, CPU processing power and performance is growing by 10-20% per year -- not nearly fast enough to keep pace with data generated from billions of active and passive data sources. GPU performance, on the other hand, is growing by an impressive 50% per year. In the latest TOP500 supercomputer rankings, more performance gains were added by GPUs than CPUs for the first time ever.
A typical CPU has 10 to 25 cores, but a GPU can have over 5,000, delivering mindboggling speed increases over 1,000 times faster than those of CPUs. With that kind of processing muscle, a small number of GPU servers can equal the performance of hundreds of their CPU counterparts in the typical data center, saving energy and space while creating a cooler, greener environment. Furthermore, the cost savings, increased performance and business application capabilities of GPUs over traditional CPU systems are dramatic.
Nvidia, world’s largest GPU maker, recently announced it will acquire Mellanox, making the value proposition to the federal government even more significant. Mellanox’s interconnect technology can connect vast numbers of GPU compute nodes over a network fabric to form a massive single compute engine. The right software capable of leveraging these GPU architectures will be a game changer.
GPUs, due to their unique parallel-processing architecture and ability to visualize massive amounts of data, are ideally suited to the era of big data. Today’s data analysts are tasked with interrogating volumes of information that would have been inconceivable just a few short years ago. GPU analytics make this task possible -- even providing zero-latency processing at scale. Such systems offer numerous advantages, including:
- Extreme ingestion speed. In benchmark testing, GPUs support the loading of millions of records per second -- a rate far exceeding traditional solutions -- using familiar JDBC, SQL and Kafka database languages.
- Elimination of manipulation workarounds including pre-indexing, down-sampling and pre-aggregation. GPU systems allow analysts to explore entire datasets, regardless of size.
- Unique rendering capabilities. Because GPUs were originally invented for graphics, they are ideal for visualization purposes. Drill down, pivoting and cross-filtering can all be done in real time, regardless of scale.
GPU analytics platforms can query and visualize multi-billion-row datasets in milliseconds (yes, that’s “billion” with a “B”). The speed and scalability of the GPU is producing dramatically new opportunities and capabilities for analysts and data scientists alike. Application examples include:
The opioid crisis. Fighting this epidemic effectively requires fresh insight and immediate, actionable information. Using the power of GPU analytics, agencies can now cross-reference public health and law enforcement records against data from prescribers, manufacturers and medical professionals in near real-time.
The results promise to be transformational. Deep dives by region, city, community, ZIP code and even by map polygons can instantly reveal bad actors who are prescribing or distributing opioids in damaging quantities. The results can not only aid in identifying current criminal activity, but also help shape future mitigation strategies.
Geointelligence. Combining disparate sources of information such as satellite and aircraft images, social media information and internet-of-things sensor data streams gives defense and homeland security analysts revolutionary levels of insight and awareness. For example, analysts can help military commanders identify patterns of life ahead of planned operations, reveal anomalies such as changes in traffic patterns or population movements and assess developing trends in a given region over time. The visualization of billions of points or polygons on a map is now possible in new ways with software that takes advantage of the power of the GPU architecture. Just a few years ago, it would been very hard and expensive to visualize even a million points or polygons.
The ability to interrogate entire datasets instead of samples or subsets provides special benefit to intelligence analysts and decision makers. In the past, an isolated sampling might not reveal the whole picture-- possibly dissuading the analyst from looking further. Today, a full analysis powered by GPUs can locate that valuable information in seconds.
Disaster assessment and relief. In any humanitarian crisis, real-time situational awareness is critical. GPU analytics can leverage unexpected sources of information to infer or determine situations that would otherwise go unidentified -- such as looking at Wi-Fi availability on a block-by-block basis to identify the impact of power outages.
During a disaster, one of the greatest factors in a successful operational response is logistics. Agencies must quickly cross-filter large and disparate datasets to see where bottlenecks exist. With GPU-based systems, operational analytics can be layered over geographic data to support deployment of food, medicine, fuel and other essentials. Decision-makers can improve deployment of personnel and equipment, optimize networks, more completely assess infrastructure, address maintenance needs and meet relief timelines.
In the 21st century, data is the new frontier, and we are the open-minded explorers. Today’s analysts seek insight, awareness and foresight -- ideals that can take us to new places and transform our collective understanding. Faced with a torrent of information, analysts look for patterns, anomalies, hidden details and sources of cause and effect. Most challenging of all, they must accomplish all this at the speed of thought.
Thanks to GPU-enabled analytics, we can now ask questions and find answers faster and with greater certainty and trust than ever before. As GPUs continue to change our thinking about computing, analysts are uncovering possibilities for knowledge like never before. For the curious mind, the discoveries are just beginning.
Ray Falcione is vice president of OmniSci.