Studies have shown that organizations that use data and analytics in higher concentration have superior business operations. Every minute, data is being generated and recorded across divisions. Understanding this data can unlock tremendous competitive advantage, and accelerate breakthrough for organizations. But time is of essence. Even a millisecond’s delay can cost a lot.
NeuroBlade is set to take big data analytics to the next level. It’s SQL Processing Unit (SPU) opens doors to exploring data faster. At the recent Cloud Field Day event, they revealed the processor and demonstrated how it powers high-throughput analytics workloads.
Computation Demand is Peaking for Data Analytics Workloads
There is a torrent of information that hides solutions to the thorniest business problems. But it brings little benefit if it is not turned into intelligence. Turning information into insights at mindboggling speeds is where the real advantage is.
“Organizations are collecting more data and they’re finding new business opportunities and ways to improve their service and operational costs” noted Mordechai Blaunstein, VP of Marketing at NeuroBlade.
As companies ramp up their AI workloads, they are increasingly turning to more powerful processors for greater compute power. The market offers a bottomless supply of hardware to choose from. But crunching data fast alone won’t cut it. Aside from raw processing power, analytics workloads require processing at massive scale.
“The scale is much larger today. You have to be able to compute the data to make it usable. Otherwise, it’s just spending money on storing the data,” he pointed out.
Companies have rushed to bolster their hardware infrastructure accruing more and more specialized components to meet the growing demand. Analytics tools with self-service capabilities are stacked to add the extra layer of intelligence to the data. In a way, it has worked to break down stats and numbers into actionable insights, but analytics has still not penetrated businesses as much as it could.
It takes colossal funds to scale the infrastructure to where it can handle the growing volume of data analytics workloads. Alternative software-oriented approaches have hit a wall due to hardware bottlenecks.
“We observed such trend and we tried to come up with a solution that unlocks performance without blowing up expenses,” said Blaunstein.
NeuroBlade SPU for Demanding Analytics Workloads
NeuroBlade has designed a class of processors that can radically reduce the processing time and cost of data analytics. Traditional hardware built with CPUs – and more recently GPUs – are costly to scale. Although highly powerful, GPUs have a long waiting line, are not meant for cloud with millions of instances running data analytics workload, said Blaunstein.
Growing data variety is another stumbling block. Any enterprise adopting AI has to deal with three kinds of data at least – structured, semi-structured and unstructured data. “We see massive models that require large amounts of compute in the ecosystem.”
Custom-made for modern data analytics infrastructures, The NeuroBlade SPU offers 10 times more speed than GPUs, and can handle large volumes of diverse datasets and formats. It leverages parallel processing for optimum query performance. SQL operations are split into smaller tasks and distributed across the entire cluster containing multiple enclosures loaded with SPUs. This allows tasks to be executed concurrently resulting in faster query performance.
“Think about it like a CPU that can do data analytic work or SQL query work much faster. It’s like the CPU will have an instruction set that runs SQL operation faster,” he explained.
The architecture is designed to minimize data movement towards the CPU. Much of the processing happens inside the SPU helping bypass the traditional compute bottlenecks, and accelerate the query flow.
Faster processing helps shorten the ETL process, and that can reduce the AI training time, said Blaunstein.
NeuroBlade SPU is cross-platform compatible, and integrates effortlessly with existing compute and storage frameworks. It can be installed on computer servers as a PCIe card that users can stick in directly without making any hardware modifications. It blends just as easily with the software ecosystem. Users are provided an open set of APIs to integrate the accelerator seamlessly with existing query engines and data analytics tools. It outputs results in native formats.
Wrapping Up
As time-to-insight gets increasingly critical in decision-making in businesses, there is growing need for a new way to compute that can enable companies to gain exclusive information in real-time. NeuroBlade SPU provides that infrastructure advantage, making it possible to run the most demanding data analytics workloads at reasonable costs.
For more information, be sure to check out NeuroBlade’s technical presentations from the recent Cloud Field Day event.
