AT&T’s wi-fi community connects Greater than 100 million subscribers from the Aleutian Islands to the Florida Keys, spawning An monumental knowledge sea.
Abhay Dabholkar runs a evaluation group that acts like a lighthouse Wanting for Definitely one of the biggest devices to navigate it.
“It’s nice, we get to play with new devices Which will make a distinction for AT&T’s day-to-day work, and As quickly as we give staff The latest and grealook at devices it provides to their job satisfaction,” said Dabholkar, a distinguished AI architect who’s been with The agency Greater than a decade.
Recently, the group look ated on GPU-powered servers the NVIDIA RAPIDS Accelerator for Apache Spark, Computer software that spreads work throughout nodes in a cluster.
It course ofed a month’s worth of mobile knowledge — 2.8 trillion rows Of information — in simply 5 hours. That’s 3.3x faster at 60 % decrease value than any prior look at.
A Wow Second
“It was a wow moment as a Outcome of on CPU clusters it takes Greater than 48 hours to course of simply seven days of knowledge — Prior to now, we had The information however couldn’t use it as a Outcome of it took such A very Very prolonged time to course of it,” he said.
Particularly, the look at benchmarked what’s referred to as ETL, the extract, rework And cargo course of that cleans up knowledge earlier than It Might be used to traInside the AI fashions that uncover current insights.
“Now we’re considering GPUs Might be utilized for ETL and all types of batch-course ofing workloads we do in Spark, so we’re exploring completely different …….
Source: https://blogs.nvidia.com/blog/2022/03/22/att-data-science-rapids/