When CUDA specialist Eri Rubin prefaced his talk at NVIDIA’s recent Tel Aviv GPU event by apologizing for a tendency to dive deep, he wasn’t kidding.
In addition to being one of Israel’s leading programmers in CUDA – which helps graphics processors handle general computing tasks – he’s the reigning national champion at free diving. He won the crown last month by staying under water – without breathing – for just over six minutes.
To prepare, the lean, tightly cropped former Army staff sergeant made regular training runs down to the Red Sea resort of Eilat, where he’d join friends in diving 40 meters – about 130 feet – off a secured inner tube, using just his lungs, while tracking wily octopi and shy eels.
Whether submerged in the calm, clear cool sea – his technology limited to a heart monitor and depth gauge – or training in a neighborhood pool, it’s a sharp contrast with his day job as head of development at SagivTech, a software firm in the Tel Aviv suburbs specializing in GPU computing and algorithms.
“The water has a magical effect, it clears out your head, especially in the evening after work,” said Rubin, 38, who was formerly a computer animation specialist, working on full-length films. “The contact of water immediately drops your blood pressure and pulse.”
Among other tasks, Rubin manages SagivTech’s involvement in an ambitious project funded by the European Union’s future emerging technologies program. Its objective is to enable real-time 3D streaming on social media of crowd-sourced videos from events like rock concerts.
The effort, undertaken with the Swiss Federal Institute of Technology and Germany’s University of Bremen, starts with collecting hundreds of amateur video feeds of a single event. Next comes cleaning up the heavily compromised content, which is gathered on shaky, handheld smartphones and tablets, with their smudged lenses and tinny recording capabilities. The separate clips are then stitched into a 3D video stream that can be shared on social networks at 30 frames per second.
That’s a staggering challenge even in the world of parallel processing. Just a few months into the project, it now takes five minutes for a powerful GPU to reconstruct a single frame of content. That means there’s a long, long way to go.
But Rubin’s joined on the project by several other CUDA and computer-vision experts at SagivTech, who are devoted largely full time to the task. One of the country’s top employers of CUDA programmers, SagivTech is run by Chen Sagiv, an energetic Ph.D. in applied math, and her husband, Nizan, a naval officer turned management exec.
In addition to the EU contract, the firm is engaged in projects ranging from real-time processing of complex algorithms on multi-GPU clusters to running parallel-processing courses for Israel’s booming high-tech sector.
Rubin comes to the work naturally. He began programming at the age of eight on an eight-bit Commodore VIC-20 home computer and is finishing a master’s degree in computer science from Hebrew University.
When he isn’t doing that … or spreading the good word on the parallel power of GPUs … or commuting an hour each way to his home on a traditional Israeli kibbutz … or honing his free diving talents, he spends his downtime on a task that’s decidedly linear in its lack of complexity.
He likes to go spearfishing for grouper in the Mediterranean.
“It’s very simple, very straight forward and doesn’t require much planning,” he said. “And you come back with lunch.”