This week at the GPU Technology Conference (GTC), fastdata.io is introducing FDIO Engine, the company’s GPU-native high performance computing software for real-time stream processing.
The Santa Monica, California-based startup also used the opportunity at GTC to announce it has raised $5 million in seed funding from CrossCut and Pelion Venture Partners, along with its original investor, NVIDIA GPU Ventures. According to CrossCut’s Brian Garrett, “fastdata.io is poised to transform the big data and streaming analytics space, utilizing GPUs to offer a level of performance and efficiency never seen before in data processing.” Ben Lambert of Pelion Venture Partners said the company will “create endless innovative opportunities for businesses in industries that need real-time stream processing like finance, security, ad-tech, telecom, cloud, AI, Internet of Things, autonomous vehicles, and robotics.”
CEO and founder of fastdata.io, Alen Capalik, believes his company is well-poised to take advantage of changing landscape of data analytics, which is being reshaped by increasingly large data volumes, the need for interactive response times, and the emergence of the Internet of Things (IoT). “This paradigm shift, from traditional batch processing where you must store data before processing to true real-time processing where you can process data before deciding to store it, will open exciting new possibilities for application development in numerous industries,” said Capalik, whose background spans work in high frequency trading and cybersecurity.
That kind of enthusiasm is based on the potential of FDIO Engine to upend the stream analytics space, which today is dominated by CPU-based platforms like Spark Stream, Flink and other Apache spinoffs in the open source space, as well as IBM, SAP, Oracle, and DataTorrent, among others, in the commercial arena.
In a nutshell, the FDIO Engine leverages the parallel computing capabilities of the GPU to accelerate real-time processing of streaming data. How exactly it does this has not been fully elaborated, but apparently the software takes advantage of the GPU Data Frame (GDF) standard to make the process more efficient.
GDF was developed by the GPU Open Analytics Initiative (GOAI), an industry group founded by OmniSci (formerly MapD), H2O.ai and Anaconda. The format is based on Apache Arrow and enables the data to reside entirely on the GPU side, avoiding costly copies from host memory. It also reduces the number of local data copies when GPUs are talking among themselves.
The company claims its FDIO Engine can run NVIDIA GPU-accelerated Spark workloads 966 times faster than CPU-powered Apache Spark. As a result, fastdata.io says processing costs can be reduced by more than 70 percent and power and space requirements by more than 98 percent.
Not surprisingly, NVIDIA has been a big booster of the company. Jeff Herbst, vice president of business development at NVIDIA, said they quickly recognized the value of using GPUs to accelerate stream analytics, which precipitated their initial investment in the company last year. “We are huge supporters of both GDF and the fastdata.io business plan and see them essentially ‘paving the freeway’ for faster, highly efficient data communication between GPU-based software solutions,” said Herbst.
Fastdata.io is demonstrating their software this week during GTC at the San Jose McEnery Convention Center.