How Speedata accelerates data analytics (as much as its CEO will say)

Israel-based startup Speedata emerged from stealth nearly a year ago with an infusion of $70 million and is now running its first-generation accelerator chip on simulators and prototypes with multiple customers, some performing demanding analytics on big data.  

“We have seen extremely interesting workloads from pharma, which is rather new to us, while we originally looked at finance, and we have workloads from cloud, the travel industry and lately adtech,” said CEO Jonathan Friedmann in a Teams interview with Fierce Electronics on Tuesday from Israel. “It’s amazing how data is changing.”

Friedmann won’t divulge which companies are trying out the Speedata chip but added, “We have not seen any workload where we do not give 20x faster performance.”  He means 20 times faster than companies relying on the fastest currently-available CPUs and other processors from various vendors used to crunch numbers in data centers for unique insights and information.

He smiles as he describes the silicon Speedata designs, calling it an APU, for Analytical Processing Unit. That smile is mainly his reaction to an explosion in three letter acronyms for processors.  “There are so many extreme acronyms, just look at the names.  DPUs are used for multiple different things, while Intel’s IPU is Nvidia’s DPU. Everybody invents these names and they are very hard to follow even if you are in the industry. We accelerate what we call analytics, which is also an abused term.”

To be clear, what Speedata has invented is not for AI.  It is for processing structured and semi-structured data on databases if a company is trying to extract information from databases, “and is definitely not AI and not for non-structured data,” Friedmann said.

Friedmann believes the future for data centers will rely on dedicated processors like Speedata’s.  “There’s simply no way a general processor can continue to keep up. It’s not an option,” he said. Having said that, “CPUs are not going away and maybe will continue in the control domain where the CPU controls dedicated processors.”

The secret sauce that gives a 20x improvement is in Speedata’s architecture, even as the architecture itself if a secret. That means it can rely on hardware that is not currently the most advanced, such as a chip built on a 7 nm process node.  And because it can rely on 7nm, Speedata has skipped the supply chain concerns affecting 3nm and 5nm, Friedmann said.

“We’re lucky to have an architecture to give such a big boost,” he said.  “We have a unique architecture which gives us such a huge advantage and do not need to go into the extreme high end of 2nm, or 3 or even 5.  We can compete with a processor two generations behind the others.”

Speedata, made up of 70 engineers, is fabless, and relies on a fab based in the “Far East” to make its 7nm accelerators, Friedmann said, smiling again. “We’re a few quarters away from [full] manufacturing of the chip,” he said. “Hopefully we will have a lot of traction from big companies.”

Coming from a company based in Israel, Friedmann brings an unusual perspective to the recently approved U.S. CHIPS and Science Act, which boosts U.S. chip manufacturing with $52 billion but also plies funds into chip research.   “The CHIPS Act was not intended to respond to any immediate needs,” he noted. “Chips are the future and the major part of the U.S. and the world’s economy. It was important for the U.S. to take a significant role and the Act says the U.S. recognizes it is losing control and has to do something, especially with fabs. Chips are a marathon and creating a fab takes a few years.”

Speedata sees the fastest way it could take advantage of the Act’s infusion of cash would be to contract with Intel as it tries to open its fabs for external customers.

“There is definitely a shortage of fabs able to manufacture in high end processes,” Friedmann added “From that respect, any place in the world we would have fabs to compete on high end processes, the better.”

Even as hardware is a critical component of Speedata’s future, it is software that might be the ticket for i early success. Customers on the Apache Spark ecosystem that use the Speedata accelerator won’t need to make a change in existing software, Friedmann claimed.

 “They would continue to use Spark and won’t see changes,” he said. “We add our own plug-ins and libraries and you integrate our libraries.”

RELATED: Intel CEO Gelsinger at Hot Chips focuses on chiplets for speed