Habana Labs from Caesarea near Tel Aviv, announced that its Goya AI inference processor, successfully passed the PCI-SIG™ compliance testing at the Taipei, Taiwan workshop. As a result, the AI processor card and the HL-1000 chip, have been added to the PCI-SIG integrators list of add-in cards and Components lists for PCI Express 3.0 x16, operating at 8 Giga-transfers per second (GT/s).
PCI-SIG approval is a milestone for Habana Labs: The PCI-SIG workshop promotes compliance with real products to eliminate interoperability issues and ensure proper implementation of the PCIe specifications. This is the second major approval for Goya processors: Working with IBM, the Goya HL-100 x16 PCI Express card interoperability was also observed operating at PCIe 4.0 speeds on IBM POWER9-based IBM Power Systems, the only commercially available enterprise computing servers with the 16GT/s PCIe 4.0 interface.
The PCIe 4.0 technology operates at 16 GT/s or up to 32 GB/s (per direction), double that of PCIe 3.0-only devices. Goya has set two industry records, by delivering 15,012 images/second throughput with 1.3msec latency on the ResNet-50 benchmark, while simultaneously attaining an unmatched power efficiency record of 150 images/second/watt.
“The successful certification of Goya AI processor chip incorporating Synopsys’ low-latency, high-performance DesignWare IP for PCI Express 4.0 demonstrates the robustness of the IP and enables Habana Labs to accelerate their development effort,” said John Koeter, vice president of marketing for IP at Synopsys.
Out of Stealth Mode
Habana had been working for a year in stealth mode until September 2018, when it announced the sampling of its first AI processor to select customers. Designed to process various AI inferencing workloads such as image recognition, neural machine translation, sentiment analysis and recommendation systems, Goya platform has been designed from the ground up for deep learning inference. It incorporates a fully programmable Tensor Processing Core (TPC), development tools, libraries and a compiler.
A PCIe card based on its Goya HL-1000 processor delivers 15,000 images/second throughput on the ResNet-50 inference benchmark, with 1.3 milliseconds latency, while consuming only 100 watts of power. Habana Labs’ AI processors offer one to three orders of magnitude better performance than solutions commonly deployed in data centers today. The Goya silicon has been rigorously tested and is ready for production,” said Avigdor Willenz, Habana Labs’ lead investor and Chairman of the Board.
In November 2018, it has secured $75 million in series B funding, led by Intel Capital. The round is joined by WRV Capital, Bessemer Venture Partners, Battery Ventures and others, including existing investors and brought the total investments in the company to $120 million. “Among all AI semiconductor startups, Habana Labs is the first and the only one, which introduced a production-ready AI processor,” said Lip-Bu Tan, Founding Partner of WRV Capital.