Intel has spent another $2bn to build out its AI processor portfolio, with the purchase of Israel-based startup Habana Labs.
Habana has two processor lines. Its Goya AI Inference processor has been “commercially available” since last summer. The company claimed 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”.
Its Gaudi training processor is sampling with selected “hyper scale” customers, and apparently delivers “an increase in throughput of up to four times over systems built with equivalent number GPUs”.
Both support mainstream AI and machine learning frameworks, such as PyTorch, TensorFlow.
The deal shouldn’t be a terrible shock to the company’s workers – the company’s VC arm Intel Capital was an early investor in Habana. Intel is clearly taking a relatively long view on AI – in 2016 it snapped up Nervana Systems for $350m as well as computer vision outfit Movidius. Other buys that feed into either AI or acceleration include $15bn for self-driving car sensor firm Mobileye and $16.7bn for FPGA firm Altera in 2015.
With that amount of cash spent, some might thinks Intel’s predictions for the AI silicon market are rather modest. It expects $3.5bn of AI-driven revenue this year, and is banking on the overall AI market to hit $25bn by 2024.
The new buy will remain an independent unit, and report into Intel’s Data Platforms Group which is headed by Navin Shenoy. Presumably this means it will operate separately to Intel’s Artificial Intelligence Platforms Group which is headed by Naveen Rao, former Nervana Systems CEO.
Shenoy said in a canned statement, “Habana turbo-charges our AI offerings for the data center with a high-performance training processor family and a standards-based programming environment to address evolving AI workloads.”
“We know that customers are looking for ease of programmability with purpose-built AI solutions, as well as superior, scalable performance on a wide variety of workloads and neural network topologies. That’s why we’re thrilled to have an AI team of Habana’s caliber with a proven track record of execution joining Intel. Our combined IP and expertise will deliver unmatched computing performance and efficiency for AI workloads in the data center.”