Wanguba Muriuki is a content crafter passionate about putting everything into writing. He is passionate about Blockchain and Traveling. He is also an experienced creative and technical writer. Everything and everyone has a story to tell. What better way to capture the real story than in words.
Intel announced that it had acquired Habana Labs in a clear indication that it aims to venture deep into the Artificial Intelligence chip market.
On December 16, Intel announced that it had acquired Habana Labs in a clear indication that it aims to venture deep into the AI chip market. Habana is an Israel-based developer of programmable AI and machine learning accelerators that are used for cloud datacenters. Analysts estimate that this deal is worth about $2 billion.
Intel further stated that it will enhance its AI strategy as Habana starts sampling its proprietary silicon to customers. Last November, Habana raised $75 million in venture capital but it will remain an independent business model that will continue to be led by its current team of managers.
Habana is known to offer two silicon products that target workloads in machine learning and artificial intelligence. They include the Goya AI Inference Processor and the Gaudi AI Training Processor. The latter is optimized for “hyperscale” environments and it is expected to power datacenters that offer up to 4 times the throughput compared to the systems set up with the equivalent number of graphics chips.
They operate at barely half the energy per chip which translates to 140 watts. In the case of the Goya processor that was introduced in June and it is currently commercially available, it delivers up to 3 times the AI inferencing performance as Nvidia Chips. In that case, throughput and latency are the primary considerations.
Gaudi is available in the form of a standard PCI-Express card or a mezzanine card that is fully compliant with the Open Compute Project accelerator module specifications. It comprises of one of the industry’s first on-die implementations of Remote Direct Memory Access over Ethernet (RDMA and RoCE) on an AI chip.
Goya is designed to enhance Intel’s in-house Nervana NNP-I, dubbed Springhill. It is based on a 10-nanometer Ice Lake processor. This processor will enable it to handle high workloads using minimal amounts of energy.
On the other hand, Gaudi will slot alongside Intel’s Nervana Neural Net L-1000 that is known as Spring Crest. Spring Crest is enhanced for image recognition. Also, its architecture is unique compared to other chips since it does not have a standard cache hierarchy. Moreover, its on-chip memory is managed directly by software.
Previously, Intel claimed that the NNP-T’s 24 compute clusters, 32GB of HBM2 stacks, and local SRAM deliver almost 10 times the AI training performance of competitors’ graphics cards. In the case of the software equation, Habana provides a development and execution environment known as SynapseAI.
SynapseAI has libraries and a JIT compiler that is well-designed to help customers deploy various solutions as AI workloads. It supports all of the basic machine learning and AI frameworks and the Open Neural Network Exchange format supported by IBM, Qualcomm, Microsoft, Arm, Huawei, AMD, and others.
The future of Intel is Artificial Intelligence. As it is evident from its books, the Santa Clara Company’s AI chip segments attained $3.5 billion in revenue this year. Moreover, it forecasts that the market opportunity will grow 30% yearly. Starting from $2.5 billion in 2017 to $10 billion by 2022. Considering all that, AI chip revenues rose from $1 billion a year in 2017.
Intel’s acquisition of Habana follows its acquisition of San Mateo-based Movidius in September 2016. In 2015, Intel acquired Altera, a field-programmable gate array (FPGA) manufacturer, and in 2016 it purchased Nervana. In August 2019, Intel acquired Vertex.ai which develops a platform-agnostic AI model suite.
Intel (INTC) closed at $57.70 in the latest trading session which marked a -0.16% move from the previous day. The markets will be looking to find some positivity from INTC as it nears its next earnings report date. The stock is projected to report earnings of $1.25 per share on that day representing a year-over-year decline of 2.34%.
The INTC stock started the year in the mid $40 levels and by April it had surged to the highs of $58.82. However, it failed to sustain the upward momentum. It dropped back to the lows of $43.5 at the start of June. The oscillation continued with the stock rising again to $53 in the last week of July.
One month later the stock was testing the $45 support but it has now bounced back once more. In the long term, the INTC stock shows some upside potential. However, the latest acquisition news has not affected the price of the shares significantly. All the daily and weekly charts show that it has maintained its gradual upside path.