Artificial intelligence opportunity
Intel (INTC) is expanding its data center offerings beyond server CPU (central processing unit) solutions to complete rack solutions. The company is shifting its offerings along with the changing needs of today’s data center space.
As we move toward the data-centric world, the need for real-time analytics is growing, driving demand for AI (artificial intelligence). Right now, AI is an emerging market that accounts for less than 10% of the data center workload. However, it’s a quickly growing segment, and NVIDIA (NVDA) is at the forefront of this growth. The company’s data center revenue rose 145% YoY (year-over-year) in 2016 as more and more data centers adopted its GPUs (graphics processing unit) for their AI workloads.
Competition in the AI space
Intel is looking to tap the AI market, but it faces strong competition from NVIDIA’s GPUs and Google’s (GOOG) TPUs (tensor processing unit), which are general-purpose chips designed specifically for AI workloads.
In order to accelerate its AI product development, Intel formed the Artificial Intelligence Product Group in March 2017. The group combines all its AI offerings and engineering expertise under one roof.
Intel’s AI offerings
Intel offers a full spectrum of AI products, from edge computing to network computing to data center computing. Its Mobileye (MBLY) autonomous car solution would offer 1 TFLOPS (tera floating-point operations per second) per watt of computing performance. In fact, Intel is looking to create a solution that will offer a performance of more than 1 TFLOPS per watt.
In the lower milliwatt range, Intel offers Movidius AI solutions for robots and drones. In the high-power range, Intel offers Nirvana’s ASIC (application specific integrated circuit), which delivers a higher performance than a GPU, for the neural network. Intel also offers the Xeon Phi processor, integrated with Altera FPGAs (field programmable gate arrays), to handle data centers’ AI workloads.
All the above AI technologies were acquired, not developed, by Intel.
Intel is also offering FPGAs as individual products so users can connect them to their existing server CPUs. One accelerator that Intel doesn’t offer is a GPU, the most widely used accelerator. Intel’s basic integrated GPUs are no match for NVIDIA’s discrete GPUs. Recently, there was a job listing for a graphics and media architecture lead on Intel’s website, indicating that the company is looking to develop GPU technology for AI applications.
Even if it succeeds in developing a strong graphics architecture, Intel could face challenges in the software space. Most GPU programs are written for NVIDIA’s CUDA platform, which has evolved over the years.
Next, we’ll take a look at Intel’s FPGA business.