Intel's (NASDAQ:INTC) data center dominance seems to be on shaky ground as there is a paradigm shift in the technology used in cloud computing servers. The huge amount of data generated by fast-growing tech trends such as the Internet of Things, smart homes, and autonomous cars requires fast computation for real-time decision making, which is bad news for Intel's data center business.
GPUs are gaining ground
More and more data centers are now using graphics processing units (GPUs) in their servers. GPUs have an inherent advantage over the central processing units (CPUs) that Intel specializes in as they can significantly boost computing capacity. NVIDIA, for instance, claims that its GPU accelerators can reduce costs by 60% and increase server computing power by five times by replacing CPUs.
This is because a GPU has thousands of cores that allow it to take on a huge amount of workload simultaneously. By comparison, AMD's latest EPYC server chip has just 32 cores at the most, while Intel's top of the line chips come with 28 cores. This is a cause for concern for Intel as the rise of GPUs could cripple its data center business, which currently supplies almost 30% of total revenue.
In fact, Intel's data center business growth seems to be moderating after a spike last year.
But Intel is determined not to let the data center opportunity slip away. It is planning to introduce a hybrid chip that could help it thwart the rising competition in the server chip market.
Intel turns to FPGAs
Intel acquired programmable chip manufacturer Altera back in 2015 for a handsome consideration of $16.7 billion. Altera specialized in making field-programmable gate arrays (FPGAs), which can be reprogrammed to serve any purpose after they are manufactured. This makes them different from application-specific chips that are custom-made to perform specific tasks.
Intel is now planning to program these FPGA chips to accelerate the performance of its x86 server CPUs. Chipzilla believes that the customizable and versatile nature of the FPGA chips makes them better at certain tasks when compared to GPUs.
Specifically, GPUs are limited to offload processing in data centers, which means that the data has to pass through the CPU before it moves to the GPU. FPGAs, meanwhile, can be programmed to perform inline processing so that the data flows straight to the FPGA for processing. As a result, FPGAs are capable of boosting server performance as compared to a stand-alone CPU, so Intel has decided to pack the two chips into a single platform.
The company's hybrid CPU-FPGA processors can perform with greater efficiency as the FPGA component can be programmed to take on different workloads as per requirement. Therefore, an integrated system-on-a-chip will be capable of running parallel workloads just like a GPU, enabling the hybrid chip to run applications related to machine learning, artificial intelligence, and financial analysis, among others.
Intel is planning to offer discrete FPGA chips that are expected to be available in the first half of 2018, followed by the hybrid chip that are expected in the second half. Therefore, Intel is trying to cut a bigger share of the FPGA pie for itself, which is not surprising as this market could be worth $13 billion by 2024.
The telecommunications market is expected to drive a substantial portion of this growth as more FPGAs are deployed in cloud computing servers. Intel has an inherent advantage in the server space as it has a monopoly like position, which makes it easier to push its FPGA chips through existing sales channels.
Additionally, the addition of FPGA could help Intel steal a march over AMD's EPYC server chip, which is reportedly outperforming Intel's current crop of server processors. Therefore, the FPGA push will catalyze Intel's data center business given the gains that it can deliver over traditional server chips and GPUs.