Intel (INTC) held its datacenter day on April 2nd, 2019 and within the event the company management team discussed the on-going expansion of various workloads tied to high-performance computing, and the potential market opportunities tied to its entire suite of silicon, which involves server CPUs, FPGAs, network interconnects, Optane memory, SSDs and so forth. Intel owns a large technology stack tied to the datacenter, and it’s what gives them a defensive moat even with the prospect of heightening competition with Nvidia and AMD.
The presentation resurfaced much of the same commentary we’ve grown accustomed to from the datacenter presentation, which provided an update on their new line-up of various products tied to their datacenter business. The biggest takeaway from the event though was their launch of the Xeon 9200 Platinum processor, which can scale up to 56 cores, but with a dual socket configuration it could reach a core count of 112 cores on a single motherboard within a server rack system. Following the event, Intel stock climbed from $54 to $55.92 per share.
Of course, most datacenter customers don’t really buy at the high-end of Intel’s server stack, but instead purchase lower-end Xeon CPUs and stack them within a rack. Hence, the bigger takeaway from the Xeon 2nd generation launch was the 33% performance improvement across the entire server CPU stack from prior generation. This refresh in server CPUs fall under the family of Cascade Lake, which is still a 14nm process, but with higher emphasis on inference based or AI-based computing, which aims more of an arrow at Nvidia, whereas the stackable core count takes aim at AMD’s EPYC server CPU line-up.
The emphasis on Deep Learning improvements keeps Intel’s Xeon stack more competitive when compared to Nvidia’s various attempts to take datacenter market share with its graphics cards, which are used for inference computing and machine learning algorithms. At the event, Intel presented the Xeon Platinum 9200 versus a Nvidia P100 GPU (used in the datacenter and also costs $5,000) and was able to deliver performance in certain benchmarks that was better than the Nvidia card.
Hence, this paints more of a troubling picture for Nvidia, which has relied on machine inference or machine learning and various AI applications for adoption of Nvidia graphics cards in the past couple fiscal years to drive financial results. However, the launch of Xeon CPUs that can actually match performance in some inference workload instances diminishes the strengthening case that Nvidia has been able to put on display.
Not that Nvidia will sit around on its new-found datacenter market, but it’s likely to face stiffening competition from Intel on the basis of performance and with the use of general computing parts, which can be re-purposed for more uses than just interference-based computing.
Intel continued the event discussing various other datacenter market opportunities tied to other silicon, such as their Optane Memory solutions, which is sort of like RAM (random access memory), but with a much higher ceiling for total GBs in comparison to conventional DRAM sticks. They also announced the Agilex FPGA, which is a 10nm part, and helps them broaden their server stack with customizable hardware design for specific workloads like network packet processing and so forth (depends on what it’s repurposed for).
Intel’s datacenter business represents just under half of the company’s total revenue ($32.9 billion) and remains the company’s main growth category. Though the other chip companies do intend to take some segments of the market, Intel’s efforts to diversify away from just the CPU business will keep the business a little more balanced even as they’re anticipated to lose some market share to AMD due to the launch of various AMD parts in the foreseeable year. Intel’s response to competition was strong this year, perhaps financially the loss of market share may translate to a slower growth rate in its datacenter business, but it might not be as detrimental for shareholders as the entire datacenter market is expanding relatively quickly when compared to personal computing.
Disclosure: The author has no position in INTC. The information contained herein is for informational purposes only.
Read more on INTC: