It is no secret that Intel is under new leadership with Pat Gelsinger, given his return to the iconic Silicon Valley company earlier this year, where he spent the first thirty years of his career. The semiconductor giant is subsequently making waves with several recent announcements. One caught my attention at last week’s Intel Architecture Day – Mount Evans. Intel Mount Evans is the company’s new foray into the data processing unit (DPU) market, and I would like to share my takeaways.
IPU versus DPU
By definition (Marvell and NVIDIA) a DPU is a compute entity that moves, processes, secures and manages data as it travels or rests. Ultimately, this enables data to be available and optimized for applications. I have written about DPUs in the past concerning Marvell. If interested, you can find that article here.
With definitions out of the way, Intel is pinning its hopes on Mount Evans to help it compete with DPU offerings from NVIDIA and Marvell. I like that Intel is attempting to differentiate itself, referring to Mount Evans as an Infrastructure Processing Unit (IPU) ideally positioned to meet hyperscaler needs. Intel says that a top cloud provider is involved in the design of Mount Evans. That is an interesting factoid and points to the fact that Mount Evans should have legs out of the gate when it becomes generally available. Moor Insights & Strategy CEO Patrick Moorhead believes that CSP is Microsoft Azure given the work Intel did with FPGAs and networking.
Under the hood
And what a design it is. From a networking perspective, Mount Evans can support the demands of cloud service providers in connecting up to four Intel Xeon CPUs. This multi-host adapter functionality coupled with improved packet processing leveraged from Intel’s Barefoot Networks acquisition makes for a compelling set of networking features. From a compute perspective, Mount Evans is packing up to sixteen Arm Neoverse N1 cores. On the surface, one might assume that Intel’s use of Arm instead of x86 is odd. However, the reality is that Arm has risen to become the defacto DPU standard, given Marvell’s use of it in its Octeon family. In my mind, Intel is wise also to adopt it in Mount Evans to ensure its long-term competitiveness.
From my perspective, Intel is clearly in the DPU game with Mount Evans, and it should be a formidable challenger to Marvell and NVIDIA. Today, I give Marvell the performance edge with its latest Octeon 10 family’s leverage of Arm Neoverse N2 cores. However, I am confident that Intel will close the gap over time as it develops future iterations of its IPU platform. I also believe that in time, Intel’s entry into DPUs/ IPUs will attract the attention of venture capitalists, and that should fan the flames for innovation and new market entrants in the segment. Competition always breeds innovation, and the DPU category should be fun to watch in the coming quarters.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.