x86 Duopoly Collaborates to Fend off Arm and RISC-V Threat

X86 architectures have dominated the data center and Personal Computer (PC) markets for over 2 decades. Intel and AMD have cemented their positions with a heavy Research and Development (R&D) moat, with cumulative spending totaling tens of billions of dollars over the years. This dominance has ensured software moved in lock-step with the architecture. For example, Windows, along with its extensive software suite, continues to run better on x86 than Arm-based processors (Apple transitioned to Arm-based processors in 2020). The new group addresses the lucrative PC and data center markets—as well as network equipment and other specialized hardware—and hopes to simplify software development across vendor platforms and allow microarchitecture designs to progress without the risk of forking along the way. This should offer customers more choice between Intel and AMD Central Processing Units (CPUs).

More recent developments include using chiplets in both data center CPUs like the 4th Gen AMD EPYC and the Intel Core Ultra processor offering, as well as the larger size of Artificial Intelligence (AI) domain-specific cloud deployments, creating the potential for “significant shifts” (in the words of the group) between Intel and AMD CPUs. By collaborating between industry leaders and major Original Equipment Manufacturers (OEMs) like Dell, Lenovo, Hewlett Packard (HP) and Hewlett Packard Enterprise (HPE), the group wants to guarantee predictability and consistency across future x86 CPUs.

The move follows the growth of Arm’s Neoverse Compute Subsystems (CSS) since its launch last year for the design of data center CPUs based on its competing RISC architecture, as well as the successful launch of Qualcomm’s Arm-based Windows platform for PCs. AMD and Intel’s x86 CPUs compete with RISC-based Arm CPUs primarily across data center, PC, and embedded system markets. However, Arm’s business model differs from the x86 giants as an Intellectual Property (IP) vendor. Chipset vendors like Ampere, Apple, MediaTek and Qualcomm innovate on top of Arm designs to varying degrees, depending on their licensing arrangement. Although the number of players in data centers and PCs is not vast, it has grown in recent years, particularly in the data center market, where captive vendors Microsoft, Google, and Amazon Web Services (AWS) have developed in-house CPUs targeting AI workloads based on Arm IP.

The main goal of this initiative is to guarantee that Operating System (OS) and software layers can run frictionless on Intel and AMD processors, regardless of each vendor’s innovations in the microarchitecture. Without this collaboration, the risk of fragmentation between Intel and AMD’s x86 ecosystems is present—despite the longstanding overlap. This is because any major architectural breakthroughs could lead to a competitive advantage in novel workloads for CPUs, including AI inferencing and confidential computing for AI databases. This can be contrasted with the dynamics of the Arm ecosystem: when disregarding the complexities of present and legacy licensing agreements, Arm can be regarded as the steering member of its software ecosystem, with which it actively collaborates to develop open system standards to guarantee compatibility, thus preventing software fragmentation and ensuring interoperability of system IP. So, the move by AMD and Intel to reduce the risk of x86 fragmentation seems sage.

AMD and Intel have a long history of collaboration on industry-defining standards like Universal Serial Bus (USB) and Peripheral Component Interconnect Express (PCIe), and more recently, Ultra Accelerator Link (UALink), although the latter, less mature, example has yet to render commercial examples due to its relative immaturity. Nonetheless, the current x86 initiative potentially goes deeper than before and ensures that any microarchitecture innovation does not preclude the other at the application software layer, ensuring compatibility for software made for x86. But there are elements of the software stack where unity is lacking. For example, considering the proliferation of AI workloads, over a year after its formation, AMD is not a member of the Unified Acceleration Foundation (UXL), a group driving an open standard accelerator programming model for heterogenous accelerated compute, despite the obvious place for its CPUs in such systems.

By guaranteeing compatibility for software going forward, Intel and AMD hope to accelerate innovation, as developers will know their applications can run on both vendors’ platforms, negating the need for concurrent projects. Still, questions remain about what exactly inspired this level of collaboration, but it likely stems from the rise of Arm in data centers and PCs, the financial frailty of Intel, and, to a lesser extent, the risk posed by open Instruction Set Architecture (ISA) RISC-V—the latter being a common enemy among all three players. Ultimately, this initiative is about the survival of x86, and the benefits will trickle down to its biggest implementers in the cloud and PC.