Arm Reveals Full Technical Details of AGI Server CPU at Hot Chips 2026

Dual-chiplet design packs up to 136 Neoverse V3 cores, 2 TB/s inter-chiplet fabric, and a focus on latency-sensitive workloads

edit
By LineZotpaper
Published
Read Time2 min
Arm provided extensive architectural details of its AGI data center CPU at the Hot Chips 2026 conference, revealing a dual-chiplet processor with up to 136 Neoverse V3 cores, a 2 TB/s UCIe interconnect, and a design philosophy that eschews the heterogeneous multi-chiplet approaches of rivals AMD, Intel, and Nvidia in favor of monolithic per-chiplet SoCs aimed at reducing DRAM latency.

Arm’s AGI CPU, first announced earlier this year as its first data-center-class processor, is now slated for commercial shipments starting in late 2026. At Hot Chips this week, the company filled in key specifications it had previously withheld, including core counts, clock speeds, cache hierarchy, and memory bandwidth.

The processor is built from two chiplets fabricated on TSMC’s N3P process, each containing 50 billion transistors and 70 Neoverse V3 cores. Configurations will offer 64, 128, or 136 active cores per package. Each core features a 10-wide front end and decode, 10-wide dispatch, 8-wide retire, and a 384+ entry out-of-order window. Cores run at 2.80 GHz to 3.70 GHz and include 2 MB of dedicated L2 cache. A distributed system-level cache of up to 272 MB (128 MB per chiplet) sits on an 8×9 CMN-S3 mesh interconnect.

Memory support is a standout feature: each chiplet integrates a six-channel DDR5 controller, enabling up to 3 TB of DDR5-8800 per chiplet (6 TB per socket) and an aggregate memory bandwidth of 844.8 GB/s. Arm claims DRAM latency can be kept under 100 nanoseconds—a critical factor for latency-sensitive workloads such as single-threaded applications and agentic AI inference.

Arm opted not to follow the heterogeneous multi-chiplet pattern used by AMD, Intel, and Nvidia, where compute chiplets and I/O chiplets are separate. Instead, each AGI chiplet is a complete SoC with its own memory controllers and I/O. The inter-chiplet connection uses 16×16 UCIe macros running at 32 GT/s, delivering an aggregate bandwidth of 2 TB/s. The CPU also provides 96 PCIe 6.0 lanes with CXL 3.0 support, plus four PCIe 4.0 lanes and other standard interfaces. Thermal design power is rated at 300W.

Arm said the AGI processor is already working as expected and that the company is on track for commercial shipments in the coming months. The detailed disclosure at Hot Chips suggests the design is mature enough to share with potential customers and partners.

§

Analysis

Why This Matters

  • Arm’s entry into the data-center CPU market directly challenges incumbents AMD (EPYC), Intel (Xeon), and Nvidia (Grace) with a fresh architecture unconstrained by x86 legacy.
  • The focus on low DRAM latency and monolithic chiplets could give Arm an edge in emerging workloads like agentic AI, where response time is paramount.
  • If successful, AGI could accelerate the shift toward Arm-based server infrastructure, reshaping procurement strategies for hyperscale cloud providers.

Background

Arm has long supplied CPU designs for mobile and embedded devices, but its push into servers has been gradual. Early attempts, such as the failed “Server-class” Cortex-A72 systems, were eclipsed by AMD’s and Intel’s x86 dominance. The 2021 introduction of the Neoverse platform marked a serious pivot, targeting cloud and edge with scalable cores. The AGI CPU, announced earlier in 2026, is Arm’s first in-house designed data-center processor rather than a licensable core. Hot Chips 2026 provided the first public architectural deep dive, revealing Arm’s design philosophy and confirming that the chip is moving toward production.

Key Perspectives

Arm: Positions AGI as a high-performance, low-latency alternative that leverages Arm’s ecosystem and power efficiency. By keeping memory controllers on the same die as compute, Arm argues it avoids the latency penalty of traversing separate I/O chiplets. Competitors (AMD, Intel, Nvidia): They likely view AGI as a niche play, given their mature platforms with advanced packaging (e.g., AMD’s Zen chiplets, Intel’s EMIB, Nvidia’s Grace Superchip). Their heterogeneous designs may scale more cost-effectively for general-purpose cloud workloads. Cloud providers and hyperscalers: They stand to benefit from increased competition and silicon diversity, but adoption will depend on software ecosystem maturity, per-core performance benchmarks, and total cost of ownership versus existing x86 and Grace offerings.

What to Watch

  • Independent benchmark results comparing AGI’s per-core and per-watt performance against AMD Turin and Intel Diamond Rapids.
  • Arm’s ability to secure design wins with major cloud providers (AWS, GCP, Azure) beyond the initial shipping window.
  • The availability of DDR5-8800 memory modules, which are critical to achieving the claimed 844.8 GB/s bandwidth.
  • Any updates on AGI’s support for advanced I/O fabrics like CXL 3.0 memory pooling and disaggregation.

Sources

newspaper

Zotpaper

Articles published under the Zotpaper byline are synthesized from multiple source publications by our AI editor and reviewed by our editorial process. Each story combines reporting from credible outlets to give readers a balanced, comprehensive view.