Technology · · 8 min read

Arm’s Cortex-X925 Reaches Desktop Performance, But the Ecosystem Remains Fractured

The chip designer's flagship core delivers 15% IPC gains and 36% total performance uplift — yet commercial adoption remains limited to MediaTek smartphones and Nvidia's $4,699 AI workstation.

Arm’s Cortex-X925, codenamed Blackhawk, represents the company’s most aggressive push into high-performance computing since launching the Cortex-X series, delivering what independent testing confirms is performance parity with AMD Zen 5 and Intel Lion Cove — but 16 months after its May 2024 announcement, the core remains largely confined to a handful of premium Android smartphones.

The X925 achieved a record 15% instructions-per-cycle (IPC) improvement over its predecessor on the Geekbench 6.2 benchmark, the largest year-over-year gain in Cortex-X history, according to Arm’s official blog. When combined with clock speeds reaching 3.8 GHz on TSMC’s 3nm process, the total performance uplift reaches 36% in single-threaded workloads.

Cortex-X925 Performance Gains vs. X4
IPC Improvement (ISO-frequency)+15%
Total Single-Core (3nm, 3.8GHz)+36%
AI Inference (Phi-3 model)+46%
Max L2 Cache3MB

But the technical achievement obscures a strategic problem: outside MediaTek’s Dimensity 9400 smartphone processor and Nvidia’s GB10 workstation chip, no major silicon vendor has committed to X925-based products for the PC market that Arm spent much of 2024 courting.

The Architecture: Massive Width Meets 3nm Efficiency

The X925 is a massive 10-wide core designed through and through to maximize performance, according to analysis from Chips and Cheese. The core features a 10-wide decode and dispatch width with doubled instruction window size, plus a 2x increase in L1 instruction cache bandwidth, per Wikipedia’s technical breakdown.

Key microarchitectural changes include 50% more vector units than the Cortex-X4 — six SIMD units up from four — which XPU.pub notes accounts for most of the core’s AI/ML performance boost. The out-of-order window doubled to 768 instructions in-flight, or 1,536 fused operations, according to WikiChip Fuse.

The CPU design includes an upgrade to the private L2 cache, increasing it from 2MB to 3MB, which Arm says significantly uplifts overall performance and power efficiency. The core is built on a 3nm process node as part of the second-generation ARMv9.2 architecture.

Technical Context

Arm’s shift to “Titan” nomenclature (X925 instead of X5) signals a strategic repositioning. The company is bundling cores into Compute SubSystems (CSS) — production-ready GDSII files optimized for specific foundry processes — rather than just licensing soft IP. This is the first time Arm has packaged cores as GDSII files that can be provided to fabricators like TSMC or Samsung for immediate production, taking into account specific quirks or features of the fabricator, according to XDA Developers.

Real-World Performance: Desktop Parity, With Caveats

Independent testing validates Arm’s boldest performance claims. Cortex X925 in Nvidia’s GB10 achieves performance parity with AMD’s Zen 5 and Intel’s Lion Cove in their fastest desktop implementations, writes Chips and Cheese in a March 2026 deep-dive.

However, Arm needs a large enough IPC advantage to overcome both a clock speed deficit and a less efficient representation of the work — Cortex X925 does achieve high IPC, but it’s not high enough in all workloads. Several SPEC CPU benchmarks require significantly more instructions on ARM versus x86, with 554.roms making X925 execute more than twice as many instructions compared to Zen 5.

The Qualcomm Oryon CPU in the Snapdragon 8 Elite clocks 19% higher than the Arm Cortex-X925 in the MediaTek Dimensity 9400 but is only 12% faster owing to the X925’s greater IPC, according to XPU.pub analysis from October 2024. That margin — meaningful in benchmarks — proves insufficient to overcome Qualcomm’s custom silicon advantage in the premium Android segment.

Market Reality: MediaTek Smartphones and Nvidia’s $4,699 Outlier

MediaTek officially announced its latest flagship SoC, the Dimensity 9400, which is built on the Armv9.2 instruction set and features a new Cortex-X925 prime core along with 3x Cortex-X4 cores and 4x Cortex-A720 cores, per NotebookCheck in October 2024. The implementation runs the X925 core at 3.62 GHz with 2MB L2 cache.

The Vivo X200 series includes three models — the X200, X200 Pro, and X200 Pro Mini — all equipped with MediaTek’s flagship Dimensity 9400 chipset, according to Gizmochina. OPPO brought the flagship MediaTek Dimensity 9400 SoC to global markets in its upcoming OPPO Find X8 and Find X8 Pro smartphones, per the company’s official announcement.

Beyond smartphones, the only commercial X925 deployment is Nvidia’s GB10 Grace Blackwell “Superchip.” The chip features a CPU tile designed by MediaTek which features a “big/little” architecture with 10 Arm X925 cores and 10 A725 for a total of 20, reports The Register. The workstation mini PC launched with an MSRP of $3,999, but the company has revealed that the system’s price will increase by $700, bringing its MSRP to $4,699, according to NotebookCheck in February 2026.

Commercial Cortex-X925 Implementations (as of March 2026)
Product X925 Config Clock Speed Launch Date Market
MediaTek Dimensity 9400 1x X925 3.62 GHz Oct 2024 Smartphones
MediaTek Dimensity 9400+ 1x X925 3.73 GHz Dec 2024 Smartphones
Nvidia GB10 (DGX Spark) 10x X925 ~3.8 GHz Jan 2025 AI Workstation

The Windows-on-Arm Problem: Qualcomm’s Lock, Microsoft’s Inertia

Arm positioned the X925 as delivering “ultimate performance for PCs,” with the company envisioning a PC chip with up to 12 Cortex-X925 CPU cores to push performance well beyond mobile. That vision has not materialized.

Qualcomm has had an exclusivity deal with Microsoft since 2017, with Snapdragon chipsets exclusively powering Windows on Arm laptops — this deal is reportedly set to expire at the end of 2024, Android Authority reported in May 2024. Yet 16 months later, none of the systems based on GB10 officially run Windows, instead shipping with a lightly-customized version of Ubuntu 24.04 Linux tailored to machine learning research and development.

Although Nvidia’s N1X chip was exposed to have entered the actual machine testing phase as early as the beginning of 2025, due to “serious software problems,” it wasn’t until the end of February 2026 that there was news indicating it would be first launched in gaming laptops by Dell and Lenovo in the summer of 2026, according to Chinese publication 36Kr.

The absence of MediaTek, Samsung, or other major silicon vendors with concrete X925-based PC product announcements suggests the Windows-on-Arm ecosystem remains fragile. Qualcomm is moving to a new custom CPU core for the Snapdragon 8 Gen 4, which means that the majority of flagship Android phones in 2025 probably won’t use Arm Cortex-X925 or Immortalis-G925 — likewise, the major wave of Windows on Arm laptops are all powered by Qualcomm’s Snapdragon X Elite platform, which uses custom CPU cores rather than Arm’s Cortex, per Android Authority.

“A 12% advantage confers bragging rights but not much else. The enthusiasts who purchase flagship Android smartphones might care but probably won’t be able to attribute differences between devices to peak single-thread CPU performance.”

— XPU.pub analysis, October 2024

The Desktop Performance Claim: Entry-Level or Competitive?

Arm’s “desktop-class performance” framing requires precision. Arm is claiming single-threaded IPC leadership with the largest core, the Cortex X925, surpassing what both Intel and AMD are capable of, which is a bold claim, noted AKEX Solutions in May 2024.

Independent verification supports competitiveness, not dominance. The X925 matches — but does not exceed — current x86 flagship performance in most workloads. In the same Geekbench tests Arm is referring to, Apple’s M4 is 27% faster still — presuming the X925 is relative to the X4 with “36 percent faster,” they’ll still be behind M3, much less M4, according to AnandTech forum discussion.

The claim is accurate for “entry-level desktop” performance — competitive with laptop-class x86 chips, but not replacing high-end desktop processors that clock significantly higher and benefit from mature software ecosystems.

What to Watch

MediaTek’s PC ambitions: There are rumors about MediaTek possibly getting into the PC game early in 2025, per Moor Insights & Strategy, but no concrete product announcements have materialized as of March 2026. Any delay beyond mid-2026 risks the window closing as Intel and AMD refresh laptop lineups.

Apple silicon comparison cadence: Apple’s M5 generation (expected late 2026) will reset performance benchmarks. If X925 successors cannot match M5 IPC gains, Arm’s licensing business faces pressure from its most successful architecture licensee.

Samsung Exynos adoption: Industry observers are guessing Samsung will bake the Exynos variant in their own fabs for their Galaxy book series, but Samsung has made no official confirmation. Any Exynos PC product would validate the X925’s viability outside MediaTek’s ecosystem.

Nvidia’s N1X Windows support: If Dell and Lenovo gaming laptops ship in summer 2026 as reported, it confirms Windows-on-Arm emulation has matured sufficiently for demanding workloads. Failure to ship — or shipping with severe software limitations — signals the ecosystem remains immature regardless of X925’s technical merits.

The X925 proves Arm can design cores that match x86 performance. Whether that matters depends entirely on an ecosystem that, 16 months post-announcement, has barely shown up.