Snapdragon X2 Elite Extreme vs Apple M5: Reading the Independent Numbers

Verified as of 12 August 2026. Figures below are attributed to the outlets and databases that produced them.

Disclosure: TechPari has not independently tested either platform. This is an analysis of published third-party testing. Where sources disagree, we say so rather than picking a number.

Two different bets

Qualcomm and Apple have arrived at opposite answers to the same question about laptop silicon.

Qualcomm went wide. PCMag's hands-on coverage of the Snapdragon X2 Elite Extreme describes a jump to 18 cores, an expanded cache, a boost clock reaching up to 5GHz, and an NPU rated at 80 TOPS - close to double the roughly 45 TOPS of the prior generation.

Apple went narrow and fast. The M5 keeps a lower core count at a higher sustained clock, with the Pro variants scaling cores while retaining the same basic approach.

Neither is obviously correct. They optimise for different workloads, and the benchmark suite you pick largely determines which one wins.

What the published numbers show

Aggregated database figures - useful for rough positioning, weaker as evidence than a controlled review:

| | Snapdragon X2 Elite (X2E-88-100) | Apple M5 | | :--- | :--- | :--- | | Cores | 18 | 10 | | Base / peak clock | 3.4 GHz | 4.61 GHz | | L2 cache | 53 MB | 28 MB | | Geekbench 6 single-core | - | ~4,133 | | Geekbench 6 multi-core | - | ~16,472 |

Source: cpu-monkey aggregated database. Treat aggregator figures as indicative. Submitted-result databases mix hardware configurations, thermal envelopes and OS versions.

PCMag, which ran its own testing, reported the X2 Elite Extreme ahead of Apple's part in its benchmark suite. Tom's Guide, comparing early Geekbench 6 results, characterised both as reaching among the highest CPU scores it had recorded and called the outcome closer than expected.

These are not contradictory. They are different suites weighted differently. That is the point.

Reading these numbers properly

Three cautions that matter more than any individual score.

Naming is genuinely confusing. Qualcomm ships at least four SKUs under similar names - X2E-88-100, X2E-90-100, X2E-94-100, X2E-96-100 - with different clocks and configurations. Apple ships M5, M5 Pro at 15-core, and M5 Pro at 18-core. A comparison that names only "X2 Elite" versus "M5" has not specified what it measured. Check the exact SKU before treating any number as applicable.

Geekbench 6 rewards Apple's design. It is a short-burst benchmark. High single-core clocks look excellent under it. Sustained multi-threaded compilation looks different, and 18 cores with 53 MB of L2 look considerably better there. Neither result is wrong; they answer different questions.

Chip benchmarks are not laptop performance. PCMag tested the X2 Elite Extreme in a specific machine - Asus's 16-inch Zenbook A16, shown at CES 2026 and named the show's best ultraportable by their team. Cooling design, power limits and firmware move sustained results substantially. The same silicon in a thinner chassis behaves differently.

The NPU question

The 80 TOPS figure is the most interesting number here and the hardest to evaluate.

TOPS is a theoretical peak throughput rating. It tells you very little about whether a given model runs well, because real on-device inference performance depends on memory bandwidth, quantisation support, and - decisively - whether the software you use actually targets that NPU.

On-device AI is a legitimate reason to care about either chip. But the useful question is not "how many TOPS" - it is "does the model I want to run have a path to this accelerator." For much of the current tooling ecosystem the honest answer remains partial, and it varies by framework.

Treat TOPS as a ceiling, not a measurement.

Who should care

Windows-on-ARM compatibility is no longer the blocker it was, but it is not fully resolved either. If your work depends on specialised native tooling, drivers, or virtualisation, verify your specific stack rather than trusting a general claim of maturity.

For sustained multi-threaded work - large compiles, heavy container workloads, parallel test suites - the wide-core approach with substantially more L2 has a clear architectural argument.

For latency-sensitive single-threaded work and for anyone already invested in Apple's ecosystem, the high-clock design plus platform integration remains a strong position.

For on-device AI, both are credible and neither is proven for your particular workload. Test with your models.

What we cannot tell you

We have not run these chips. Specifically unresolved from published sources:

• Sustained thermal behaviour across chassis designs. Reviews test specific laptops. Generalising to the silicon is a leap.
• Battery life under mixed real-world load. Vendor claims and standardised loops both diverge from actual use.
• Real-world NPU utilisation across common development and creative tooling.

If any of those three decide your purchase, wait for testing on the specific machine you intend to buy.

---

Sources

• PCMag - Snapdragon X2 Elite Extreme benchmark testing
• Tom's Guide - Apple M5 vs Snapdragon X2 Elite Extreme early benchmarks
• cpu-monkey - X2 Elite vs M5 specification comparison