Intel

A new benchmark published on August 7, 2026, by hardware reviewer Jeff Geerling confirms that the Intel Core 5 320 "Wildcat Lake" processor achieves a genuine performance-per-watt advantage over Apple's A18 Pro in one specific, historically significant test. But the full picture of independent lab data — including a published Tom's Guide head-to-head review — tells a more complicated story: Intel wins real-world battery life by a full hour, while Apple leads by wide margins on GPU performance, AI inference throughput, and single-core CPU speed. The efficiency question turns out to be the wrong question. The right question is: efficient at what?

What the Benchmark Actually Measured

The efficiency figure at the center of the story comes from Geerling's custom HPL FP64 benchmark suite. HPL — High Performance Linpack — is the same benchmark that powers the Green500 list, which ranks the world's supercomputers by energy efficiency. It specifically measures double-precision floating-point (FP64) throughput divided by power draw.

Running a Dell XPS 13 (Core 5 320, 8GB RAM, running Fedora 44) against a MacBook Neo (A18 Pro, 8GB unified memory), Geerling measured: the Core 5 320 achieved 127.91 GFLOPS at 20.6 watts of sustained draw, yielding 6.21 GFLOPS per watt. The A18 Pro achieved 57.012 GFLOPS at 10.6 watts, yielding 5.38 GFLOPS per watt. The data is documented in Geerling's Dell XPS 13 benchmark repository. The Intel advantage works out to 15.4% in this test. Geerling reports the Core 5 320 is #2 on his efficiency list — second only to the M4 Mac mini — and the first x86 chip he has ever placed in his top 10.

What this means: the Intel chip converts a larger fraction of every watt into FP64 floating-point work. What this doesn't mean: HPL FP64 is not a simulation of what a $699 laptop does when someone is editing a document, scrolling through a browser, or running an AI photo tool. Those workloads run at different power levels, different instruction mixes, and entirely different computational units — including the Neural Engine, which does not appear in HPL FP64 at all.

To understand why that distinction matters, consider the raw power draw. The Core 5 320 pulled 20.6 watts under HPL load; the A18 Pro pulled 10.6 watts. The Intel chip uses nearly double the power — and it still wins the efficiency ratio only because its FP64 output is more than double. That is genuinely impressive engineering. It is also not what happens when someone streams a video, opens a spreadsheet, or asks their laptop's AI assistant to rewrite a paragraph.

The more relevant consumer number comes from Tom's Guide's full review, published July 25, 2026. In continuous Wi-Fi web surfing with the display at 150 nits — the standard mixed-productivity battery test — the Dell XPS 13 lasted 14 hours and 26 minutes. The MacBook Neo lasted 13 hours and 26 minutes. Intel wins real-world battery life by exactly one hour. That result is meaningful, attributable to Intel's Wildcat Lake architecture using Darkmont low-power efficiency cores for light tasks at very low sustained draw, and directly applicable to a buyer who wants to know which laptop will last through a workday without a charger.

How Intel 18A Makes This Possible

The engineering story behind the battery win is rooted in Intel's 18A process node — the most advanced chip-making technology in commercial production in the United States, and the first to simultaneously deploy both RibbonFET transistors and PowerVia backside power delivery. The Intel 18A process page details the full specification.

RibbonFET is Intel's implementation of gate-all-around (GAA) transistors. Where the prior FinFET design wrapped the gate around three sides of the transistor channel, RibbonFET wraps it around all four. The additional gate contact gives engineers finer control over current flow, reduces leakage at low voltages, and enables the chip to operate efficiently at lower power states — directly relevant to the Darkmont LP E-cores that handle light productivity workloads. Technical details are documented on Intel's 18A process specification page.

PowerVia moves the power delivery network to the back of the silicon wafer, separating it from the signal interconnects on the front. The conventional design stacks power and signal lines together on the front, creating wiring congestion that limits both density and efficiency. With PowerVia, Intel's PowerVia test results report approximately 30% reduction in voltage droop and a 6% boost in operating frequency at the same power level. Standard cell utilization exceeds 90%.

The combined effect: Intel 18A delivers up to 18% higher performance at the same power, or 38% lower power at the same performance, compared to Intel's prior Intel 3 node. The Wildcat Lake Core 5 320's compute die is approximately 70mm² — remarkably small — which keeps manufacturing cost low and yield high, explaining how a chip on the newest process node can power a $699 laptop.

The A18 Pro, by contrast, uses TSMC's N3E (3nm-class) process. TSMC N3E is a mature, proven node. Intel 18A achieves approximately 238 million transistors per square millimeter by industry estimates; TSMC's newer N2 node leads density at around 313 million transistors per square millimeter, but N3E — the A18 Pro's node — predates both. Apple's efficiency advantage in the MacBook Neo's form factor comes not from transistor density alone but from the A18 Pro's ARM architecture (which requires fewer transistor toggles per instruction than x86), Apple's tight software-hardware optimization, and the chip's deeply specialized silicon blocks.

Where Apple Still Wins: GPU, AI Inference, and Single-Core

The battery advantage and the HPL FP64 efficiency number are real. What they don't change is Apple's lead in three other dimensions that matter significantly to the buyer population these laptops target.

Single-core CPU speed. In Geekbench 6, the platform-level measure of single-task responsiveness, the MacBook Neo (A18 Pro) scored 3,535 in single-core; the Dell XPS 13 (Core 5 320) scored 2,684 in Tom's Guide's testing. Apple leads by 38% in single-core throughput. This is what makes applications open faster, renders individual web pages more quickly, and produces the responsiveness gap that PhoneBuff's robot-arm testing captured — Microsoft Word opening in 7 seconds on the MacBook Neo versus 12 seconds on the XPS 13.

GPU performance. The Core 5 320 ships with two Xe3 integrated graphics cores. The A18 Pro in the MacBook Neo (note: the MacBook Neo uses a five-core GPU variant, not the six-core version in the iPhone 16 Pro) includes Apple's full graphics pipeline with hardware-accelerated ray tracing, as confirmed in Apple's A18 chip specifications. In Tom's Guide's 3DMark Wildlife Extreme test, the XPS 13 scored 2,455 at 14 frames per second; the MacBook Neo scored 3,661 at 21 fps — a 49% GPU advantage for Apple. In Jeff Geerling's GravityMark test, the MacBook Neo scored 14,893 versus the XPS 13's 8,281 to 8,648 — a lead of 72% to 80% depending on the API. For photo editing, video rendering, or any compute that routes through the GPU, the gap is substantial.

AI inference. The Core 5 320's NPU 5 block delivers 16 TOPS (tera-operations per second) of INT8 inference performance. The A18 Pro's Neural Engine is rated at 35 TOPS. Apple leads by 2.2x in dedicated AI throughput. This gap directly determines which laptop runs AI-powered features faster: photo enhancement using computational photography, real-time audio transcription, AI writing tools, and any application built around Apple Intelligence. For a buyer who specifically wants a laptop for AI-assisted creative work or productivity, the NPU lead is significant and architecturally persistent — not something that firmware or software updates can close.

The NPU gap also carries a practical consequence beyond feature performance: Microsoft's Copilot+ PC certification requires 40 TOPS of dedicated NPU performance. At 16 TOPS, the Core 5 320 falls well short of that threshold. The Dell XPS 13 (Core 5 320) is not a Copilot+ PC and cannot run Windows Recall, the real-time NPU-powered captioning features, or other Copilot+ capabilities that Microsoft has been marketing aggressively.

What Intel Wins Beyond Battery

The XPS 13's strengths extend beyond battery and FP64 compute. Tom's Guide's Blackmagic SSD test recorded write speeds of 3,714 MB/s and read speeds of 5,026 MB/s from the XPS 13's PCIe 4.0 NVMe drive, compared to the MacBook Neo's 1,440 MB/s write and 1,585 MB/s read. The storage advantage is substantial — and the XPS 13's 2230-sized NVMe slot is user-upgradeable, confirmed by Geerling's teardown and benchmark data.

On sustained CPU workloads, Intel's multi-core lead persists. In Handbrake (converting a 4K clip to 1080p), the XPS 13 completed the task in 10 minutes and 38 seconds; the MacBook Neo finished in 9 minutes and 57 seconds. Apple wins by 41 seconds — meaningful for regular video editors, but not a decisive gap for occasional use. In PassMark, the Core 5 320 scores 15,222 in multi-core — confirmed on the PassMark benchmark database — leading the A18 Pro's 12,000–12,860 range by 21–27%.

The XPS 13 also weighs 2.2 pounds (1 kilogram) — lighter than the MacBook Neo at 2.7 pounds (1.2 kg), a meaningful advantage for buyers who carry their laptop all day.

And the price equation has equalized since the original draft was written. As of June 25, 2026, the MacBook Neo costs $699 — no longer $599 — after Apple raised prices citing an AI-driven DRAM shortage that pushed LPDDR5X contract prices up roughly 89% in a single quarter. Both laptops now sell at identical retail prices. Students can access $599 pricing from either: Intel via Dell's verified-student discount (expiring November 2, 2026) or Apple via the Apple Education Store.

What the Efficiency Story Actually Says About Intel 18A

The deeper significance of Geerling's HPL finding isn't which laptop is "better." It's what the Core 5 320 result says about Intel's 18A manufacturing process at the budget tier.

Before Wildcat Lake, no x86 chip had broken 4 GFLOPS per watt in Geerling's benchmark. The prior best x86 result — an Intel Core Ultra 9 265K, a premium desktop chip — measured 2.71 GFLOPS per watt, documented in Geerling's top500 benchmark repository. Intel's 18A process node, built for efficiency through RibbonFET's gate-all-around transistors and PowerVia's voltage-droop reduction, has delivered a real-world efficiency gain that is measurable, independently reproducible, and documented by a credible third-party tester. That result holds up on a $699 budget chip — not just on a premium flagship.

Intel's own marketing claims for its Panther Lake (Core Ultra Series 3) chips at CES 2026 — including 50% better efficiency than AMD's Ryzen AI 300 series and 30% lower single-thread power draw — have not been independently verified in direct controlled comparisons against Wildcat Lake specifically. But Geerling's HPL result, and the Tom's Guide battery result, provide independent corroboration that 18A's efficiency promises are not entirely marketing. They're partially real, specifically in x86 compute-per-watt at sustained loads and in light-load battery endurance via the Darkmont cores.

Whether Intel can extend this efficiency narrative to GPU performance and AI inference — the two dimensions where Apple still leads significantly — depends on whether future Wildcat Lake revisions or Panther Lake products increase Xe3 core count and NPU throughput without sacrificing cost. A rumored Wildcat Lake Refresh, expected as part of Intel's "Core 400" series in 2027, reportedly doubles the P-core count to four (from two), per a Tom's Hardware Wildcat Lake Refresh report. Whether NPU throughput increases proportionally remains unreported.

Who Should Buy Which

For buyers at the $699 price point, the data now points to a clear workload-based decision:

Buy the Dell XPS 13 (Core 5 320) if: your primary workloads are general productivity — document editing, browser use, videoconferencing, email — and you want the longest real-world battery, a touchscreen, 120Hz display, faster storage, lighter weight, or a Windows environment. The one-hour battery advantage over the MacBook Neo and the SSD speed advantage are real, and the efficiency gains from Intel 18A make this the most competitive x86 budget laptop Intel has fielded in years.

Buy the MacBook Neo (A18 Pro) if: you use AI-powered features regularly — photo enhancement, transcription, Apple Intelligence writing tools — or do GPU-intensive work like video editing or graphics at this price point. Apple's 2.2x NPU lead and 49–80% GPU lead are architectural gaps that no amount of battery endurance can substitute for. For single-task responsiveness — the speed of opening apps, rendering web pages, and completing quick operations — the A18 Pro's 38% single-core Geekbench lead is also meaningful.

The 15% HPL FP64 efficiency advantage for Intel is real. For most buyers at $699, however, real-world battery life (which Intel also wins, by a different and more direct measure) and the AI inference gap (which Apple wins substantially) are the numbers that will determine whether the right choice turns out to be an x86 laptop or an ARM one.

Frequently Asked Questions

Does the Dell XPS 13 (Core 5 320) actually have better battery life than the MacBook Neo?

Yes, in Tom's Guide's standardized battery test — continuous Wi-Fi browsing at 150 nits — the Dell XPS 13 (2026) lasted 14 hours and 26 minutes, compared to 13 hours and 26 minutes for the MacBook Neo. That's a one-hour advantage for the Intel laptop. The result reflects Intel's Wildcat Lake architecture using Darkmont low-power efficiency cores for light workloads, which keeps sustained draw very low during the kinds of tasks that dominate typical laptop use — web browsing, document editing, and light media.

What is HPL FP64 efficiency, and why does it matter for laptop buyers?

HPL (High Performance Linpack) is a floating-point computation benchmark used to rank the world's most powerful supercomputers. Jeff Geerling adapts it for laptops to measure how many double-precision floating-point operations a chip can deliver per watt — a metric of deep interest to scientists, engineers, and developers who run numerical simulations or high-performance computing workloads. For most $699 laptop buyers, HPL FP64 efficiency is not directly predictive of battery life in everyday use. The real-world battery test (see above) is the relevant consumer figure. The Geerling result does matter as evidence that Intel's 18A process node is delivering genuine efficiency improvements at the transistor level — just not in every workload category.

Is the Dell XPS 13 a Copilot+ PC?

No. Microsoft's Copilot+ certification requires a minimum of 40 TOPS of dedicated NPU performance. The Core 5 320's NPU 5 delivers 16 TOPS — well below the threshold. This means the XPS 13 cannot run Windows Recall, the NPU-powered real-time captioning features, or other AI capabilities Microsoft designates as Copilot+-exclusive. Buyers who specifically want on-device AI acceleration should look at Intel's Panther Lake-based laptops (Core Ultra Series 3, with a 50-TOPS NPU) or choose the MacBook Neo, whose 35-TOPS Neural Engine exceeds the equivalent Windows Copilot+ threshold.

How does Intel's 18A process compare to the A18 Pro's manufacturing process?

The Apple A18 Pro is built on TSMC's N3E (3nm-class) process. Intel's 18A is the first production node to simultaneously combine RibbonFET gate-all-around transistors with PowerVia backside power delivery — architectural innovations that improve efficiency through better transistor control and cleaner power routing. Intel 18A delivers up to 18% higher performance at the same power level, or 38% lower power at the same performance level, compared to Intel's prior Intel 3 node. TSMC's newer N2 node leads in transistor density (around 313 million transistors per square millimeter), but N2 is not yet used in consumer laptops at this price tier. The Wildcat Lake Core 5 320 is built on a more advanced process node than the A18 Pro's N3E — a meaningful reversal from earlier Intel generations that relied on older nodes than TSMC's best.

Originally published on Tech Times