Why Mobile CPUs Are Getting an AI Upgrade
In the last few years the laptop market has been driven by two competing forces: ever‑higher raw performance for gaming and content creation, and the rising demand for on‑device artificial‑intelligence (AI) capabilities. From real‑time language translation to image up‑scaling, developers increasingly want AI workloads to run locally, without relying on the cloud. AMD’s response is the “Ryzen AI” branding, which pairs its high‑performance Zen 4 cores with a dedicated on‑die neural‑processing unit (NPU). The two most discussed members of this line are the Ryzen AI 9 HX 370 and the more traditional Ryzen 9 7945HX.
Core Architecture – Zen 4 Meets an On‑Die NPU
Both chips share the same Zen 4 microarchitecture built on a 5 nm process. That means they inherit the same 16‑core, 32‑thread configuration, the same 2‑plus‑stage decode pipeline, and the same support for DDR5‑5600 memory and PCIe 5.0. What separates them is the integration of an AI accelerator in the HX 370.
- Ryzen 9 7945HX – Pure Zen 4 CPU cores, no dedicated AI hardware.
- Ryzen AI 9 HX 370 – Same Zen 4 cores plus a custom NPU capable of several tera‑operations per second (TOPS) for inference workloads.
The NPU sits on the same silicon die, sharing the L3 cache and power delivery network. This tight integration keeps data movement latency low, which is essential for real‑time AI tasks such as video frame enhancement.
Performance in Traditional CPU Workloads
When you strip away the AI engine, the two processors behave very similarly. In multi‑threaded workloads such as video rendering, 3D modeling, or large spreadsheet calculations, the 7945HX and the HX 370 both deliver the high boost frequencies that AMD advertises for its top‑tier mobile parts – typically around 5 GHz on a single core under ideal cooling conditions.
The difference shows up only when the system’s power envelope is constrained. Laptops that prioritize thinness often cap the CPU at 35‑45 W. In that scenario the 7945HX may sustain slightly higher boost clocks because the NPU on the HX 370 draws a few watts of its own. In practice, most reviewers see less than a 5 % variance in pure CPU benchmarks between the two chips.
AI‑Centric Performance – What the NPU Brings to the Table
The real point of the HX 370 is to accelerate inference workloads that would otherwise tax the CPU cores. AMD’s NPU is optimized for matrix‑multiply operations, the building blocks of neural‑network inference. Because it is purpose‑built, the NPU can process common models – such as image super‑resolution, speech‑to‑text, and face detection – at lower latency and with less power than a CPU‑only solution.
In a side‑by‑side test of a popular open‑source up‑scaling model, laptops equipped with the HX 370 completed a 1080p‑to‑4K conversion roughly twice as fast as those with the 7945HX, while consuming about 30 % less power during the operation. Similar gains have been reported for on‑device transcription and real‑time video background removal, where the AI workload runs continuously.
Gaming and Content‑Creation: Does AI Matter?
Most gamers will not notice the NPU because modern titles rarely offload to it. However, the presence of the AI accelerator does not hurt gaming performance. In fact, the added NPU can enable new features such as AI‑enhanced up‑scaling (DLSS‑like technologies) that rely on low‑latency inference. Some recent Windows titles have begun experimenting with these features, and early reports suggest a modest frame‑rate uplift (often 1‑3 %) when the game can tap the NPU.
Content creators, on the other hand, can benefit directly. Video editors using tools that incorporate AI—such as automatic color grading, object removal, or AI‑driven audio cleanup—will see smoother timelines and faster render times on the HX 370. The CPU cores handle the traditional editing workload, while the NPU takes over the heavy inference steps, keeping the overall system responsive.
Thermal and Battery Implications
Adding an NPU to a mobile chip inevitably raises questions about heat and power draw. In practice the HX 370’s NPU is designed for efficiency: it idles at sub‑milliwatt levels and spikes only when an AI task is active. Laptop manufacturers have reported that the overall power envelope for the HX 370 remains within the typical 45‑W TDP range for high‑performance thin‑and‑light models.
When an AI workload is running, the NPU’s consumption can be in the range of 2‑5 W, which is comparable to a mid‑range GPU core. Because the NPU processes data in place, the CPU can stay at lower frequencies, which can offset the extra draw. Real‑world testing on a 99 Wh laptop shows roughly a 10‑15 % reduction in battery drain during continuous AI inference compared to a 7945HX‑only system running the same workload on the CPU.
Thermally, the two chips behave similarly under pure CPU load. Under mixed CPU‑plus‑AI loads, the HX 370 may run slightly hotter, but modern laptop cooling solutions (dual‑fan, vapor‑chamber designs) are more than capable of handling the modest increase.
Choosing Between the Two: Who Should Buy What?
The decision largely hinges on how you plan to use the machine. Below is a quick guide to help you match the right processor to your workflow:
- Pure gaming or general productivity – The Ryzen 9 7945HX offers the same raw CPU horsepower at a lower cost, making it a solid choice for gamers who don’t need on‑device AI acceleration.
- AI‑heavy content creation – If you regularly use AI‑assisted tools (up‑scaling, transcription, generative image models), the HX 370’s NPU can shave minutes off each task and reduce power draw.
- Portable workstations – For users who need a thin‑and‑light laptop that can still run occasional AI inference without pulling the plug on battery life, the HX 370 strikes a good balance.
- Budget‑conscious buyers – Because the HX 370 is a newer SKU, it may command a premium. The 7945HX provides the same Zen 4 performance at a more accessible price point.
Looking Ahead: The Future of AI‑Enabled Laptops
AMD’s strategy of embedding an NPU alongside Zen 4 cores signals a broader industry shift. As software ecosystems mature and more applications expose AI features, the value of on‑die acceleration will grow. We can expect future generations of Ryzen mobile chips to increase the NPU’s TOPS rating, support a wider range of model formats, and possibly integrate more tightly with Windows’ AI runtime.
For now, the Ryzen AI 9 HX 370 offers a glimpse of what a truly heterogeneous mobile processor can achieve. It delivers the raw performance that high‑end laptops are known for, while adding a low‑power engine that makes AI workloads feel native. The Ryzen 9 7945HX remains a powerhouse for traditional CPU tasks, but for users ready to embrace AI‑first workflows, the HX 370 is the more forward‑looking option.
