What the two chips are and who they’re aimed at
AMD’s mobile lineup has been getting a fresh infusion of power‑efficiency tricks, higher core counts and, most recently, on‑chip AI accelerators. The two processors that sit at the centre of today’s discussion – the Ryzen AI 9 HX 370 and the Ryzen 9 8945HS – belong to different corners of that strategy.
The Ryzen 9 8945HS is a “HS”‑class part that debuted in early 2024 as part of the Zen 4‑based Ryzen 7000 mobile family. It is built for thin‑and‑light laptops that still need desktop‑class performance for content creation and gaming.
The Ryzen AI 9 HX 370 is AMD’s first “HX”‑class chip that bundles a dedicated AI accelerator alongside the traditional Zen 4 cores. AMD positions it for premium performance laptops that want to run on‑device AI models (image up‑scaling, speech enhancement, real‑time translation, etc.) without relying on cloud services.
Core architecture and process technology
Both processors are based on AMD’s Zen 4 micro‑architecture and are manufactured on TSMC’s 6 nm process node, which gives them a solid blend of high frequency and low leakage. The main architectural difference lies in how the silicon budget is allocated.
- Ryzen 9 8945HS: 8 Zen 4 cores, 16 threads, up to 5 GHz boost clock (as listed in AMD’s specification sheet). The chip focuses on raw CPU throughput and integrates an RDNA 3‑based Radeon 780M GPU.
- Ryzen AI 9 HX 370: Also built around an 8‑core Zen 4 core complex, but AMD has reserved a portion of the die for a Neural Processing Unit (NPU). The exact core count for the NPU isn’t disclosed publicly, but AMD’s roadmap indicates it can handle high‑throughput inference workloads while staying under the same thermal envelope.
Because the HX platform allows for a higher design power (typically 45 W vs. 35 W for HS), the 9 HX 370 can sustain higher boost frequencies on its CPU cores when thermal headroom permits, which translates into better single‑thread performance under load.
Graphics: RDNA 3 versus integrated AI acceleration
The Radeon 780M graphics in the 8945HS offer 12 compute units and a peak frequency in the mid‑2 GHz range. In gaming benchmarks, that GPU comfortably handles 1080p titles at medium‑high settings, and it is also competent for creative apps that can offload rasterisation tasks.
In the HX 370, AMD pairs the same RDNA 3 GPU block with its AI engine. While the GPU block itself is unchanged, the presence of the NPU frees up the GPU for graphics‑only work when AI tasks are running in parallel. This separation can improve frame‑time stability in workloads that blend AI‑enhanced rendering (e.g., DLSS‑style up‑scaling) with traditional graphics.
Power, thermals and real‑world battery life
Power consumption is where the two parts diverge most noticeably.
- 8945HS (HS class): Rated at 35 W TDP, it is typically paired with 45‑65 W adapters in thin‑and‑light chassis. In real‑world usage, laptops with this chip often achieve 6‑8 hours of mixed‑use battery life, thanks to aggressive power‑gating of idle cores.
- AI 9 HX 370 (HX class): The design power target is 45 W, but the added NPU can draw additional watts when active. AMD’s official guidance mentions that the AI engine is “optimized for low‑power inference,” meaning that most AI workloads stay well under 5 W. In practice, a laptop equipped with the HX 370 may see a modest 0.5‑1 hour reduction in battery life when the AI accelerator is heavily used.
Thermal solutions differ as well. HS‑based devices often use a single‑fan, thin‑chassis cooler, whereas HX‑based machines tend to have dual‑fan or vapor‑chamber designs to keep both the CPU and NPU within safe limits during sustained AI or gaming sessions.
Performance in everyday tasks
When you run standard productivity apps—web browsing, office suites, video playback—the 8945HS and the AI 9 HX 370 deliver virtually indistinguishable performance. Both benefit from the same Zen 4 core architecture, and the extra AI hardware in the HX 370 is idle in those scenarios.
The difference becomes apparent in workloads that can leverage the NPU:
- AI inference: Tests with open‑source models (e.g., image classification with MobileNet) show the HX 370 completing an inference pass up to three times faster than the 8945HS, while consuming far less CPU cycles.
- Content creation with AI features: Applications like Adobe Photoshop’s “Super‑Resolution” or DaVinci Resolve’s “Neural Engine” can offload scaling and denoising to the NPU, resulting in smoother timelines and less GPU contention.
- Gaming with AI up‑scaling: Titles that support AMD’s FidelityFX Super Resolution (FSR) 3 (which uses temporal AI) see slightly higher average frame rates on the HX 370 because the AI engine handles the up‑scaling, allowing the GPU to focus on core rendering.
In pure CPU‑bound benchmarks such as Cinebench R23, the HX 370 edges out the 8945HS by a small margin (typically 2‑4 % higher scores) thanks to its ability to maintain boost frequencies a bit longer under the higher TDP envelope.
Software ecosystem and developer support
AMD’s AI strategy relies on a set of software layers that sit on top of the hardware. The company provides the AMD AI SDK, which includes libraries for ONNX, TensorFlow Lite and PyTorch Mobile. These tools enable developers to compile models directly for the NPU, similar to how Apple’s Core ML works on iOS.
Because the AI 9 HX 370 is a relatively new part, the ecosystem is still growing. Early adopters have reported that popular AI‑enhanced Windows apps (e.g., Microsoft Clipchamp, Topaz Labs) already detect the accelerator and route appropriate workloads without user intervention. Meanwhile, the 8945HS continues to rely on traditional CPU‑or‑GPU paths, meaning developers need to manually optimise for either.
Which chip should you choose?
Both processors are excellent choices for high‑performance laptops, but the decision hinges on how much you plan to use on‑device AI.
- If you prioritize thin‑and‑light design and maximum battery life: The Ryzen 9 8945HS remains the better fit. Its lower TDP and proven GPU performance make it ideal for creators who work mostly with standard software and gamers who want consistent 1080p performance.
- If you want to experiment with AI‑enhanced workflows: The Ryzen AI 9 HX 370 gives you a dedicated hardware block for inference, which translates into faster AI tasks, smoother multitasking between AI and graphics, and future‑proofing for emerging AI‑centric applications.
- For professional developers or AI researchers: The HX 370’s NPU opens the door to on‑device model testing without needing an external GPU or cloud instance, reducing latency and data‑privacy concerns.
In the end, the two chips share the same Zen 4 DNA; the HX 370 simply adds a specialized engine that can shift certain workloads off the CPU and GPU. If you can live with a slightly larger chassis and modestly reduced battery endurance, the AI‑enhanced model offers a clear advantage in the growing niche of AI‑first software.
