Ryzen AI 300 puts a 50-TOPS XDNA 2 NPU beside Zen 5 and RDNA 3.5
AMD Ryzen AI 300 is a mobile processor family designed around CPU, GPU and neural-processing hardware on the same platform. It targets premium Windows laptops and the wider Copilot+ PC category.
The Ryzen AI 9 HX 370 combines four Zen 5 cores and eight compact Zen 5c cores for 12 cores and 24 threads, together with Radeon 890M integrated graphics and AMD’s XDNA 2 NPU.
TechnologyBlog.co.za has not independently benchmarked Ryzen AI 300. Performance varies substantially with the laptop’s cooling, power configuration, memory and software.
XDNA 2 supplies up to 50 TOPS for local AI
AMD rates the XDNA 2 NPU in Ryzen AI 300 at up to 50 TOPS. Microsoft’s Copilot+ PC specification requires a sufficiently capable NPU for certain local AI experiences.
TOPS measures theoretical operations per second under particular numerical formats. It is useful for classifying hardware but does not tell users how quickly every AI model will run.
Software support, model precision, memory bandwidth and whether a task can actually execute on the NPU all affect real performance.
Zen 5 and Zen 5c share the CPU workload
The Ryzen AI 9 HX 370 uses 12 CPU cores: four Zen 5 and eight Zen 5c. AMD’s compact Zen 5c cores implement the same instruction-set family while optimising physical area and efficiency.
The processor supports simultaneous multithreading for 24 threads and boosts to as much as 5.1GHz under suitable conditions.
AMD allows laptop makers to configure power across a broad range, so two machines with the same processor name can behave differently under sustained load.
RDNA 3.5 graphics keep light GPU work on the processor
Radeon 890M integrated graphics provide 16 graphics cores on the HX 370. This is designed to handle display, media, GPU compute and gaming without requiring a discrete graphics processor in every system.
Integrated graphics share system memory, so memory configuration can materially affect performance.
For demanding 3D workloads, a laptop with a discrete GPU can still offer substantially more graphics throughput.
Who is Ryzen AI 300 for?
The family suits thin-and-light and performance laptops that need a capable general CPU while supporting local AI features without waking a discrete GPU for every inference task.
Buyers should evaluate the whole laptop rather than buying on the processor name alone. Cooling, display, battery size, memory capacity and OEM power limits often shape the experience as much as the silicon.
Ryzen AI 9 HX 370 reference specifications
| Specification | Details |
|---|---|
| CPU architecture | Zen 5 + Zen 5c |
| CPU cores / threads | 12 / 24 |
| Maximum boost | Up to 5.1GHz |
| Default TDP | 28W |
| Configurable TDP | 15–54W |
| NPU architecture | AMD XDNA 2 |
| NPU performance | Up to 50 TOPS |
| Integrated graphics | Radeon 890M |
| GPU cores | 16 |
| Memory support | DDR5 and LPDDR5X |
| Process | TSMC 4nm |
A South African lens
In South Africa, model availability can be uneven, so the exact Ryzen AI 300 SKU and laptop configuration matter more than the family name on a retailer page. That matters because many advanced components arrive through finished systems rather than direct component channels. Warranty, firmware support and the configuration chosen by a laptop, server or device vendor can have more practical impact than the theoretical maximum in the silicon data sheet.
The useful question is workload fit
The strongest way to judge AMD Ryzen AI 300 Series is to start from the workload: what data type is processed, how much memory it needs, how latency-sensitive it is, how long the device must sustain performance and which software APIs it uses. Once those questions are clear, headline specifications become much easier to interpret. Without that context, large numbers risk becoming marketing shorthand instead of engineering information.
The architecture matters more than one peak number
AMD’s Strix Point mobile processor family combining Zen 5 CPU cores, RDNA graphics and an XDNA 2 NPU. The platform reflects how notebook performance is now split across CPU, GPU and NPU rather than measured by CPU frequency alone. This is why peak clock speed, TOPS, bandwidth or capacity should be read as one characteristic of a larger compute system rather than a universal performance score. TOPS figures describe peak AI arithmetic and do not guarantee application speed; software must target the right engine and memory bandwidth still matters. Workloads behave differently, and the software stack decides which block actually does the work.
Data movement is becoming the hidden performance tax
Modern processors spend substantial energy moving data between memory, caches and execution units. AI has made that problem more visible because matrix engines can consume data faster than conventional memory systems can supply it. Designers therefore spend as much effort on cache, memory bandwidth, interconnect and packaging as they do on arithmetic throughput. AMD Ryzen AI 300 Series should be evaluated through that systems lens: a faster engine that waits for data can still underperform a better-balanced design.
Software determines whether specialised hardware earns its silicon
NPUs, DSPs, vector extensions and dedicated accelerators only help when operating systems, frameworks and applications know how to target them. A feature can exist in hardware for years before it becomes commonplace in mainstream software. Developers also care about debugging, model conversion, libraries and fallback behaviour. That means platform support and toolchains can be as important as the block diagram when a company decides whether to build around a processor family.
What buyers and engineers should check
Laptop buyers should compare the complete notebook: cooling, battery size, memory configuration, screen, firmware and vendor power limits can change the experience of the same processor. It is also worth separating a family name from the exact part number. Vendors often use one brand across devices with different core counts, memory capabilities, clocks or I/O. A procurement sheet should therefore record the precise SKU and the system configuration around it, not only the product family.
Five questions worth asking before committing
Before adopting AMD Ryzen AI 300 Series, write down the problem it is meant to solve, the metric that will show improvement, the systems or people it depends on, the failure mode that would hurt most, and the support path when something goes wrong. Laptop buyers should compare the complete notebook: cooling, battery size, memory configuration, screen, firmware and vendor power limits can change the experience of the same processor. That exercise prevents a technically impressive product from becoming a solution in search of a problem. It also creates a baseline for later review: if the expected outcome does not improve, the organisation can change configuration, training or even the product choice instead of defending the original purchase.
Sources and verification
AMD Ryzen AI 9 HX 370 specifications. AMD Ryzen AI 300 architecture announcement.
