AMD has launched its new Ryzen AI 300 series “Strix Point” APU-powered laptops, which see the combination of Zen 5 CPU cores, RDNA 3.5 GPU cores, and an XDNA 2 NPU for AI workloads.
7VIEW GALLERY – 7 IMAGES
AMD had the first “AI PC” processor with its Phoenix APUs back in 2023, improving things across the board with the Hawk Point APUs with higher TOPS out of the NPU. Now, we have the far more powerful and advanced Strix Point APUs which make them the most powerful Copilot+ PC processors on the planet.
The company is doing things a little different here with the Ryzen AI 300 series “Strix Point” APUs as AMD is converging its portilio and doesn’t rely on a fixed TDP for a selected SKU anymore. This means that there are no U, H, HS series branding, and that AMD is now giving OEMs the flexibility to adjust Strix Point APUs to their demands with TDPs ranging between 15W and 54W.
The new “HX” branding marks that particular Strix Point APU as the best SKU, while non-HX Strix Point APUs will be the power-efficient chips.
7AMD’s new Zen 5 architecture launches with the Ryzen AI 300 series “Strix Point” APUs, joined by new RDNA 3.5 GPU cores and new XDNA 2 architecture powering the NPU side of things for AI workloads. Zen 5 will make an appearance on the desktop with the upcoming Ryzen 9000 series “Granite Ridge” desktop CPUs in August.
AMD is using a monolithic die for its Strix Point APU based on TSMC’s newer 4nm process node packing up to 12 cores and 24 threads of a mix of Zen 5 and Zen 5c cores. AMD splits maximum power to the full Zen 5 cores, and maximum efficiency to the Zen 5c cores. We have up to 5.1GHz boost, 36MB of cache, 50 TOPS NPU, and 16 CUs of RDNA 3.5 GPU.
RDNA 3.5 receives some enhancements in the form of optimized performance per watt, optimized perf/bit, and better power management that leads Strix Point APU-powered laptops to enjoy longer battery life.
AMD RDNA 3.5 GPU features:
- 2x Texture Sampler Rate: The Subset of the most common texture sampling operations is now double rate for common game texture ops.
- 2x Interpolation and Comparison Rates: Most of the rich vector ISA for interpolation & comparison is now double the rate for common Ops in shaders.
- Improved Memory Management: Improvements to the primitive batch processing to reduce memory access, better compression techniques, workload reduction and optimized LPDDR5 access.
AI is everything now, and now AMD owns the Copilot+ PC ecosystem with its Ryzen AI 300 series “Strix Point” APUs and their upgraded XDNA 2 NPU that’s capable of 50 TOPS. THis beats out Qualcomm’s new Snapdragon X series processors, and will beat what AMD estimates Intel will deliver with its upcoming Core Ultra 200V series “Lunar Lake” processors with 45 TOPS.
7AMD is launching just two of its Strix Point SKUs today: the Ryzen AI 9 HX 370 and Ryzen AI 9 365. The Ryzen AI 9 HX 370 features 4 x Zen 5 cores and 8 x Zen 5c cores with up to 5.1GHz boost clocks, 36MB of cache (24MB L3 + 12MB L2) and the RDNA 3.5-based Radeon 890M GPU with 16 compute units or 1024 cores. The XDNA 2-based NPU pumps out an industry-leading 50 TOPS for AI workloads.
That’s a rather large 50% increase in cores and threads, 33.3% more compute units, and 3.12x the NPU performance over the previous-gen flagship Ryzen 9 8945HS, making AMD’s new Ryzen AI 300 series “Strix Point” APUs an impressive feat.
7AMD says that its new XDNA 2 NPU features enhanced AI tiles that provide 2x the multi-tasking capabilities for increased compute and that the new NPU is twice as efficient with specific improvements for GenAI workloads.
7This means that the XDNA 2-based NPU is 5x faster than last-gen offerings in LLMs, with this not even being the best performance out of Strix Point. There’s the higher-end Ryzen AI 9 HX 375 that features 55 TOPS (up from 50 TOPS) from its NPU for AI workloads, already spotted inside of the upcoming HP OmniBook Ultra laptop.
AMD’s second Strix Point SKU launching today is the Ryzen AI 9 365 (non-HX) with 10 cores and 20 threads that boost up to 5.0GHz, 34MB of cache, and an RDNA 3.5-based Radeon 880M GPU with 12 compute units or 768 cores. Inside, we’ve got 4 x Zen 5 cores and 6 x Zen 5c cores, with the same NPU performance at 50 TOPS from the XDNA 2-based NPU for AI workloads.