Microsoft roadmap leak reveals AMD Ryzen AI 500 and Intel Core Ultra 4 CPUs
Microsoft has shared its plans for next-gen Windows PCs, and in the process revealed the names of upcoming AMD and Intel processor series, plus the product categories these chips will cover in its lineup. Called Hybrid Intelligence, this roadmap includes four device tiers, ranging from budget 8GB laptops to supercomputers powered by Nvidia’s GB300 platform.
A leaked slide from the event shows two unreleased CPU families. The first is AMD’s Ryzen AI 500 lineup, which is possibly based on the Zen 6 architecture. The second is Intel’s Core Ultra 4 range of CPUs, based on the Nova Lake architecture. There is even a mention of Intel’s upcoming Crescent Island GPU (officially branded as the Intel GPU D-Series). Interestingly enough, Microsoft seems to have split its product tiers based on their total system memory rather than raw processing power.
Tiered Intelligence: From 8GB to 480GB
Microsoft’s strategy relies on a four-tier system that uses total system memory as the primary differentiator for AI capabilities. This approach suggests that the company views memory bandwidth and capacity as the true bottlenecks for local AI tasks, rather than just the theoretical TFLOPS of the processor.
Entry-level and Copilot+ Tiers
At the bottom of the stack, Tier 0 targets budget-friendly laptops with 8GB of memory. These systems will utilize AMD Ryzen 100 and 200-series CPUs, Intel Core Series 3 and 4, and Qualcomm Snapdragon C/X/X Plus SoCs. Microsoft positions these for 'local context and actions,' essentially basic tasks that require minimal compute. This follows recent claims from Microsoft that Windows 11 multitasking performance on 8GB devices has been improved by up to 40%.
Tier 1 introduces Copilot+ certified devices, split into three sub-categories based on RAM. The 16GB and 32GB models will feature Intel Core Ultra Series 3/4, AMD Ryzen AI 400/500, and Qualcomm Snapdragon X2 Plus/X Elite/Extreme processors. For the 64GB segment, the roadmap lists AMD’s Ryzen AI Max+ 400, Nvidia’s RTX Spark N1X, and an unnamed Intel Core Ultra Next chip, marking the threshold where local AI processing becomes a primary focus.
High-End AI Workstations
The highest tiers, Tier 2 and Tier 3, are aimed at advanced users running large language models (LLMs) locally, requiring 128GB of memory or more. Tier 3 is particularly ambitious, listing Intel’s Crescent Island (D-Series) GPUs based on the Xe3P architecture. These high-end configurations are described as having up to 480GB of LPDDR5X memory, providing 1.5TB/s of bandwidth and 1.3 petaflops of FP8 compute power—specs theoretically capable of handling models with up to 300 billion parameters on a single card.
Гика
