AMD Looks Set to Sacrifice Integrated Graphics for AI in Future Ryzen Processors

In its next-gen desktop Ryzen processors, AMD is reportedly planning to swap out the integrated graphics for a dedicated AI accelerator. We're talking about the Ryzen 10000 series, built on the Zen 6 architecture and expected to land sometime in 2027.

These chips will make the jump to TSMC's N2P process node, but the headline change isn't hidden in the compute cores—it's in the I/O die (IOD). According to the leak, that's exactly where AMD will move the NPU for local AI task acceleration. At the same time, the IOD will lose its minimal Radeon Graphics video adapter, a two-CU affair that has been a standard feature on every regular AM5 "rock" since the Ryzen 7000 generation.

This move makes pragmatic sense for AMD, but for end-users, it's a decision that brings tangible inconvenience. Previously, even the simplest integrated video chiplet allowed you to boot your system without a discrete graphics card, run diagnostics, update the BIOS, or do an initial system setup. Now, for any kind of display output, you'll absolutely need to have a separate graphics card on hand—even if your main one bites the dust. The new CCD chiplet will pack up to 12 Zen 6 cores and 48 MB of L3 cache. By combining one or two of these chiplets, AMD will be able to offer a wide range of configurations—from 6 to 24 cores, all retaining SMT multithreading. The list includes familiar 6- and 8-core parts, as well as intermediate 10-core models on a single CCD, and dual-chiplet variants boasting 16, 20, and 24 cores. The 3D V-Cache versions aren't being forgotten either; they're also slated to appear in the lineup.

The AM5 platform will maintain physical compatibility, but new motherboards are in for a serious upgrade. They'll come with Wi-Fi 7 support, expanded I/O capabilities, and a full suite of EXPO 1.2 profiles. The latter is particularly important for working with DDR5 CUDIMM memory modules, which have their own built-in clock generator, making it easier to hit frequencies beyond 8000 MHz and achieve low-latency operation via ULL profiles.

The release of Olympic Ridge is pegged for roughly the same timeframe as the debut of Intel's desktop Nova Lake-S. The two companies will once again offer fundamentally different approaches: Intel, according to early reports, is doubling down on a hybrid configuration with up to 52 total cores, while AMD is focusing on the NPU as the platform's killer feature, alongside maintaining AM5 compatibility. The tangible benefit of an NPU in a classic desktop PC isn't immediately obvious, especially when stacked against the utilitarian value of a diagnostic video output.

How do you view this reallocation of resources in favor of the NPU—a rational step toward the future, or a detrimental blow to basic functionality? Let us know your thoughts in the comments.