Rumor mill: Intel's Nova Lake client platform is shaping up to be as much a GPU-focused update as a CPU overhaul. The company is standardizing on its Xe3 graphics architecture while reserving the newer Xe3P blocks for select high-end SKUs targeting heavier workloads and AI acceleration. This shift helps clarify months of conflicting leaks regarding how much next-generation graphics IP Intel will actually bring to the Core Ultra 400-series refresh.
According to longtime Intel watcher Jaykihn, Nova Lake's integrated graphics will be built around Xe3, the same generation used in Panther Lake integrated GPUs. Jaykihn had previously suggested that Nova Lake would include an Xe4 media component but has since walked that back, stating that "there is nothing Xe4 on Nova Lake" and that the blocks he had previously labeled as Xe4 "are actually Xe3P."
The updated information aligns with internal Linux driver work, which indicates that Nova Lake combines two graphics-related IP blocks: the baseline Xe3 and the more advanced Xe3P – also known by the codename Crescent Island – for display and media functions.
These driver updates and leak revisions close the door on expectations that Nova Lake might introduce Xe4 "Druid" IP into Intel's integrated stack ahead of schedule. Instead, Intel appears to be betting on a refined, higher-volume deployment of Xe3 while using Xe3P selectively.
– Mofumofu (@Mofumofu_Ship2) April 4, 2026
VideoCardz describes a Nova Lake lineup split across U, H, S, HX, and UL segments, with most configurations sticking to Xe3. Nova Lake-H, however, is slated to ship with a 12 Xe3P configuration as the successor to Panther Lake's 12-Xe3 design. In practical terms, this suggests that Intel is focusing on incremental gains and broader enablement rather than making a disruptive change to its graphics architecture at this stage of the roadmap.
On the CPU and platform side, Nova Lake remains a substantial refresh of Intel's hybrid P-core and E-core design, targeting late-2026 availability with volume ramping into early 2027. The platform is designed to support DDR5-8000 memory at stock settings, without overclocking profiles, pointing to a more capable integrated memory controller than the one found in today's Arrow Lake Refresh parts.
Current controllers top out around DDR5-7200 support; pushing the certified ceiling higher should help keep the GPU and AI engines better fed, especially in systems without discrete graphics where bandwidth is shared.
Power budgets are also increasing. The flagship Nova Lake configuration is expected to scale to 52 cores with a 175W TDP, alongside 65W variants for more mainstream builds. This represents a notable jump from the 125W base TDP on the current Core Ultra 9 285K and its refresh but aligns with what is needed to sustain higher all-core boost clocks and more aggressive integrated GPU and NPU utilization.
For AI workloads, Intel is targeting more than 100 TOPS of INT8 performance on Nova Lake by combining an onboard NPU with the Xe3 graphics engine. This signals that local inference and mixed CPU – GPU – NPU pipelines are central design goals for this generation rather than optional extras.