Intel Nova Lake-H leaked specs reveal 16-core CPU and 25W TDP for Core Ultra Series 4
Intel is preparing a massive architectural pivot with its upcoming Nova Lake processors, aiming to reclaim the productivity crown through aggressive core counts and a redesigned cache hierarchy. Codenamed Core Ultra Series 4 (or Core Ultra 400), these chips are expected to arrive by late 2026, positioning the platform as a powerhouse specifically tailored for the demands of agentic AI tasks.
Recent leaks provide a detailed look at the mobile side of this rollout, specifically a Nova Lake-H SKU spotted on an Advantech board. This particular mobile processor utilizes a 16-core configuration consisting of eight performance cores, four efficient cores, and four low-power efficient cores. The chip is reportedly clocked at a 2.0 GHz base and 4.8 GHz turbo, supported by 18 MB of last-level cache. Perhaps most striking is the memory support, as the leak indicates compatibility with DDR5-8000 MT/s and capacities up to 128 GB, representing a significant jump over the DDR5-7200 support found in Panther Lake.
Despite the high clock speeds, this specific Nova Lake-H variant is rated for a 25W TDP, suggesting a focus on efficiency for mainstream mobile devices. The integrated graphics are expected to utilize the Xe3P architecture, with the leaked SKU featuring a 2.3 GHz clock speed. While official confirmation is pending, higher-end variants in this family may carry up to 12 GPU cores.
Desktop Architecture and Platform Shifts
The transition to the desktop segment, known as Nova Lake-S, suggests an even more aggressive approach to performance. Leaks indicate a wide range of core counts, with variants spanning from 8 cores up to a massive 52-core configuration. Early samples spotted on OpenBenchmarking show a 28-core chip reaching 5.2 GHz turbo boost, which may carry a PL2 power limit as high as 296W. To support these demands, Intel is reportedly introducing the LGA 1954 socket, with an emphasis on forward socket compatibility to extend the platform's lifespan.
To compete with high-cache competitors, Intel is adopting a "Big Last Level Cache" (bLLC) architecture. This shift is paired with a broader expansion of I/O capabilities, including integrated Wi-Fi 7 and support for up to eight SSDs across PCI Express 4.0 and 5.0. The platform is also expected to debut integrated Thunderbolt 5, offering bidirectional speeds of 80Gbps and potential single-direction speeds of 120Gbps, alongside support for four independent display outputs.
Instruction Sets and AI Integration
Beyond raw throughput, Nova Lake marks the return of 512-bit AVX-512 vector instructions to consumer chips, a feature that had been absent since the 11th-generation Rocket Lake processors. Because hybrid architectures require a uniform instruction set across all cores to prevent thread migration errors, AVX-512 is expected to be implemented in both the Coyote Cove performance cores and the Arctic Wolf efficiency cores. This return is particularly relevant for scientific calculations, engineering workloads, and local AI acceleration.
This focus on AI is further bolstered by the inclusion of an NPU 6. By combining the NPU with high core counts—such as the 52-core variant—Intel is targeting a future where "agentic AI" allows users to run complex AI agents in the background while maintaining heavy productivity workloads in the foreground. Power consumption across the broader Nova Lake family is expected to range from 35 watts to 175 watts, with the highest-end desktop models potentially offering GPU-less configurations.
The convergence of AVX-512, bLLC, and Thunderbolt 5 suggests that Intel is moving away from incremental updates in favor of a foundational reset. If these leaks hold true, the Core Ultra Series 4 will represent a shift toward a high-margin, AI-centric productivity tool, though the ultimate success of the platform will depend on Intel's ability to resolve current manufacturing limitations and successfully deploy the new LGA 1954 ecosystem.

