Hot Chips 2026: Intel Wildcat Lake and the Rise of Heterogeneous Computing

At Hot Chips 2026, Intel presented Wildcat Lake, the architecture behind Intel Core Series 3. Designed to maximize value for mainstream systems, the processor takes what Intel calls a “right-sized” approach to compute, balancing CPU, GPU and NPU capabilities around the workloads the platform is designed to serve.

That approach highlights a broader shift in processor architecture: different types of compute increasingly work together within the same system, each handling the workloads they are best suited for.

Different workloads, different compute.

Wildcat Lake combines CPU cores, integrated Xe graphics and an NPU delivering up to 17 TOPS, with Intel positioning the overall architecture for AI-ready systems spanning PCs and edge applications.

The Q&A following the presentation made this division particularly clear.

When asked whether Intel had changed the CPU architecture specifically for AI inference, the presenter distinguished between different parts of AI workloads. CPU cores can handle areas such as orchestration, tool use and conventional CPU workloads, while inference can benefit from specialized acceleration. Wildcat Lake's NPU, meanwhile, is designed for background AI workloads such as video conferencing effects and semantic indexing.

It is not a workload running on one type of processor: CPU, GPU or NPU. Increasingly, computing systems divide work between different architectures according to the characteristics of the task.

Industry commentary around Hot Chips reflected the same trend. Computerworld's analysis of the event highlighted analyst observations that enterprise AI infrastructure is becoming increasingly heterogeneous, with workloads moving between CPUs, GPUs and specialized processors as system architectures evolve.

Adding another capability alongside the CPU

Flow approaches this heterogeneous computing landscape from the CPU side.

The Flow Parallel Processing Unit® (PPU) is designed to work alongside conventional CPU cores, adding closely coupled acceleration for scalable general-purpose parallel computation. The CPU continues handling sequential execution, control and existing software, while suitable parallel sections can execute on the PPU.

This means the PPU is not intended to replace CPUs, GPUs or other specialized accelerators. Instead, it adds another complementary capability to the system.

A heterogeneous architecture could therefore include CPU for general-purpose and sequential execution, GPU or NPU for specialized accelerated workloads, and PPU for suitable general-purpose parallel workloads alongside the CPU.

Architecture is about trade-offs

Wildcat Lake also provides an interesting example of how these architectural choices extend beyond individual compute engines.

Intel also discussed the cost and design trade-offs behind Wildcat Lake’s packaging. To meet its cost targets, Intel moved away from the Foveros packaging approach used in higher-end designs and selected a two-die organic multi-chip package, using UCIe for die-to-die communication. The decision illustrates how processor architecture involves balancing performance, integration and cost rather than maximizing any single characteristic.

More broadly, it reflects a shift towards optimizing how different resources work together rather than maximizing any single specification.

Flow PPU follows a similar principle at the workload level. Increasing conventional CPU resources alone does not always translate into proportional parallel performance because memory access, synchronization and communication can introduce scaling overhead.

Flow PPU adds a complementary architecture alongside the CPU, allowing suitable general-purpose parallel workloads to execute on an architecture designed for scalable parallel computation.

CPU + the right architecture for the workload

Wildcat Lake's CPU + GPU + NPU design is one example of where computing is heading: heterogeneous architectures designed around different workload requirements.

Flow extends that idea with another possibility, adding closely coupled parallel acceleration alongside the CPU.

As workloads become more diverse, the question may increasingly shift from which processor architecture is best? to which combination of architectures is best for the workload?

CPU + GPU + NPU + Flow PPU each have different roles to play, and the future of computing may depend increasingly on making those architectures work together.

Explore Flow's architecture and latest performance results →