Intel Core Ultra 9 285K CPU Review

The Intel Core Ultra 9 285K is one of the flagship desktop SKUs based on the Arrow Lake-S family. This is a different beast from the previous monolithic Core i9 chips. Instead of being just one big slab of silicon, the 285K employs a multi-tiled package which uses Intel’s Foveros 3D packaging, with separate dies for compute, graphics, and I/O functionality. Intel’s goal is obvious: with a performance-critical set of dies and using the most advanced process node available, Intel switched it up for LGA1851; it aims to compete with AMD by using Foveros 3D packaging.

A breakdown of the individual tiles and process node for Intel’s Core Ultra 9 285K shows that it is architecturally superior to its previous chips. Here’s a quick breakdown of the tiles themselves:

• Compute tile (CPU Cores and Cache) is built on TSMC N3B
• SoC tile (memory controller, NPU, media, display, and platform logic) is built on TSMC N6
• I/O tile (extra PCIe, connectivity plumbing) is reportedly also built on TSMC N6
• Graphics tile (small iGPU) is built on TSMC N5
• Base tile/active interposer is built on an older 22nm class Intel node, which is used for dense die-to-die wiring designed for functionality, not performance

The most important point here is that Intel isn’t just using chiplets for no reason; the 285K’s lithography is listed as TSNC N3B, which is Intel’s way of saying tiles do matter, especially on more advanced nodes with advanced CPU packaging, with Foveros 3D taking center stage. The compute tile, which houses all of the cores, including 8 x Lion Cove performance (P) cores, with 16 x Skymont E-Cores arranged in alternate rows for better heat and latency management. All of the cores share a big 36MB L3 cache block, which moves away from hyper-threading foir a thread per core approach.

Focusing on the most important tile, compute, this is where the real performance uplifts come from. Intel is using 8 x Lion Core P-cores, with 16 x efficient running Skymont E-cores. Hyper-threading itself has been eliminated, which shows Intel’s focus on one core per thread for better performance and latency throughput. Instead of inflating core count with HT, Intel’s opting for more advanced packaging and nodes, rather than old monolithic ideas with more threads.

Intel has gone hard on cache for this generation, with 36MB of Intel Smart L3 cache, and a large 40MB of L2 cache too. Arrow Lake uses Intel’s typical approach of distributing L3 across slices along the ring interconnect, rather than using a single monolithic slab as in previous iterations. Latency depends heavily on how long the ring is and how many stops it has to traverse, but that trade-off shows up with Arrow Lake being designed to improve on bandwidth, but with higher memory and cache access latency, which is usually expected from a high-performance desktop chip.

It’s important to consider Intel using TSMC’s advanced N3B for its compute tile. For many years, Intel’s desktop chips were constrained by process and power usage. Moving to Arrow Lake’s compute tile on TSMC N3B, this is designed to enable a claw back of efficiency overall; Intel hasn’t been known for efficiency on its high-end desktop process for years, with small TDP claims, but real big power pulls when in full gear and boosting off the charts with high frequencies such as 5.4 GHz. Arrow Lake is positioned to enable high boost clocks of up to 5.7 GHz, but without the chip acting as a space heater the moment it ramps up; Intel positions the architecture as a shift towards better performance per watt, and its silicon choices certainly reflect that focus.

Arrow Lake uses a base tile, often described as an active interposer that sits beneath the functional tiles. Instead of routing everything through longer on-package traces, Intel uses extremely dense die-to-die wiring in this base layer, enabling the compute, SoC, and graphics tiles to talk to each other. The downside is packaging complexity, with extra silicon dedicated to glue logic and interconnect, and a constant risk that low-latency workloads will get stuck paying for Intel’s architectural ambition.

The end result is a hub-and-spoke layout: compute and GPU tiles talk into the SoC tile, and the whole package behaves more “monolithic” in bandwidth terms than a typical chiplet CPU.

With 8 Performance-cores and a staggering 16 Efficient-cores, resulting in a total of 24 cores and 24 threads, with a boost clock of up to 5.7 GHz and an expansive 36MB L3 cache, the 285K is a pretty serious bit of kit. Of course, more cores and higher speeds will also result in more heat. However, with Intel’s strong focus on efficiency for these 1st Gen Core Ultra CPUs, we’re not expecting it to be the thermal time bomb that previous i9 CPUs tended to be.

While the Core Ultra 5 and the Core Ultra 7 are strong options for both gaming and daily workflow, the Core Ultra 9 is here for the enthusiast users who want the best performance possible. With more cores, and bigger clock speeds, it should dominate for all workflows and will have clear benefits for content creators. But of course, let’s put that to the test and find out!

Specifications

• Cores and Threads: 18 cores (8 performance-cores + 10 efficient-cores) with 36 threads. This is a significant jump in core count compared to the Ultra 5.
• Cache: A large 36 MB Intel Smart Cache. More cache generally means faster performance.
• Clock Speed: Blazing fast speeds with up to 5.70 GHz clock speed, thanks to technologies like Intel Thermal Velocity Boost.
• Power: 125W base power with a maximum turbo power of 250W. This chip will likely need robust cooling.
• Graphics: Integrated Intel UHD Graphics with 4 Xe-cores. Similar to the Ultra 5.
• AI: Intel AI Boost for accelerated artificial intelligence tasks.
• Memory: Supports up to 192 GB of DDR5 6400 MT/s memory. Same as the Ultra 5.
• Socket: Uses the same FCLGA1851 socket as the Ultra 5.

Intel Core Ultra Series

Intel’s Core Ultra CPUs are here at last, marking a big change in the Intel naming structure, albeit not a particularly slick one. These are technically the 15th Gen Intel Core CPUs, but they’re the first generation of the new Intel Core Ultra series! These processors codenamed “Arrow Lake-s” mark the debut of Intel’s new nomenclature, replacing the familiar “i” series with the “Ultra” branding. This change signifies more than just a name update; it reflects a fundamental change in architecture and focus, a new starting point for future generations of CPUs, with the first gen having a significant focus on thermal performance and power efficiency for Intel to build upon.

AI and Beyond

One of the most prominent features of the Core Ultra series is its emphasis on artificial intelligence (AI). These CPUs are the first to feature a dedicated Neural Processing Unit (NPU), a specialized hardware component designed to accelerate AI tasks. This allows for significant performance gains in AI-powered applications, ranging from content creation and gaming to productivity and security. Intel claims that the Core Ultra CPUs deliver up to 50% faster performance in AI-enabled creator applications compared to competing processors.

Power Consumption and TDP

The new architecture features a hybrid design of performance cores (P-cores) and efficient cores (E-cores), similar to what we’ve seen on previous CPUs, which can deliver up to 14% faster multi-threaded performance compared to the previous generation. Additionally, Intel has achieved a 40% reduction in package power, which is great news, as the previous gens were notoriously power-hungry and had lofty cooling requirements. The Core Ultra series also introduces a new platform and socket (LGA1851), requiring users to upgrade their motherboards to accommodate these CPUs, which is why today we also have a series of new motherboard reviews for you to read.