Inside Intel Tech Tour 2025: Panther Lake And Clearwater Forest Constructed On 18A

We not too long ago had the chance to attend Intel’s Tech Tour 2025 and dive deep into the corporate’s upcoming Panther Lake and Clearwater Forest CPU architectures. We additionally bought the prospect to tour Intel’s Ocotillo campus in Chandler, Arizona, which is the positioning of Fab 52 and Fab 62, the place high-volume chip manufacturing utilizing superior Intel 18A course of know-how is being executed. There was an abundance of knowledge shared at ITT – an excessive amount of to function in a single article. Panther Lake and Clearwater Forest, nevertheless, would be the cornerstones of Intel’s shopper and information middle choices within the coming months and each of them are partially manufactured on Intel Foundry’s modern 18A fabrication course of, so we’re going to deal with these three issues right here.

Intel 18A: Touring Intel’s Ocatillo Campus

We weren’t allowed to take any photos and video throughout our fab excursions on the Ocatillo campus, however wished to spend a while making an attempt to offer you a way of scale at this plant, and drive residence the purpose that Intel is arguably manufacturing among the world’s most superior chips proper right here within the U.S. – TODAY.

Intel’s Ocatillo campus is located on greater than 700 acres of land and incorporates a multitude of buildings and amenities. There are a number of wafer fabrication crops on-site, utility buildings that home chillers and boilers, and wastewater therapy crops, and a large number of different buildings with all the pieces from cafés to emergency mills. To learn or hear information about Intel’s Ocatillo campus is one factor, however to expertise in particular person is one thing solely totally different. The Ocatillo campus is actually a village or small metropolis. It’s large.

The buildings are all intertwined and leverage shared methods which might be vital to manufacturing chips of every type. The main target of our tour was 18A, so we donned our bunny fits, walked by a few superior clear rooms and bought to see the fab in motion. Though the transistors manufactured on 18A are among the smallest on this planet, all the pieces within the fab is massive. Even the dimensions of the ASML Excessive-NA EUV lithography instruments dwarf the older EUV instruments. The dimensions of the power must be seen first-hand to be appreciated.

On the scale Intel requires, the huge footprint and useful resource necessities of the Ocatillo campus are troublesome to fathom. Intel has a webpage obtainable detailing among the energy and water consumption particulars and different environmental issues and when you test it out, you’ll see what I imply. However what’s equally fascinating is that Intel has put in a ton of labor to make sure the encircling communities are minimally impacted. The campus generates tens of millions of kilowatt-hours of electrical energy from photo voltaic panels, which it pumps out into the native grid as properly. It is also geared up with waste-water reclamation amenities, and solely 4% of the waste generated truly makes it to the landfill.

Exploring Panther Lake: CPU Structure Deep Dive

At this level, when you comply with the processor house in any respect, you’re in all probability properly conscious of what Intel’s Panther Lake is all about – we’ve been speaking about it and posting leaks and different information, together with quite a few photos of the chips, for properly over a 12 months. To summarize Intel’s official messaging on Panther Lake although, it’s an evolution of Lunar Lake that leverages Intel’s superior 18A course of know-how, with extra versatile reminiscence configurations, designed to scale efficiency and effectivity throughout all its processing engines (CPU, NPU, IPU, GPU).

Like the previous couple of processor generations, Panther Lake is a disaggregated design, composed of a number of tiles, that every one occur to be manufactured on totally different processes, linked collectively by way of a base die on a single package deal utilizing Intel’s superior Foveros-S 2.5D packaging know-how. The compute tile, which is residence to the CPU cores and media engine, is constructed on Intel’s superior 18A course of, the GPU tile is constructed on Intel 3 or by TSMC, relying on the actual configuration, and the platform controller tile can also be manufactured at TSMC.

Though the first marketplace for Panther Lake can be next-generation AI laptops, it’ll seem in a wide range of designs and type components, at a myriad of worth factors, from robots and level of sale methods, to laptop imaginative and prescient methods and edge AI IoT units.

8 And 16-Core Intel Panther Lake CPU Variants

Out of the gate there can be three primary Panther Lake chip configurations, an 8 CPU core variant (4P, 4LPE) with a 4-core GPU tile, a 16 CPU core variant (4P, 8E, 4LPE) with the identical 4-core GPU tile, and a top-end 16 CPU core mannequin (additionally 4P, 8E, 4LPE) with a a lot increased performing 12-core GPU tile. The platform controller tile on all three configuration is generally related, and gives as much as 20 PCIe lanes (8x Gen 4, 12x Gen 5), 4x Thunderbolt 4 posts, 10x USB ports – cut up between USB 2 (8x) and USB 3.2 (2x), Wi-Fi 7, and Bluetooth Core 6.0. The one actual distinction between the platform controller tiles is that every one 20 lanes can be found on the mid-level 16-core Panther Lake configuration, whereas solely 12 lanes can be found on the entry-level 8-core and prime 16-core / 12-Xe3 fashions.

The three primary Panther Lake chip configurations additionally differ when it comes to their reminiscence assist. The 8-core mannequin can use both DDR5 or LPDDR5X reminiscence, at speeds as much as 6,400MT/s or 6,800MT’s, respectively. The mid-level 16 core mannequin will use DDR5 or LPDDR5X reminiscence, however at speeds as much as 7,200MT/s or 8,533MT/s. And the top-end 16-core / 12-Xe3 mannequin will use 9,600MT/s LPDDR5X reminiscence solely. Intel truly has a strict requirement in place to pair the 16-core / 12-Xe3 variant with solely 9,600MT/s reminiscence, to make sure most efficiency, although that reminiscence will be soldered down on the PCB or use LPCAMM modules. We must also be aware that the utmost reminiscence capability for the 8 and 16-core fashions is 128GB, however the 16-core / 12-Xe3 tops out at 96GB.

Panther Lake CPU Core Upgrades: Cougar Cove And Darkmont Unveiled

Panther Lake leverages Intel’s newest CPU core designs. The “massive” high-performance P-cores are based mostly on Cougar Cove and the “small” power-efficient E- and LPE-cores are based mostly on Darkmont. At their foundations, theses aren’t completely new CPU core microarchitectures – Cougar Cove is an evolution of Lion Cove and Darkmont is an evolution of Skymont – however they’re each optimized for 18A and have IPC enhancements and supply elevated efficiency over previous-gen designs at decrease energy. Intel is claiming as much as 10% higher single-thread efficiency, with 50% higher multi-thread efficiency at related energy ranges to Lunar Lake, or related MT efficiency at 30% decrease energy. With the highest finish 12-Xe3 core GPU mannequin, Intel can also be claiming and 50% extra GPU efficiency.

The width or depth of Cougar Cove is unchanged versus the previous-gen Lion Cove, however there have been enhancements made to enhance efficiency and effectivity. Panther Lake may have 16MB of L2 and as much as 18MB of L3 cache, which is a large, 50% enchancment, together with improved heuristics, an up to date thread director, and extra clever AI based mostly energy administration.

There are additionally TLB enhancements with 1.5X the capability for contemporary workloads. Improved department prediction, and utilizing the superior 18A course of gave intel the power to develop some inner constructions to scale back latency. The Department Prediction Unit has been refactored for effectivity as properly, and has improved accuracy, decrease latency, and extra prediction bandwidth over Lunar Lake.

Darkmont, nevertheless, is wider and deeper than Skymont, and gives 26 dispatch ports, extra vector all through, extra L2 bandwidth, improved nanocode efficiency, and higher, dynamic AI based mostly prefetching just like the P-cores – the design can basically resolve if it needs to do kind of perfecting utilizing clever algorithms, relying on the workload being run on the time.

The Darkmont cores are organized in two clusters, with as much as 8 customary E-cores and 4-LPE cores – the microarchitecture is identical for each E and LPE cores.

Versus Lunar Lake, the LPE cluster is a giant improve, and it now options 4MB of shared L2 cache. The LPE cluster not solely gives extra efficiency, however it helps a wider vary of duties over Lunar Lake, whereas sustaining low energy. Panther Lake additionally provides a reminiscence facet cache, with 8MB of bodily cache, to lowered DRAM visitors, which in flip reduces energy, enhances latency, and supplies further caching for the IO engines. The 8 customary E-cores additionally reside on the L3 cache ring, which isn’t the case on Lunar Lake.

Intel Thread Director And Energy Optimization Revamped For Panther Lake

As a consequence of the entire enhancements made throughout the varied CPU cores, Intel additionally revamped Thread Director in Panther Lake. In case you recall, Thread Director basically classifies the LPE, E, and P-core capabilities and supplies dynamic, real-time suggestions tables to the working system. The suggestions desk updates dynamically based mostly on the actual workloads which might be in-flight, to optimize scheduling selections. On Panther lake, Thread Administrators algos are enhanced and optimized to supply higher thread steering throughout a wider set of contemporary use circumstances. Intel optimized the core classification modules to raised decide which workload needs to be dispatched to a specific kind of core, and in addition improved energy administration enter, and expanded busy use case protection. Panther Lake’s design additionally permits for simultaneous execution throughout P, E, and LPE cores.

As a result of the E and LPE cores are extra performant in Panther Lake, effectivity is paramount, and Thread Director will usually begin with essentially the most environment friendly LPE cores if the work suits, and if the workload exceeds the capability, Thread Director strikes on to the E-cores, or in the end to P-cores.

Intel’s deal with effectivity with Panther Lake additionally ends in some adjustments to how OS energy modes work. Panther Lake will supply what Intel calls an Clever Expertise Optimizer, which consists of a number of firmware and software program enhancements. Although the normal energy mode slider in Home windows will perform equally, if a system is about to Balanced, the Clever Expertise Optimizer will mechanically lean in direction of providing extra efficiency relying on the workload, however when it detects the necessity for extra effectivity, it’ll transfer in that route. Intel improved the telemetry for the SoC so it will probably make higher energy administration selections. These algos additionally transfer from software program to {hardware} for tighter, extra responsive management and to supply a extra constant unplugged versus wall-powered expertise. Intel claims these dynamic optimizations will assist shut the AC / DC energy delta we usually see on x86 based mostly notebooks and supply an approximate 19% uplift versus baseline DC efficiency.