Introduction

Indeed, it is that time of the year when the highly anticipated series of desktop processors from Intel, codenamed Arrow Lake, is finally released for public consumption. In this era of technological advancements, the ecosystem is poised for progress, starting with upgraded motherboards (the Z890 chipset), faster clocked RAM, and cutting-edge m.2 NVME drives (Gen 5). This period is characterized by tireless efforts, building new platforms, optimizing, and benchmarking. Tech enthusiasts like us thrive on this excitement, and when combined with the approaching festive season in India, the result is an exhilarating, adrenaline-fueled time of heightened activity.

Arrow Lake CPUs: Unveiling the Core Ultra 200S Series

Behold, Intel presents its latest Arrow Lake CPUs, carrying the brand name “Intel Core Ultra” alongside Intel’s Lunar Lake laptop CPUs. For desktops, this is Intel’s inaugural series bearing the “Intel Core Ultra” nomenclature, accompanied by a novel numbering scheme. Intel has departed from the traditional “Intel Core” branding and the subsequent XX-XXX numbering of its CPUs. Embracing a simplified naming convention, Intel now employs the “XXX” designation, assigning 285K, 265K, and 245K to the Ultra 9, Ultra 7, and Ultra 5 chips, respectively. Furthermore, Intel has eliminated the “i” prefix that previously denoted its portfolio, resulting in the discontinuation of i9, i7, i5, and i3 branding.

The recent significant shift in naming conventions suggests a new direction for the organization. In earlier generations, the primary objective was consistently increasing raw performance with each new generation of products. This trend was evident in Intel’s journey from its 6th generation CPUs, when it introduced the “Core iX-XXXXX” naming scheme. With each successive generation, Intel pushed the boundaries of performance, resulting in increased power consumption and heat generation of the chips. However, for this new generation, Intel has a distinctively different vision:

• Increase performance per watt, with the target set to reduce package power by 40%.
• Deliver >15% generational multithreaded performance while maintaining gaming performance
• Expand AI acceleration
• Integrate Xe-LGP for latest media support
• Reduce package temperatures by ~10°C while gaming.

As we progress through this article, we will elaborate on each point. The first bullet point emphasizes efficiency, which is Intel’s primary motivation for this series. Intel’s decision to focus on efficiency is understandable. Over the past five to seven years, we have witnessed an increase in the size of power supply units and CPU coolers. This trend has reached a point where a 360mm water-cooled system is almost a prerequisite to fully utilize the potential of i7 and i9 chips. Consequently, cabinets have also become larger. To further complicate matters, motherboard manufacturers have opted for their own unique CPU adjustments, leading to difficulties for PC builders in ensuring consistent performance from their 13th and 14th-generation systems.

We applaud Intel’s decision to shift its focus toward efficiency, which, in our opinion, is the most significant improvement offered by the new chips. Frankly, there was no other viable path for Intel to restore its reputation. This is the right approach, and we are pleased to see it become a reality.

Intel Core Ultra 200S – Architecture

Let’s explore the architectural changes Intel has implemented in the Core Ultra 200S series compared to Raptor Lake to achieve its goals. Intel divides its cores into Performance-cores (P-core) and Efficiency-cores (E-core). The P-cores are meant to handle heavy tasks and have high power limits. The E-cores are meant to handle light tasks with considerably lower power limits. Why Intel uses two different types of cores – efficiency. For a low powered task, a fully equipped core (P-core) will use high power and generate more heat, so it becomes more efficient to just use the E-core to handle light loads.

On the external, Intel is using the industry leading FOVEROS 3D packaging, with a layer-based configuration for each module of the chip. The new chips get native Wi-Fi 6E support, integrated Thunderbolt 4, PCIe Gen5 support, built-in NPU, better DDR5 support and a new silicon security system. Intel claims that it has integrated an AI-based power management system in the 200S series. Clock intervals has been reduced to 16.67MHz for finer tuning. The Intel Core Ultra 200S CPUs now also support Modern Standy: Windows transitions the SoC from idle mode to active mode a minimum of every 30 seconds to perform kernel maintenance tasks. This maintenance activity is extremely brief in duration (typically no more than a few hundred milliseconds) and cannot be adjusted. Thus you can put your system on standby without worrying about excessive power consumption. Only Apple CPUs have had this feature, up until now.

With the Arrow Lake S series, Intel has rejigged the placements of its E-cores (codename: Skymont) and P-cores (codename: Lion Cove). Previously, the P-cores were located in its own colony, and E-cores in its own. Now, E-cores are distributed evenly between the P-cores (see diagram). This new floorplan brings in many benefits, like consistency in performance, better head spreading and more efficient power consumption. To give you an example: for an office productivity task, the scheduling starts off on the E-cores, and switches between the P-cores as more performance is needed. In principal, this is the technology that has resulted into upto 30% power usage decrease from the previous gen without dropping performance.

The cache hierarchy has been slightly adjusted too. Shared cache remains the same as last-gen, at 36MB. The allocation to P-cores has increased from 2MB to 3MB. A 1MB difference might not sound monumental but in hindsight, that is a 50% increase in cache size per individual P-core.

Intel claims that its P-cores have integrated AI-based power management systems in the 200S series. Clock intervals has been reduced to 16.67MHz for finer tuning. There is an increased multi-threading performance for the E-cores, even though the number of threads has gone down from the previous generation.

The E-Cores have been designed with the goals of increasing workload coverage, doubling the vector and AI throughput and increasing scalability. You can get the gist of the Arrow Lake CPUs: Intel hasn’t added much, they’ve simply refined what was already there. To draw a parallel, it is much like the latest version of the superbike Suzuki Hayabusa. The maximum horsepower has gone down from the previous generation of the bike, and with that some top-end. But what you get now is much refined version of the outgoing bike.

Integrated GPU – Intel Xe

With a new generation we also get a completely revamped integrated Graphics Processing Unit – “Intel Xe GPU”. This GPU has full DirectX 12 support from top to bottom. Intel has equipped the Xe GPU with upto 2x compute resources for a larger GPU configuration. The clock frequences have been increased significantly, and the performance per flop has also gone up. The Xe GPU consists of 4 Xe-cores meant to handle graphic intensive tasks. There are also four dedicated Ray Tracing unit, so you get proper Ray Tracing support. Add to that, 64 vector engines, 1 geometry pipeline, 4 samples, 2 pixel backends and 4MB dedicated cache memory fabric. All of this advancement can yield a up to 8 TOPS of raw processing power. That is roughly more than double of the i-GPU performance in Intel’s 14th Gen CPUs. These GPU advancements are significant and could be a game changer for those opting to not buy a dedicated GPU. One has to think about the extreme high prices of GPUs currently. You can count on Intel Xe to support the most modern of games.

NPU 3 – AI Architecture

With the Intel Core Ultra 200S comes Intel’s third generation NPU architecture. This architecture has 2 neural compute engines, 2 large INT8/FP16 MAC arrays, 4MB scratchpad RAM and 4 Shave DSPs. It can reach raw performance of up to 13 TOPS, and, as claimed by Intel, is the first NPU for enthusiasts. It supports tons of Creator and generative AI workloads, and coupled with the P and E-cores, can handle light and real-time AI with ease. The NPU allows the new Intel chips to have native support for apps such as Wondershare – AI video effects, Audacity – AI music generation, XSplit – NPU enabled AI background segmentation and auto-framing: without affecting the performance of the CPU.

Let’s talk about memory. The Intel Core Ultra 200S series can run DDR5 memory upto 6400Mhz. Per DIMM (per RAM stick), a maximum of 48GB can be supported. That means that the entire system can have a maximum of 192GB memory capacity. You do get ECC and Dual channel support.

Intel Core Ultra 200S Series Launch CPUs

The Arrow Lake, or Intel Core Ultra 200S series will launch with principally three chips, with each having their K and KF variants. More variants will be launched in the future, notably the Intel Core Ultra 3 chips.

	Total Cores	P Cores	E Cores	Total Threads	GPU Cores	TOPS NPU	Max GHz
Intel Core Ultra 9 285K	24	8	16	24	4	13	5.7
Intel Core Ultra 7 265K	20	8	12	20	4	13	5.5
Intel Core Ultra 7 265KF	20	8	12	20	0	13	5.5
Intel Core Ultra 5 245K	14	6	8	14	4	13	5.2
Intel Core Ultra 5 245KF	14	6	8	14	0	13	5.2

Intel 800 Series chipset

The 200S chips will require a new socket, so that means that you can only use them with new (and expensive) 800 series motherboards. This is the only sore move by Intel that has not sit well with us. The advancements that Intel has achieved in its new chip, along with its iterations, really do not require a completely new socket. We guess that this was a collective decision of Intel and the chipset manufacturers (ASUS, Gigabyte, MSI etc.), and a pure capitalist one at that. People with expensive Z790 motherboards will have to buy a new motherboard just to use the new Intel chips. They might also have to buy completely new cooling units. It’s not fair for them, that’s all. Most iterations are at best slightly incremental and don’t warrant a new socket.

The Intel 800 Series chips has a total of 48x PCIe lanes which includes 24x PCIe 4.0 lanes and 20x PCIe 5.0 lanes. You also get up to 32x USB 3.2, 5x 20G, 10x 10G and 10x 5G. On discrete systems you can now have 4x Thunderbolt 5 slots, Wi-Fi 7 (5.8Gb/s), 2.5GbE ethernet and Bluetooth 5.4.

Performance

Alright, that is more than enough theory for today. Now, let us dive deep into the performance of the Intel Core Ultra 9 285K, which is why most of us are here in the first place, isn’t it?

Test System

CPU	Intel Core Ultra 9 285K
Motherboard	Gigabyte Z890 AORUS PRO ICE
Cooler	Hyte Thick 240mm
RAM	Corsair 16×2 DDR5 RAM
GPU	ZOTAC GeForce RTX 4080 SUPER 16GB
Storage	WD_BLACK SN850X 2TB
PSU	Corsair 850E
Cabinet	Hyte Y70 Touch
Monitor	BenQ EX3210U 4K 144Hz
Keyboard	Logitech MX Master Keys S
Mouse	Logitech G Pro X Superlight 2

So you can see our test system is equipped with the cutting edge of technology. We would have love to add a Gen5 NVMe drive too, but they aren’t so easy to find around here. We are using Gigabyte Z890 AORUS PRO ICE motherboard, which is exactly the same like its Z790 cousin, with just a different socket.

The Gigabyte BIOS offers three Intel Default performance profiles for the CPU under the setting GIGABYTE PerfDrive – BaseLine, Performance and Extreme. We will run our benchmarks using the Performance and Extreme profiles. The Extreme profile removes the 255W power limit.

Cinebench R23

First and foremost, the most recognizable CPU benchmark by Maxon called Cinebench R23. This benchmark renders a cinematic image using the CPU’s power.

Cinebench 2024

3DMark Steel Nomad

Power Usage

Conclusion