In October this year, the company officially announced the release of the code-named "Arrow Lake" and built on TSMC's process.Core Ultra 200S series desktop processorsThis time, I also used the first wave of Core Ultra 9 285K and Core Ultra 5 245K for testing to see what changes have been made to Intel's desktop processors based on the "Lunar Lake" design.
A brief review of the two processor designs
"Arrow Lake" also targets "AI PC" application needs. Many of its designs are based on "Lunar Lake" designs. It is also produced using TSMC's 3nm process, with the substrate made using Intel's 18A process technology, and is produced in conjunction with Intel's Foveros 3D packaging technology.
In terms of performance, Intel claims that "Arrow Lake" will be able to reduce power consumption by up to 30%, improve multi-core computing performance by more than 10%, and even more than double display performance. It also further enhances artificial intelligence computing capabilities and cooperates with more than 2 software companies to correspond to more than 100 artificial intelligence application functions and support more than 300 artificial intelligence models used in the market.
The Compute Tile utilizes eight performance cores (P-Cores) codenamed Lion Cove and 8 energy-saving cores (E-Cores) codenamed Skymont. The performance cores are designed for AI computing, with 16 execution channels and an L18 cache memory increased to 36MB. Combined with an L3 cache memory increase to 3MB, this expands the computational prediction range by eight times, thereby improving AI execution efficiency.
The energy-saving core also improves computing prediction capabilities and doubles the L2 cache memory bandwidth. Combined with an increase in shared L2 cache memory capacity by 4MB, this increases data throughput during artificial intelligence computing, significantly improving the computing accuracy of the energy-saving core compared to the previous generation.
On the other hand, with the enhanced L3 cache memory design, "Arrow Lake" can more efficiently allocate and schedule computing tasks between different cores through the Intel Thread Director function, allowing the processor to switch to the performance cores more efficiently to quickly complete tasks and quickly return to the energy-saving cores to perform general work tasks, thereby reducing overall power consumption by up to 30%.
However, Arrow Lake continues the Lunar Lake architecture, eliminating Hyper-Threading and replacing it with additional L3 and L2 cache memory. While it boasts of using Intel Thread Director to more efficiently allocate and schedule tasks across different cores and reduce overall power consumption, its actual performance is no better than the previous generation of processors, prompting widespread criticism.
In terms of display performance, the GPU design integrates the Xe-LPG display architecture, which combines 4 Xe display cores, 64 vertical computing engines, 4 samplers, and 4 real-time ray tracing computing elements. The L2 memory is increased to 4MB, and it is also advertised as compatible with Microsoft DirextX 12, thereby corresponding to more games and software compatibility. At the same time, it can also improve display performance through XeSS technology and Arc software stacking.
In terms of artificial intelligence computing, in addition to improving the AI computing capabilities of the CPU core, Intel has also increased the proportion of NPU component computing applications in "Arrow Lake" and adopted the third-generation NPU architecture design, allowing processor application products to correspond to higher AI computing performance while also supporting more than 300 AI software application functions.
As for connection specifications, "Arrow Lake" supports the use of DDR 5 6400 memory modules and can improve memory access efficiency and reliability through the DDR PHY controller. It supports up to a single 48GB memory module, with a maximum usable memory capacity of 192GB. It also supports ECC error correction and dual-channel configuration.
The motherboard chip has been upgraded to the 800 series, along with the new LGA1851 socket, meaning users must replace their motherboards to use it. Furthermore, Arrow Lake supports a total of 48 PCIe lanes, including up to 20 PCIe Gen 5.0 lanes or 24 PCIe Gen 4.0 lanes. It also natively supports two Thunderbolt 2 connections and can handle up to 4 USB 32 lanes or 3.2 USB 14 lanes. Wireless connectivity includes Wi-Fi 2.0E and Bluetooth 6, as well as 5.3GbE wired network connections. Intel KiLLER technology can also be used to improve network transmission efficiency.
Through external connections, it can be expanded to use 4 sets of Thunderbolt 5, support Wi-Fi 7 and Bluetooth 5.4, and can also use wired networks with a transmission bandwidth of up to 2.5GbE. It can also be used with Intel KiLLER technology to improve network transmission efficiency.
Actual test experience
The test specifications include Core Ultra 9 285K and Core Ultra 5 245K processors, ASUS ROG STRIX Z890-E GAMING WIFI motherboard, NVIDIA GeForce RTX4070 Ti graphics card, and ASUS ROG STRIX LC II 360 ARGB water cooling kit. Kingston Fury Renegade DDR5 RGB 32GB memory (2 x 16GB) is used, and the storage component is mainly Crucial T700 Pro PCIe 5.0 SSD with 2TB capacity.
The Core Ultra 9 285K is comprised of 8 P-Cores and 16 E-Cores, for a total of 24 threads. The P-Cores have a base operating clock of 3.7GHz and a maximum operating clock of 5.7GHz, while the E-Cores have a base operating clock of 3.2GHz and a maximum operating clock of 4.6GHz. It also features 40MB of L2 cache and 36MB of Smart Cache, with a thermal design power consumption of 125W and a Turbo Boost limit of 250W.
The Core Ultra 5 245K is composed of 6 P Cores and 8 E Cores, corresponding to a total of 14 execution threads. The P Core has a base operating clock of 4.2GHz and a maximum operating clock of 5.2GHz, while the E Core has a base operating clock of 3.6GHz and a maximum operating clock of 4.6GHz. It is equipped with a total of 26MB of L2 cache memory and 24MB of Smart Cache. The thermal design power consumption is 125W, and the Turbo Boost limit can reach 159W.
In addition, the ROG STRIX Z890-E GAMING WIFI motherboard used this time is designed with ATX specifications, and the heat sink is presented with ROG gradient pattern. The heat sink on the south bridge chip is also laser-engraved, which will have different light refraction effects at different viewing angles.
As for the LGA1851 power supply corresponding to the processor, it adopts an "8+8" solid pin design, and the motherboard's power socket also adopts a solid pin design, and is equipped with an LED display that can display debug codes and a power button. The four DDR4 memory slots are equipped with ASUS's exclusive NitroPath DRAM technology that allows memory overclocking performance. In addition, it is also equipped with BIOS DIMM FLEX and AEMP III technology to make the memory more stable when operating at high clock speeds. It can correspond to the installation of up to 5GB of memory capacity and supports CUDIMM memory with built-in clock driver.
The motherboard offers a PCIe 5.0 x16 slot with a PCIe Q-Release Slim quick-release design, allowing users to quickly remove graphics cards. Another slot corresponds to the PCIe 4.0 x16 specification and also supports x4 operation. There are a total of seven M.2 slots, three of which are PCIe 7 x3 and four are PCIe 5.0 x4. The PCIe 4 x4.0 slot located above the PCIe 4 x5.0 slot features an M.16 Q-Release and M.5.0 Q-Slide quick-release heatsink, while the remaining slots use the screwless M.4 Q-Latch quick-release design.
The I/O ports include DisplayPort 1.4, HDMI 2.1, two Thunderbolt 2 ports (capable of 4Gbps transfer speeds), three USB 40 Gen3 ports (blue), and nine USB 3.2 Gen1 ports (red, seven of which are Type-A, one is USB-C, and one supports 9W PD fast charging). There's also a Realtek 3.2GbE network port, a Wi-Fi 2 antenna connector, an optical S/PDIF port, and a 7mm gold-plated analog port. BIOS and CLEAR CMOS buttons are also provided.
Other parts include USB3.2 Gen2x2 and USB3.2 Gen1 connection interfaces, as well as four SATA 4Gb/s connection interfaces on the side and an additional USB 6 Gen3.2 connection interface. It also supports Wi-Fi 1 connection specifications, using the Intel BE7 wireless chip to correspond to 200QAM, 4096MHz bandwidth and 320Gbp transmission speed performance.
• Overall performance
Geekbench and CrossMark tests of the processor's single-core and multi-core performance, as well as its performance in various computing scenarios, reveal that the Core Ultra 9 285K and Core Ultra 5 245K don't differ significantly in single-core performance, but there's a significant difference in multi-core performance. The Core Ultra 9 285K also has an advantage in multitasking and creative applications.
• Rendering operations
Evaluating the rendering performance of the two processors using Cinebench and Blender reveals similar performance differences in single-core performance, but noticeable differences in multi-core computing performance.
• 3DMark
Using 3DMark to evaluate the performance of both processors in physics computing and custom simulation computing, we can see that the Core Ultra 9 285K still achieves higher performance in multi-threaded computing.
• PCMark
Using PCMark to evaluate the overall computing performance of the two processors, the Core Ultra 9 285K still has the advantage overall, but looking at the detailed performance results, the Core Ultra 5 245K actually has some performance.
Conclusion
Overall, Arrow Lake, built on TSMC's process and featuring a new architecture, does offer improvements in single-core performance and significantly reduces overall power consumption. However, the removal of multi-threading and the reconfiguration of cache memory are detrimental to performance in most games, potentially leading to higher performance compared to previous-generation processors.
However, judging by the overall design, Arrow Lake's changes are primarily focused on AI PC applications, with reduced power consumption also a key design priority. Furthermore, the use of a new architecture and TSMC process may have led Intel to consider different product strategies, deprioritizing gaming usage. This may lead to a different approach in future product launches.
However, judging from the major architectural changes and improved single-core performance, it is still attractive to a certain extent. However, if you mainly play games, you may still want to wait and see.
Furthermore, the price difference between the Core Ultra 9 285K and the Core Ultra 5 245K is US$280, or NT$9400. If gaming is your primary need, you might consider investing more of your system setup costs in graphics card specifications to enhance gaming performance.
