Intel has officially announced the availability of the next generation of processors for desktop PCs. The Core Ultra 200S desktop chips for the new LGA1851 platform have a lot of architectural, structural, and even functional improvements. In two weeks, the first CPUs of the Arrow Lake-S family will be available for sale, but for now, we can evaluate the characteristics declared by the manufacturer.
Traditionally, the transition to the next generation of desktop processors begins with the announcement of top models for the most demanding enthusiasts.
The lineup will initially include five chips. The series is headed by the flagship Core Ultra 9 285K, a 24-core processor with 8 performance cores (P-Core) based on Lion Cove architecture and 16 energy-efficient cores (E-Cores) based on Skymont. The overall functional formula is 8P+16E (3.7/5.7 GHz + 3.2/4.6 GHz). The chip also received 36 MB of L3 and a total of 40 MB of L2.
In the new Arrow Lake-S processors, the developers have abandoned support for Hyper-Threading, and apparently compensate for the lack of HT with architectural improvements to the computing cores.
Core Ultra 7 265K/KF has 20 cores – 8P+12E (3.9/5.5 GHz + 3.3/4.6 GHz), 30 L3 and 36 L2, while Core Ultra 5 245K/KF chips have a 14-core layout – 6P+8E (4.2/5.2 GHz + 3.6/4.6 GHz) and 24 MB L3 and 26 L2. The models with the “K” index are equipped with integrated graphics with Xe-LPG architecture, while the chips with the “KF” designation have a deactivated video core.
All of the announced Arrow Lake-S chips have a base TDP value of 125 W, with a maximum power consumption of 250 W for Core Ultra 9/7 and 159 W for Core Ultra 5.
Core Ultra 200S processors support only DDR5 RAM. The transition period for the introduction of the new standard is apparently over, so further support for DDR4 no longer makes sense. The new chips officially support DDR5-6400 mode and 48GB modules. Thus, the maximum capacity of a set of dual-channel kits is 196 GB.
The support for CUDIMMs with their own clock is a hint that DDR5-6400 is only the beginning, and Core Ultra 200S will be able to work with much faster memory kits. DDR5-8000 is already predicted to be the “gold standard”, and even DDR5-10000 will not be the limit of the possible.
Core Ultra 200S is Intel’s first desktop processor with a tile structure, which involves the use of multiple functional crystals. The main unit (Compute Tile), which includes high-performance and energy-efficient computing cores, is manufactured using the improved 3-nm TSMC N3P process. For the SoC Tile, the 6-nanometer TSMC N6 process is used, and for the crystal with the graphics core, 5 nm (TSMC N5P). Intel’s proprietary Faveros 3D technology is used to combine all the tiles.
Initially, it was assumed that Compute Tiles for Core Ultra 200S processors would be manufactured in-house using the new Intel 20A process. However, Intel decided to focus on preparing for the launch of large-scale crystal manufacturing according to Intel 18A (1.8 nm class) in early 2025. For this reason, it became more rational to order computing tiles for Arrow Lake-S from a Taiwanese partner.
This silicon hodgepodge, compared to the single-crystal design, allows the manufacturer to combine blocks and potentially reduce the overall cost of manufacturing the processor.
The new chips have significant architectural differences from the previous generation Raptor Lake processors. To summarize, according to the manufacturer’s assurances, P-Core performance per clock cycle has increased by an average of 9%, while the IPC (instructions per cycle) for energy-efficient computers has increased by as much as 32%.
Intel also equipped the new processors with hardware units to accelerate AI computing. Core Ultra 200S received NPU 3 with a declared performance of 13 TOPS (int8). So, in fact, the same module is used for Arrow Lake-S as in Meteor Lake (Core Ultra 100) mobile processors. Unfortunately, the fast NPU 4 unit with 48 TOPS remains the highlight of Lunar Lake chips (Core Ultra 200V). Consequently, desktops based on Core Ultra 200S without a discrete graphics card will not be able to pass Copilot+ PC certification. In general, the presence of NPU is an interesting feature and a competitive advantage over AMD solutions. However, it will be significant only with the active implementation of software that requires local AI computing.
Intel emphasizes a radical improvement in the energy efficiency of new CPUs. Depending on the nature of the workload, power consumption can be up to 58% lower than Raptor Lake.
Under multi-threaded workloads, Arrow Lake-S can deliver the same performance as previous generation chips with twice the power consumption. This is a significant change, because in the case of the top Core chips of the 13th/14th generations, it was the high level of power consumption that caused the most complaints.
Intel claims improved efficiency during gaming sessions as well, promising a reduction in total platform consumption of up to 165 watts at identical fps. Such results seem to be a bit of an exaggeration that needs to be independently confirmed in practice. The difference in CPU+board consumption during gaming cannot be that significant.
But what seems quite likely is a drop in operating temperatures. When using a liquid cooling system (AIO 360 mm), the average heating of the Core Ultra 9 285K during gaming is ~13C lower than that of the Core i4-14900K.
In terms of performance, according to Intel, in single-threaded mode, the top-end Core Ultra 9 285K is on average 8% faster than its predecessor Core i9-14900K and has a 4% advantage over the Ryzen 9 9950X.
In the case of a multi-threaded workload, the Core Ultra 9 285K is already 15% ahead of the Core i9-14900K and 13% ahead of the 16-core 32-threaded Ryzen 9 9950X.
In the gaming “Clash of the Titans” – Core Ultra 285K vs. Ryzen 9 9950X – no winner has been identified. In some projects, the performance of Intel’s new chips is better, while in others, AMD’s model is ahead. However, this requires objective, unbiased testing. The chart shows the performance for certain games when using Intel AOP (Application Optimization), which really improves performance, but requires preliminary “profiling”.
There are no comparisons with our predecessors. It seems that the performance here is similar, or even a minimal regression. At least on one of the slides, the performance of the new chip is a few fps lower than that of the Core i9-14900K. But the new platform achieves a similar number of fps with significantly lower power consumption.
Comparing the performance of the Core Ultra 285K with that of the Ryzen 9 7950X3D is quite revealing. AMD’s top-of-the-line model with a large additional cache buffer has an advantage in games (with the exception of Sid Meier’s Civilization VI: Gathering Storm), and Intel’s new product is unbeatable when working with media content.
However, even in this area, the balance of power depends on the specific tasks. With optimized codecs, the advantage in video processing can be significant, while in other cases the new AMD 16-core processors will be better.
It’s not hard to guess the balance of power in applications that may involve AI computers. Although such a software segment is still in its infancy, the presence of a hardware NPU with even 13 TOPS will in some cases significantly improve performance and allow Intel to proudly use definitions such as “Expanding AI PC to Desktop.”
For Core Ultra 200S processors, a new desktop platform, LGA1851, is provided. This means that Arrow Lak-S chips will be available with motherboards based on the top Intel Z890 chipset. The degree of functional integration of the CPU is already so high that only the peripheral “harness” remains for PCH. Here, the Intel Z890 additionally provides support for up to 24 PCI-E 4.0 lines, a considerable number of high-speed USBs, and up to 8 SATA ports.
Note that the Core Ultra 200S has 20 PCI-E 5.0 lines, so on the new Intel platform, you no longer have to split the links allocated for the PCI-E x16 slot to connect high-speed SSDs. Other features of Arrow Lak-S include integrated support for Thunderbolt 4 and Wi-Fi 6E. Even faster options – Thunderbolt 5 and Wi-Fi 7 – can be implemented by board manufacturers using additional discrete controllers.
The new LGA1851 platform should be quite interesting for enthusiasts who want to experiment with overclocking. This includes a 16.67 MHz frequency step resolution for P/E-Cores, separate base frequencies for different tiles, the ability to change the coefficients for the inter-tile bus, and other options that will allow you to get a little more performance. No significant acceleration of P-Cores is expected, but energy-efficient ones can be further “pumped”.
In general, the new LGA1851 with Core Ultra 200S chips looks like a very interesting platform for research and experiments. After that, it will be possible to make assessments and draw conclusions. Intel will have to make efforts to regain the favor of active desktop users, who have recently been favoring the opponent’s solutions.
Sales of new Intel Z890 processors and motherboards are scheduled to start on October 24. At the same time, the NDA for the publication of independent reviews will be finalized, which will allow answering practical questions that remain after a superficial acquaintance with the key platform updates.
Finally, let’s talk about the most interesting thing – processor prices. The flagship of the Core Ultra 9 285K line is priced at $589. Core Ultra 7 256K/KF chips are priced by Intel at $394 and $379, respectively. Whereas the 14-core Core Ultra 5 245K/KF is priced at $309/$294. These figures for the US market are close to those we had for 14th-generation Core chips. Some new products are even $5-15 cheaper than their respective predecessors.
It should be noted that the KF version without integrated graphics is now $15 more affordable than the version with an active iGPU. Previously, the difference was $25. We are waiting for the start of sales of the new Intel platform in Ukraine and are preparing for the predicted initial “exaggerations”.
Loading comments …