Intel is making its energy and AI revolution with the Core Ultra 200S for desktop PCs

Finally ! That’s the first word that comes to mind when looking through the slides of the press presentation for the new generation of desktop processors that Intel is announcing today. The Core Ultra 200S, first presented under the code name Arrow Lake, are now a reality. And they bring with them three major improvements: a much more efficient CPU architecture – which results in a design seeking the best performance per watt – and Intel’s first desktop AI accelerator.

Advertising, your content continues below

Energy consumption (and temperatures) finally falling

At the heart of the Core Ultra 200S, we find the same two CPU microarchitectures as in the mobile Core Ultra 200V. Namely the powerful cores (P-cores) called Lion Cove and the small economical cores (E-Cores) called Skymont.

Operating under the leadership of a ThreadDirector which arbitrates the distribution of tasks on the two types of cores, the CPU part of this chip for desktop PCs benefits from the same gain in energy efficiency. An efficiency that has already made Intel’s mobile chips the processors with the best performance per watt ratio ahead of Qualcomm’s ARM Snadpragon X Elite chips!

To achieve this efficiency, Intel removed HyperThreading. Which leads this generation of desktop chips to only offer very moderate performance gains compared to its previous generation of chips (14th gen).

An approach of efficiency which will be felt at the socket, but also in your case: the temperature gains are very significant. Depending on the scenarios, the chip would display -10°C and up to -17°C compared to the previous generation – with an average usage around -13°C. Enough to soften the bills, and protect the ears a little.

Advertising, your content continues below

However, we can criticize Intel here for having given users a free hand: if its chips give the best performance between 65 W and 125 W – a consumption level where it equals the previous generation at 250 W! – the Core Ultra 9 can still be pushed to 250 W…

Intel’s first desktop NPU (but not the most powerful)

For the first time in the history of Intel desktop PC processors, the Core Ultra 200S accommodates a neural processor (or NPU or AI accelerator). Already implemented since last year in the Meteor Lake architecture (3rd generation after external accelerators), then this summer in Lunar Lake, the NPU integrated into this new generation of processors will be able to unload the CPU (and avoid requiring the GPU ) in certain tasks. We are thinking here of so-called maintained tasks, such as real-time noise suppression filters for videoconferencing.

One downside, however: the NPU integrated into these chips is 3rd and not 4th generation. It is therefore three times less powerful (13 TOPS) than the NPU integrated into Lunar Lake mobile processors (48 TOPS), yet smaller and less energy-consuming!

Several (potentially concomitant) scenarios are possible regarding the choice of a 3rd generation NPU. There is the place – and the position in relation to the other elements – on the SoC on the one hand. There is also the possibility that the team in charge of designing the chip made its choice on this NPU very early on. Or the fact that given that desktop PCs have very powerful additional GPUs, a simple acceleration of +10 TOPS is enough for real-time audio filters, and that the GPU will take care of the rest.

Advertising, your content continues below

If we regret this choice from an energy point of view (and raw power), we nevertheless celebrate the arrival of such an accelerator in desktop chips – even if at only 36 TOPS, this chip is far from its equivalent mobile which displays up to 120 TOPS! Hoping that Intel (and others) go even further for the next generations of chips.

Another slightly dated element: the GPU integrated in three of the five references. No second generation graphics architecture (Battlemage) as is the case in mobile chips, but four small Xe cores. Which still have the good taste of coming with the video encoding/decoding part and Quick Sync Video technology, one of the most effective in the industry.

Three processors make five references

If Intel is announcing five chip references today, it is in fact three different processors that are arriving, including two with variants. The K-branded chips integrate four graphics cores, the two KF variants (Core 7 and Core 5) being devoid of this graphics part.

The flagship is the Core Ultra 9 285K, a nasty SoC packing 8 P-Cores and 16 E-Cores for a total of 24 cores and therefore 24 threads – remember, there is no HyerThreading on this generation.

Below the Core Ultra 7 265K/KF lose four E-Cores (8+12) and the Core Ultra 5 245K/KF only has 6 P-Cores and 8 E-Cores – and loses the Intel in the process TurboBoost.

Anecdotally, but nevertheless appreciably, we notice that the prices are lower than the previous generation by a few tens of dollars. All for a platform that is the richest on the market, since Intel is the only player to ensure the development of Thunderbolt (here in version 4).

Advertising, your content continues below

Much like Meteor Lake, it appears that the batch of Arrow Lake/Core Ultra 200S chips is a transitional platform. Which requires – and this is regrettable – a new motherboard with Z890 chipset. It is therefore impossible to update your 12th, 13 or 14th generation platform. One of the reasons given is the definitive transition to DDR5, the previous families of chips being hybrid DDR4/DD5.

We therefore cannot say whether Intel, which is in the process of reorganization, will release these Core 200S chips into other references. Or just occupy the market for a while, before moving on with a completely new architecture.

The end of the wattage madness?

On paper, we see that Intel’s new chip does, roughly, as well or a little better for less energy dissipated (and consumed!) and for less cost. For us, it’s called progress – as opposed to the race for power for the sake of power.

Inspired by its change of mentality first at Meteor Lake then Lunar Lake for its chip architectures for laptop PCs, the approach to researching energy efficiency is one of the turning points that Intel was slow to take. We can only rejoice and hope that it lasts and is now more than an objective: a real mantra.

The unlocking of the turbo mode of the Ultra 9 and Ultra 7 chips at 250W suggests that the question is bothering Intel internally, undoubtedly torn by its DNA (and certain fans) of racing for absolute power. Let’s hope reason prevails.

Advertising, your content continues below