On many hardware sites I read the new announcement of the Intel Core Ultra 200S CPU, codenamed Arrow Lake-S. Intel’s latest work is supposed to be the biggest step in the desktop environment in years. I see it differently.
Always those benchmarks
The reports contain numerous benchmarks for applications and games, the smallest improvements in megahertz and RAM speed, what feels like a thousand new interfaces, a new chipset, and so on. All this data is rather boring for the average user. For example: It can handle DDR5–6400, but DDR5–5600 will remain the standard for the time being. It supports ECC RAM, but this will probably never find its way to the desktop (it’s more for Xeon CPUs). Do you find this as exciting as I do?
Or the announcement that the CPU finally supports Thunderbolt 4. Some of the articles mention something new on the desktop. Do the editors know that Thunderbolt 4 has been announced for 2020? Or that the 40 Gbps transfer rate is the same as Thunderbolt 3? Of course, there are advantages between the two generations in general, but to sell that as a major feature? I don’t want to bore you with unimportant details.
What is remarkable is the attention to detail with which even the smallest and most unimportant innovations to the average consumer are covered. The editors really jump on anything Intel’s press department throws at them.
Nevertheless, I read the articles about the many innovations and details.
Some of it is interesting, but I’m starting to get the feeling that these innovations aren’t all that new. I already know many of the features from previous architectures, such as the 3-year-old Raptor Lake or its direct predecessor, the Meteor Lake.
At some point I reach a point that makes me wonder:
Most reviews talk about single-core performance, multi-core performance, and gaming performance. But something is missing from the list, something that is completely changing our IT right now: AI performance.
Where is the NPU for AI?
For those wondering what it is, an NPU (Neural Processing Unit) is a specialised processor optimised for AI applications and machine learning.
Many people believe that a powerful graphics processing unit (GPU) can do the job of a neural processing unit (NPU). However, this is only partially true. While GPUs excel at parallel processing, they are less optimised for AI-specific operations than NPUs. The key difference is power efficiency: NPUs perform AI computations at a fraction of the power consumption of a GPU. It is this specialisation and efficiency that makes NPUs indispensable for modern AI applications, especially when it comes to local processing on devices. They enable powerful AI capabilities while keeping power consumption low - a combination that GPUs cannot offer in this form.
For example, Microsoft recently set a minimum NPU performance of 40 TOPS (trillion operations per second) for its Copilot+ PC. Apple uses around 38 TOPS in its M4 and is laying the groundwork for many new AI features in iOS 18.
In Intel’s official presentation, I found some NPU details buried deep at the end: Intel’s brand new Core Ultra 200S CPU can process a whopping 13 TOPS. Nevertheless, Intel promotes its CPUs with the slogan 'AI everywhere'.
This is well below Microsoft’s specifications and even below the level of the iPhone 14 from two years ago (17 TOPS). This could be one of the reasons why Microsoft has decided to remove Intel processors from its Surface range altogether. Instead, the company has opted for the ARM-based Qualcomm Snapdragon X CPU, whose NPU offers 45 TOPS of performance.
Some will say that AMD’s current Ryzen 9000 series does not need an NPU at all. That’s true, but its successor is already on the horizon: AMD says the Ryzen AI 300 series (‘Strix Point’) will support at least 50 TOPS. What AMD will do with that remains to be seen. Now that the cat is out of the bag, I can well imagine AMD rubbing its hands and wondering what to do.
Why Intel is only offering 13 TOPS is also explained in the press release. There Intel explains surprisingly clearly that they deliberately don’t use larger NPUs because performance per CPU core is more important on the desktop for now and they don’t want to waste space for it (at IFA 2024 at the end of August Intel sounded quite different).
The new reality
A new technology and software that is increasingly changing and determining everyday life is not being integrated because Intel does not consider it important enough?
This is indeed a bit incomprehensible, as Intel already has a 48 TOPS NPU in the ultra mobile space with the Core Ultra 200V. Apart from mobility, I don’t see much difference between a notebook and a desktop.
In my opinion, Intel is completely ignoring reality here. For example, last week there was an update to a well-known developer IDE that now allows local AI to run so that it can better and automatically complete code for the user. Without an NPU, this is almost impossible.
Let’s take a look at general software that doesn’t make you think of AI: modern antivirus software relies on local AI functions to better detect threats. Modern image editing software and translation programs also use local AI. And so on.
Reduced graphics support and no more hyperthreading
The graphics support on the Core Ultra 200S is similarly disastrous. This is on the same level as its predecessor and is really only suitable for desktop PCs. Intel had actually shown that it could be done differently with the ultra-mobile Core Ultra 200V and Core Ultra 200H, where fast internal graphics were installed (8-core ‘Xe2’ and 8-core ‘Xe XMX’). The new Core Ultra 200S only has the 4 cores of the previous ‘Xe’ generation. What a disappointment.
Intel recommends installing a discrete graphics card. But not everyone, especially companies, will install a discrete graphics card in every desktop for their employees. They will be the minority. Perhaps Intel should rethink its target audience for desktops.
It is also questionable why the new CPU no longer supports hyper-threading, unlike AMD products.
AI, a key enabling technology
In summary, we have a fundamentally new key enabling technology in AI that is very likely to completely change the market, programmes and life in the next few years.
And what NPU is too small and underpowered for it: Intel’s brand new Core Ultra 200S CPU. Congratulations to Intel.
This means that Microsoft’s new AI features are out of the question, and only time will tell if the power of 13 TOPS is enough for small local AI software. Microsoft and Apple certainly don’t think so.
Given this situation, is it wise for individuals or businesses to invest in a computer with the new Core Ultra 200S CPU if it is likely to support only limited future technologies and the manufacturer is underestimating the importance of AI support?
Who is this CPU aimed at? Is it aimed primarily at hardcore gamers who have little interest in future AI developments? Sooner or later, it could also turn out that, for example, a virus scanner with local AI functionality is significantly slower without an NPU.
Something positive
If there is anything positive to be said about the new CPU, it is that it finally works in a power-saving way. The competition has been doing this for a while, but after the last unstable Core processor disaster, it was important.
A flop
Personally, the new CPU is a flop. Not enough NPU power, not enough interest from Intel in new possibilities. Either they didn’t get it right technically, or they need to rethink their marketing department. How can you design a CPU that misses the market so completely? Apple, Microsoft, ARM and others have realised this. Intel can no longer see the forest for the trees and has been clinging to old patterns for too long.
AI everywhere should feel different.
P.S.
To all the hardware sites and editors who no longer notice that they uncritically copy Intel press releases or get lost in technical details: It is possible to be critical of big companies like Intel and their products. Yes, a critical stance could jeopardise preferential access to future product launches. But is that more important than journalistic integrity?
Let’s be brave and report on technology in a holistic way, regardless of the possible consequences. Only in this way can we provide real value to our readers.
NPU and AI are important new key technologies, and those who are not on board will miss the boat. Disruption is not stopping at the big technology companies.
Best regards,
Frank
