Results of 2024: PC processors

After a rather sluggish 2023, the past year has poured out a waterfall of news on the processor market, many of which deserve the epithet “sensational.” During this time, both manufacturers of x86 processors not only changed the architecture of the products they offer for desktops and laptops, but also fundamentally revised their solutions in many ways. For example, AMD increased the number of processing cores in processors for mobile PCs, and rethought 3D V-Cache technology for desktop processors. At this time, Intel switched to producing chips at TSMC facilities, and was also actively mastering the tile (chiplet) design of processors. There’s also a third major player in the Windows ecosystem: Qualcomm is finally shipping its promising Arm processors aimed at PCs.

A separate big topic of the past year was computers with artificial intelligence. In the wake of the AI ​​boom, Microsoft formulated the concept of Copilot+ PC, describing the requirements for modern PCs capable of locally running applications with neural network algorithms embedded in them.

The first such application, according to the company’s plan, was to be the Windows 11 operating system itself. In its standard applications, Microsoft promised to introduce a set of AI capabilities for working with graphics and texts, and also came up with the Recall function, which allows you to search for information in the history of PC use. This initiative became an important leitmotif of the evolutionary development of processors: a significant part of the CPUs released last year tried to fit into the requirements of Copilot+ PC, which spurred a kind of “NPU race” – an unspoken competition between manufacturers in increasing the power of hardware AI accelerators built into processors.

However, the year 2024 will be remembered primarily not for this, but for the story that Intel got into. The company, which until recently seemed to be an unshakable pillar of the entire computer industry, began to stagger suspiciously under the blows of fate, giving rise to very gloomy forecasts. It is not surprising that what is happening in the blue camp quickly became one of the central topics of news feeds, and if you look at the statistics of publications, for example, on our website, it turns out that in a year at least one and a half times more articles were written about Intel than about AMD . It was impossible to ignore such a resonant topic in the final material on PC processors, so we will begin a retrospective look into the recent past with a story about what is happening at Intel.

⇡#Topic of the year: what happened to Intel and will the company survive?

The past year was, if not fatal, then at least a turning point for Intel. The company faced colossal losses, a loss of investor confidence in its future, and the subsequent double collapse of shares. And at the end of the year, the creator of the i486 processor, the legendary Pat Gelsinger, was expelled from the post of Intel CEO, whose return to Intel in 2021 was perceived as nothing less than a turn of the company towards an imminent renaissance. And this obviously means that, under pressure from shareholders, the company’s board decided to deviate from the intended path and think about fundamental changes in strategy. What this means for the company is still unclear, but we can say with confidence that Intel will never be the same.

When Gelsinger took over Intel three and a half years ago, he brought with him a plan to rebuild the company as the world’s leading semiconductor maker. It was based on the development of five technical processes over four years with the launch of Intel A18 angstrom technology in 2025, plus the development within Intel of semiconductor contract manufacturing services for third-party customers, which would have to cover the gigantic costs of technical re-equipment and maintaining positions in the race of technological processes with TSMC.

And this plan did not come out of nowhere. Before Gelsinger’s arrival, Intel had been gradually losing ground for many years and missing out on growth opportunities one after another. The company slept through the smartphone processor market, lost Apple as a client, lagged behind in the development of new technological processes due to skepticism regarding EUV lithography, ignored the emergence of the GPU market and was never able to take advantage of the AI ​​boom, on which other technology companies are now actively profiting . Gelsinger’s idea to make Intel a contract manufacturer of semiconductors in the likeness of TSMC seemed to be almost the only chance to maintain the company’s former scale and consolidate its role as a key player in the global semiconductor industry. And such a plan seemed quite feasible against the backdrop of the desire of the US authorities to eliminate the country’s dependence on Taiwanese production, which had already hit the economy hard once during the coronavirus crisis. Thanks to the political situation, Intel managed to enlist the support of the government and receive billions of dollars in subsidies and government orders, which became one of the important pillars of the company’s revival plan.

Thus, Gelsinger focused his efforts on correcting one of the mistakes of the past. Despite the fact that Intel invested billions of dollars in ASML in the 2010s, the company did not purchase first-generation EUV equipment from it, which subsequently led to a serious delay in the development of 10-nm technology and a many-year lag behind TSMC in technological standards. But now, in order to make up for lost time and return to the framework of Moore’s Law, Gelsinger was able to agree with ASML on priority supplies of new generation lithographic EUV scanners with a high numerical aperture, necessary for the development of future angstrom-level technological processes.

And as a result, the part of the plan concerning the accelerated development of new technological processes does not look like a failure at all. Gelsinger promised that Intel should master the 18A process in 2025, and, apparently, this milestone is quite achievable – Clearwater Forest server chips, as well as the Panther Lake desktop chips that are planned to be released based on it, are still on the company’s roadmap for next year.

However, the problem is that in pursuit of an ambitious goal, the achievement of which would allow Intel to talk about being ahead of TSMC in the technology race, the company literally overextended itself. Tens of billions of dollars were spent on new equipment, and this was by no means free funds in the accounts. The company had to resort to reducing dividends, refusing to pay bonuses to employees and cutting staff (15 thousand people were laid off last year alone), delaying the construction of enterprises in Europe, selling existing Arm shares and cutting costs in other areas. The savings also affected Intel’s main source of profit – the processor business. As a result, some promising projects were completely curtailed, and some were sequestered. Even the flagship Arrow Lake processor had to be cut with a knife, which lost the additional Adamantine L4 cache, which hurt the consumer qualities of the new product.

Moreover, Intel began to save on pre-release testing of processors, which resulted in even more serious troubles: the epic with the spontaneous degradation of Raptor Lake became one of the biggest scandals in the company’s history, and Intel engineers have to finish the new Arrow Lake on the fly after their release .

It was far from the best idea to postpone the production of some of our own chips at TSMC facilities. The new Lunar Lake and Arrow Lake processors consist almost entirely of chips produced by a third-party contractor, and this greatly increased their cost without providing any significant technical advantages. Even Gelsinger himself admitted that Lunar Lake ultimately turned out to be a “financial nightmare” for the company, since Intel had already had to pay TSMC $5.6 billion, and upcoming expenses on production outsourcing are estimated at another $9.7 billion.

It is not at all surprising that a series of miscalculations against the backdrop of a rapidly deteriorating financial situation overwhelmed the patience of Intel board members, who sent Gelsinger to resign. The main complaint of the board of directors against Gelsinger, obviously, was that under his leadership the processor direction began to turn from Intel’s core business into a “cash cow” for the manufacturing revolution. And this is a dangerous path, so in the very first comment the board of directors promised to shift the focus to specific products and customer interests, as well as return the emphasis to the processor direction.

Moreover, the ongoing production revolution does not yet give reason to hope for a positive financial result. During the time that Gelsinger was at the helm, he was never able to turn Intel into any successful contract chip manufacturer. A serious flaw was discovered in his plan: it turned out that it was not so easy to persuade large customers to change contractors. In the three years that Intel has been trying to transform itself into a semiconductor maker for third-party customers, it has only signed three major contracts—with the U.S. Department of Defense, Microsoft, and AWS. Several more companies are considering the possibility of cooperation with Intel, but the available information is not too optimistic – many potential partners are not satisfied with the parameters of Intel’s advanced technical processes, including the low yield of usable crystals at this stage.

It is still unclear what will happen to Intel next. Its processor business continues to generate $2.5 billion in quarterly revenue, but given the huge spending of its manufacturing division, the situation is looking pretty bleak. In the first quarter of last year, Intel’s total loss amounted to $400 million, in the second it increased to $1.6 billion, and in the third it reached a fantastic $16.6 billion, which was the worst moment in the entire 55-year history of the company.

Therefore, many naturally expect that Intel will want to separate resource-guzzling production and focus exclusively on designing processors, while becoming a factoryless manufacturer in the image of AMD. However, an obstacle may be government support received by Intel in the amount of $8 billion. It is aimed specifically at the development of the production line, and therefore the conditions for receiving it directly prohibit Intel from losing control over the relevant division. In addition, it is completely unclear whether Intel’s semiconductor production is viable without financial support from the parent company in principle, because it still does not have a sufficient number of external clients, and the technical processes at its disposal are clearly focused on internal needs.

However, it is obvious that we won’t have to wait long for the outcome. Time is not on Intel’s side: the company right now continues to lose the trust of not only investors, but also customers, and therefore, ultimately, profits. Therefore, she is quickly looking for a new head, and as soon as a suitable candidate is found, we will find out what changes in the “blue” camp will occur in 2025. One thing is clear: the restructuring initiated by Gelsinger not only did not please the board and shareholders, but ran into fundamental obstacles in the form of a lack of finance, and therefore is unlikely to be fully completed. Therefore, the timely implementation of the 18A technical process, and even more so the subsequent 14A, is no longer guaranteed. It is also questionable whether Intel is even able to continue to exist in its current form without radical steps like selling some parts.

⇡#Disaster of the Year: Intel Raptor Lake

Even those who don’t follow corporate news could clearly see that something was wrong with Intel. In 2024, a serious scandal broke out around the company’s flagship processors for desktop systems – users began to massively complain about the degradation and failure of older Core models of the 13th and 14th generations.

The first reports of emerging and growing problems in the stability of Raptor Lake appeared at the beginning of the year. The most common reason for complaints was the unexpected appearance of “blue screens” that occurred even when the system was operating in nominal mode. Motherboard manufacturers have been struggling with this problem for some time, bringing the default settings in line with Intel recommendations. But that didn’t really help. Intel identified the main cause of the problems by the beginning of summer. It turned out that due to errors in the microcode, under certain conditions, the processor begins to request higher and higher voltages, and this ultimately leads to degradation of the silicon crystal and instability even in nominal mode. It took the company some time to find a solution, and as a result, BIOS versions that finally eliminated errors in the implementation of internal voltage control algorithms were released only in early autumn.

However, one important question remains in limbo. The fact is that silicon degradation is an irreversible process. This means that even once a processor encounters overvoltage, it automatically falls into the “risk group” and may begin to malfunction or fail after some (sometimes quite a long) time. Unfortunately, Intel could not find any technical measures to identify such cases and simply offered to replace processors under warranty to all users who applied. To quell the wave of dissatisfaction that arose, the warranty period for Raptor Lake was increased from three to five years, but this, of course, does not completely solve the problem.

The discovery of certain defects in already released products is a fairly common situation that periodically arises in the history of any company. However, the case with Raptor Lake became egregious and seriously damaged Intel’s reputation. Firstly, the company did not respond properly to user reports for a long time, did not admit guilt and tried to make the users themselves or motherboard manufacturers responsible for CPU degradation. Secondly, an investigation into the causes of degradation revealed a whole bunch of unrelated errors in various internal algorithms, the elimination of which took several months and required the release of several versions of the microcode. And thirdly, users have not received an answer to the question of how they can understand whether a specific processor instance has been affected by degradation or not.

Thus, this whole story not only revealed shortcomings in the organization of pre-release testing of Intel’s new products, but also showed that the company’s technical support is not capable of responding to widespread problems in a sufficiently qualified and prompt manner. This is precisely the destructive power of this story – the company’s customers became convinced that Intel was not trying very hard to protect them from potential problems. And moreover, it is obvious that representatives of the Raptor Lake family will, one way or another, now have an increased failure rate, and they will have to deal with this for a long time.

In addition, there is no answer to the main (rhetorical) question. How could it possibly happen that for a year and a half, Intel supplied the market with processors with defects affecting two successive generations of products?

⇡#Worst processor of the year: Intel Arrow Lake

The troubles that Intel got into also affected new products. The launch of the new generation of Core Ultra 200S desktop processors, codenamed Arrow Lake, which replaced the problematic Raptor Lake, was another failure for the company. Although these processors brought a lot of new things from a technological point of view, they were received rather coldly by users. The reason is in performance, which has not increased so noticeably compared to the performance of Raptor Lake, and in some cases (for example, in games) has even seriously deteriorated.

Even during the preparation of Arrow Lake for release, Intel warned that the main goal at this step of evolution for it was to reduce the energy appetite of the CPU, and not to improve performance. But users did not appreciate the relatively small improvement in efficiency at the cost of a drop in performance, and Arrow Lake was met with sharply negative reception. Therefore, immediately after the announcement, Intel had to make excuses that the release of Core Ultra 200S “did not go according to plan” and promise to correct the situation by adjusting the microcode and drivers. Which, frankly speaking, only worsened their image, since it caused obvious associations with the epic with the repair of Raptor Lake, which has not yet been erased from memory.

In addition, there are quite serious doubts that the characteristics of Arrow Lake will noticeably improve after the release of all updates. Their performance suffered due to a number of circumstances that cannot be corrected programmatically. Firstly, Arrow Lake does not support Hyper-Threading technology. Secondly, in the pursuit of energy efficiency, they lost clock speeds. And thirdly, in their memory controller, due to its distance from the processor cores, the delays have increased significantly. Therefore, it is very likely that the new generation of Intel desktop processors will no longer become popular and will remain “passing,” especially since the transition to Core Ultra 200S requires the use of completely new motherboards with the LGA1851 socket.

However, speaking about Arrow Lake in the context of the year’s results, one cannot help but admit that technologically it is a very promising product. Processors of this family have moved to P- and E-cores with progressive Lion Cove and Skymont architectures with a significantly increased IPC. And if in the case of Lion Cove we are talking about “only” a 9% increase in specific performance compared to the cores of the previous generation, then the new efficient Skymont cores received performance level of the P-cores of processors of the Alder Lake and Raptor Lake generations.

In addition, following the Meteor Lake and Arrow Lake mobile processors, they switched to a chiplet (tiled in Intel terms) layout. The new CPUs include four semiconductor chips (CPU, GPU, SoC and I/O), assembled into a single unit on a silicon substrate using Foveros technology. Moreover, all the main components of Arrow Lake are produced not by Intel independently, but with the help of TSMC – using the most modern technical processes. This step allowed Intel to even temporarily get ahead of AMD in the race of technological standards – the main Arrow Lake crystal is produced at 3 nm standards, while the most advanced competitor processors still use 4 nm technology.

And finally, Arrow Lake became the first Intel desktop processors with a built-in NPU – a special hardware accelerator for neural network tasks. However, in this case, its addition is more like a formal step. The performance of this accelerator is only 13 TOPS, which is three times lower than Microsoft’s requirements for Copilot+ PC. Accordingly, Arrow Lake is not suitable for running the entire set of Windows 11 AI functions (including Recall), and its NPU can only be used in some optimized applications.

⇡#Surprise of the year: Intel Lunar Lake

But to say that Intel was unable to release a single successful product over the past year would be unfair. The Core Ultra 200V or Lunar Lake family of mobile processors, in which Intel took rather unexpected steps, turned out to be an excellent foundation for modern thin and light laptops. At the same time, with their release, Intel was by no means trying to take another step in the competition with AMD. Lunar Lake has a completely different goal – they were a response to the emergence of the Windows on Arm platform and primarily to the Snapdragon X Elite processors. And this answer was more than convincing: Intel was able to clearly show that x86 processors can be both quite productive and energy efficient, and they are not accompanied by any compatibility problems, which one way or another have to remember when it comes to running Windows on Arm processors.

In technical texts, Lunar Lake is usually called a mobile relative of Arrow Lake, but this is quite a stretch, since these processors do not have much in common: perhaps only the architecture of the Lion Cove and Skymont cores, plus production at TSMC enterprises. The differences in reality are much greater, because Lunar Lake is a 17-W CPU with a 4P+4E nuclear formula, powerful integrated graphics on the Xe2 architecture (Battlemage) and a large NPU unit with a performance of 48 TOPS, which not only fully meets the requirements of Microsoft Copilot+ PC, but and outperforms the NPU in Snapdragon X Elite.

Accordingly, Lunar Lake and Arrow Lake do not have any of the same silicon components. Although the mobile processor, like its desktop counterpart, is almost entirely manufactured at TSMC using a 3nm process technology, the crystals in it are completely different, and there are only two of them – a computing one (with a CPU, GPU and NPU inside) and a platform controller (responsible for the operation of Wi-Fi). -Fi, Thunderbolt and PCIe). But the most interesting thing is that in Lunar Lake, Intel for the first time took the example of Apple and placed 16 or 32 GB of integrated LPDDR5X-8533 memory in the processor. However, although this turned out to be useful for both increasing productivity and reducing consumption, the company immediately announced that it would no longer do this, since it was painfully expensive (does Apple really produce processors for Mac at a loss?).

One way or another, Lunar Lake turned out to be quite productive: the eight-core representatives of the Core Ultra 200V series were able in some cases to surpass even the 16-core Meteor Lake in computing tasks and turned out to be one and a half times stronger than them in the graphics load. Of course, they lose to the older multi-core versions of the Ryzen AI 300 in creative applications and games, but, firstly, they are much more economical, and secondly, the lag is not too significant and is not always observed. In other words, Intel really has a good balanced processor for thin and light laptops.

However, it is not yet possible to say that Lunar Lake has become a bright spot for Intel in dark times, since it lacks widespread adoption in real laptops for unconditional market success. Unfortunately, mobile computer manufacturers have not been very warm to this processor, since it deprives them of the usual flexibility in choosing components. As a result, Core Ultra 200V can currently be found in a fairly small number of systems. But the situation may still change in the coming year, at least Lunar Lake should be included in the new generation of Microsoft Surface in 2025.

⇡#AMD’s biggest mistake of the year: Ryzen 8000G

Do not think that against the backdrop of the turmoil plaguing Intel, AMD was successful in all its endeavors. The company actually made fewer mistakes, and they were not so fatal, but they still happened. The most notable is the release of desktop APUs of the Ryzen 8000G series (Phoenix), which were rather coldly received by the market and, despite the absence of any obvious flaws, were unable to repeat the success of their predecessors – the Ryzen 5000G hybrid processors (Cezanne).

When preparing the Ryzen 8000G, AMD decided to follow the beaten path – to transfer the quite successful Hawk Point mobile processors to the desktop segment, combining up to eight Zen 4/Zen 4c computing cores on a 4-nm monolithic chip, high-performance integrated graphics with RDNA 3 architecture and AI- Ryzen AI accelerator. And formally, such a combine looked attractive – AMD even said that the older representatives of the Ryzen 8000G series allow you to assemble systems that are fully suitable for 1080p games without discrete graphics, which is generally true.

However, users were skeptical about this proposal. Such hybrid processors could be an interesting option for entry-level gaming builds, but AMD set prices for them such that no savings were out of the question. Additionally, the situation was aggravated by the need to use rather expensive Socket AM5 motherboards and DDR5 memory with the Ryzen 8000G. As a result, buying a new generation APU obviously became a financially losing enterprise – a more productive configuration with inexpensive discrete graphics of the Radeon RX 6500 XT level could have cost significantly less.

The NPU unit included in the Ryzen 8000G did not save the situation either. Its performance is only 16 TOPS, which means it doesn’t meet Microsoft’s requirements and has a fairly narrow scope of applicability. As a result, even today, the Ryzen 5000G series of processors from three years ago remains more in demand among buyers than progressive APUs from 2024. And moreover, AMD itself seems to have realized the failure of the desktop Ryzen 8000G, as it continues to extend the life cycle of the Ryzen 5000G series by releasing additional models with a very attractive price.

⇡#Best Mobile Processor of the Year: AMD Ryzen AI 300

It’s no exaggeration to say that 2024 marks a major milestone in the evolution of laptop processors. The emergence of the Windows on Arm platform forced x86 chip manufacturers to act more boldly, which had a positive effect on the characteristics of all mobile new products. This is also true for Intel Lunar Lake, but even more striking new products that can radically change ideas about the capabilities of modern laptops are AMD Strix Point and Ryzen AI 300. In these CPUs, the developers did not chase exceptional energy efficiency and increase the battery life of any price, but put performance at the forefront (without forgetting about everything else). As a result, the Strix Point series chips, while maintaining a configurable TDP in the range from 15 to 54 W, fell into a new weight category and offer up to 12 productive cores simultaneously with powerful graphics based on 16 CUs.

Ryzen AI 300 is strikingly different from all those laptop chips that AMD has offered previously. And the point is not so much in the new architecture of computing cores (Zen 5) and integrated graphics (RDNA 3.5), but in the fact that the monolithic 4-nm Strix Point crystal now holds much more, because mobile processors of previous generations of the same class received no more than eight cores and graphics with a maximum of 12 CUs. It is not surprising that, having increased their arsenal, the new Ryzen AI 300 can easily match the performance in applications of solutions like the Core i5-13500HX or Ryzen 7 7745HX with a TDP of 55+ W, and in terms of performance in games they are comparable to the desktop GeForce GTX 1650, allowing some reduction in picture quality when playing on integrated graphics in modern AAA games 1080p resolution.

AMD was able to achieve such radical progress by changing its design approach. Ryzen AI 300 actually became the company’s first full-fledged attempt not only to assemble a laptop processor from available parts, but to do this by first modifying them to better suit its goals. As a result, Strix Point is based simultaneously on two types of cores – productive Zen 5, of which there are only four in the processor, and energy-efficient Zen 5c. The second version of the cores does not differ from the first in terms of IPC, but has a quarter smaller size due to the compaction of transistors on a silicon chip. This reduces their maximum frequencies by one and a half times, but increases energy efficiency and allows you to pack a larger number of them inside a monolithic mobile chip – up to eight pieces.

In addition, the mobile version of the Zen 5 architecture is different from Zen 5 for desktop processors. In order to reduce consumption and reduce the area of ​​cores, AMD abandoned 512-bit work with vector instructions in Strix Point. Therefore, mobile processors based on Zen 5 architecture operate with AVX-512 instructions in the same way as their predecessors based on Zen 4 – in two steps. And this is not the only such moment. The architecture of the RDNA 3.5 graphics core used in Strix Pont is also exclusive. Here we are talking about an adapted version of RDNA 3 with additional optimizations in terms of memory bandwidth use and energy saving.

Another unique feature of the Ryzen AI 300 is the built-in NPU hardware unit, in honor of which these processors received the AI ​​suffix in their names. It has a current record performance of 50 TOPS (and in some versions of Ryzen AI 300 – 55 TOPS), which allowed AMD not only to fit into the requirements of Microsoft Copilot+ PC with a margin, but also to beat the solutions of Apple, Intel and Qualcomm in hardware AI capabilities . However, so far this leadership has not yielded clear results. Microsoft is stuck with the development of AI additions to Windows 11, as a result of which the ambitious Recall function started working on the Ryzen AI 300 in test mode only in December, and other uses for the built-in NPU are not yet visible…

Moreover, despite the strengths of the Ryzen AI 300, AMD processors have failed to make it into Surface devices. Between 2019 and 2022, Microsoft used AMD solutions for its laptops, but in the past year its preference has shifted to the Snapdragon X due to its exceptional power efficiency. And this is a rather disappointing loss for AMD, which is actively fighting to increase its share in the mobile market. But Ryzen AI 300 still received a fairly warm reception: processors in this series can be found in devices from Acer, Asus, HP, MSI, Lenovo and other major manufacturers.

⇡#Twist of the Year: AMD Ryzen 9000

The Ryzen 9000 family of desktop processors, based on the same Zen 5 architecture, promised to be as successful as the Ryzen AI 300. In preparation for its release, AMD boasted of a 10-20% performance increase compared to its predecessors while simultaneously reducing energy consumption. However, in reality, the release of the first models of the Ryzen 9000 series turned out to be a disappointment for users: the company’s promises were justified only in terms of energy efficiency, and the real improvement in the performance of the new products was limited to a few percent.

This happened because AMD literally strangled the Zen 5 architecture in the new processors, despite the fact that it initially included a 16% improvement in IPC, which is quite consistent with the traditional performance increase when changing generations. But Ryzen 9000, compared to Ryzen 7000, received lower clock speeds, which, although it had a positive effect on consumption and temperatures, significantly reduced the possible advantage of the new products. In addition, AMD refused to upgrade the memory controller in Ryzen 9000, which again artificially limits bandwidth and does not allow one to take advantage of the transition to high-speed DDR5.

As a result, the new processors for desktop systems were met with rather indifference by the public, and in order to save sales, AMD even had to drop prices by 15-25% relative to the initially established values. In addition to this, the company slightly corrected the performance situation with the release of AGESA updates and Windows 11 scheduler patches. These measures helped to smooth out the initial negative a little, and at the end of the year, the older representatives of the Ryzen 9000 series were even able to enter the top twenty (but not the top ten) CPUs, the most in demand among retail buyers.

However, despite the dubious debut of Zen 5 architecture carriers in the desktop segment, AMD was still able to turn the situation around and confirm its status as a manufacturer of the best gaming processors. The plot twist occurred thanks to the appearance at the end of the year of the eight-core Ryzen 7 9800X3D, enhanced with 3D cache, which created a real sensation, since it not only became the fastest CPU in modern games, but also provided a hitherto unimaginable lead over competitors, reaching 20-25%.

AMD helped achieve this by rethinking 3D V-Cache technology, which involves a multiple increase in the volume of the L3 cache of a conventional processor due to an additional SRAM crystal. In the Ryzen 7 9800X3D, for the first time this crystal was placed not on top of the processor cores, but below them, which allowed AMD not to be afraid of local overheating and not to limit the clock speed of the new product. As a result, the new eight-core processor with 3D cache easily outperformed all other Ryzen 9000 models in base frequency.

Thus, AMD ended last year with a convincing victory in the desktop CPU market. And this victory seems especially crushing against the backdrop of the competitor’s failures with Raptor Lake and the dubious announcement of Arrow Lake. But even if we ignore what is happening in the “blue” camp, the Ryzen 7 9800X3D became exactly the processor that the vast majority of users wanted from AMD – productive, with moderate heat dissipation and not too expensive. Perhaps the only drawback that turns out to be inherent in the Ryzen 7 9800X3D is its deficiency, which AMD has not been able to satisfy at the moment. However, there is nothing strange about this: outstanding products always enjoy increased popularity.

⇡#Debut of the year: Qualcomm Snapdragon X

Until recently, Qualcomm was perceived as a company that produces processors for smartphones. Of course, there were exceptions, such as the Snapdragon 8cx, but in any case, Qualcomm was not trying to become a significant player in the PC market. However, now everything has changed. After acquiring the startup Nuvia, the company acquired the powerful Oryon core with Arm architecture, developed by Apple, with which Qualcomm began aggressive PC expansion. With an eye on Windows laptops, the company launched the 12-core Snapdragon X Elite and 10-core Snapdragon X Plus processors last year.

These processors cannot be called adapted mobile chips – they are initially aimed at Windows PCs, and signs of this are easy to see in their structure. In particular, Qualcomm abandoned the big.LITTLE concept, making all cores homogeneous in architecture. Moreover, to improve energy efficiency, they are grouped into quad-core clusters, each of which has its own maximum frequency and a common 12-MB L2 cache. The processor also has a 6 MB L3 cache, which can be used not only by Oryon cores, but also, for example, by integrated Adreno X1 graphics. By the way, it is also adapted for working on a PC – it has support for DirectX 12.1, hardware ray tracing and a familiar (not tile-based, like in mobile Adreno) rendering pipeline. And the theoretical performance of such a GPU is increased to 4.6 Tflops, which is comparable to the performance of integrated Intel Meteor Lake graphics.

But the most significant advantage of Snapdragon X, which spurred a wave of interest in them, was the presence of a powerful NPU with a performance of 45 TOPS. Thanks to the hardware AI engine, Qualcomm was able to enlist the support of Microsoft, which chose Snapdragon X as the base solution for AI computers of the Copilot+ PC standard. As a result, the Microsoft Surface Pro 11 and Surface Laptop 7 laptops based on these processors clearly confirmed the seriousness of Qualcomm’s ambitions – they demonstrated impressive autonomy (comparable to the autonomy of modern MacBooks), excellent performance (at the level of solutions on Meteor Lake and Hawk Point) and even the ability to execute modern games at medium quality settings (albeit with a slightly lower FPS compared to Meteor Lake).

However, Snapdragon X also has plenty of problems. The main one concerns the compatibility of familiar software with Windows systems based on processors with Arm architecture. Microsoft tried to smooth out this issue to the best of its ability and implemented the Prism software emulator in the OS. However, it is far from perfect: some x86 software in systems running Snapdragon X does not work at all, and some significantly lose performance. A separate layer of inconsistencies arises when launching gaming applications. Nevertheless, all such problems are gradually being resolved, and at the same time the list of native Arm applications is growing. The growing cooperation between Qualcomm and Microsoft is also making itself felt. Laptops with Snapdragon X processors are currently the only class of PC on which the Windows 11 Recall AI function fully works.

However, there is not much reason to talk about Qualcomm’s successful invasion of the Windows PC market yet, because as of November, only 720 thousand laptops based on Snapdragon X were sold. The x86 universe is giving a serious rebuff to the stranger, and the new AMD Ryzen AI 300 and Intel Core Ultra 200V with built-in NPUs could become a worthy alternative to the Qualcomm solution, not inferior in the ratio of performance and autonomy and without any problems with software compatibility.

But it’s still very early to put an end to the history of Snapdragon X. In the coming year, Qualcomm is going to increase the supply of systems on its processors by more than five times, and the company has everything it needs to implement such a plan. So far, Snapdragon X processors could only be used in thin and light laptops without discrete graphics and at a fairly high cost. But the company is working to reduce the price of laptops on its processors to $800 and has already presented an inexpensive eight-core version of the Snapdragon X. Moreover, in the future such laptops should become even more affordable – Qualcomm expects to set the lower price level at $600. At the same time, the company’s idea is that low-cost Windows PCs based on Arm architecture should not lack a powerful NPU, which means that even budget computers with Qualcomm processors will meet the requirements of Copilot+ PC and will be able to work with the AI ​​functions of Windows 11.

⇡#Result of the year: AMD is becoming more popular

The crisis in which Intel found itself affected not only the financial position of the company. He undermined its position in the market. Savings gradually began to lead to a deterioration in the consumer qualities of processors, and this has already resulted in a noticeable decrease in its market share. Historically, Intel owned 80% of the x86 processor market – this situation has been observed for the last several years, and AMD could not increase its share with any announcements. However, during 2024, the situation began to change rapidly, and the ratio of Intel and AMD shares in the desktop processor segment shifted to 70 to 30.

Source: Mercury Research / tomshardware.com

Obviously, this happened not so much due to the appearance of the Ryzen 9000 family, but because of problems with degrading Raptor Lake processors, which Intel was unable or unwilling to quickly fix. Recognizing these problems, finding their source, and fixing the defects took nearly nine months, during which time users of the Intel desktop platform suffered from both a lack of information from the manufacturer and a lack of understanding about whether they should worry about their existing CPUs. All this seriously undermined confidence in Intel products, and Arrow Lake, released towards the end of the year, could not give a clear answer whether the problems with Raptor Lake were a tragic accident, or whether it was an external manifestation of a systemic crisis within Intel. After all, with Arrow Lake, the company seems to have stepped on an old rake again: the processor was released in a raw form, and now Intel again has to fix problems using microcodes, however, this time they are related not to stability, but to performance.

It would seem that Intel’s advantage as a leading player in the desktop processor market continues to be overwhelming, and it will be quite difficult to lose its more than twofold sales advantage over AMD. However, do not delude yourself: a string of Intel failures can quickly weaken its position, especially since the trend is obvious – users easily change their preferences under the influence of circumstances. Thus, since the end of 2023, Intel’s share in the desktop segment has fallen from 80.2 to 71.3%. And 2025 could easily be a tipping point: support for Windows 10 ends next October, forcing enterprise customers to replace millions of PCs. In whose favor in the current realities they will make their choice is far from obvious.

However, Intel did not make catastrophic mistakes in the mobile market segment, and the situation there is more or less favorable for it. Here the pressure from AMD is not so strong, and the company’s share continues to remain at 77.5%. However, due to the release of successful new products such as the Ryzen 8000 Hawk Point and Ryzen AI 300 Strix Point, AMD was able to increase its share from 16.4 to 22.5% during 2024. It seems that the gradual shift of interest in favor of AMD’s mobile CPUs will continue to occur. At least, AMD’s laptop processors presented at CES 2025 looked somewhat more convincing than Intel’s. But let’s not get ahead of ourselves.

⇡#Instead of a conclusion: new processors in the first days of 2025

Having summed up the results of last year, let’s look at what we entered 2025 with. The fact is that just the other day the CES 2025 exhibition took place, which is traditionally used for the main announcements of new equipment. And judging by what CPU manufacturers presented this time, we can say with confidence that in the coming year the main battles in the processor market will again take place in the mobile segment.

Here Intel is going to deploy an army of mobile versions of Arrow Lake, which in this version look clearly more interesting than in desktop form. Core Ultra 200H series processors with a 6P+8E+2LE nuclear formula and powerful graphics will have to replace Meteor Lake in high-performance laptops, and desktop-like Core Ultra 200HX processors with an 8P+16E formula will traditionally be aimed at gaming laptops with discrete graphics cards. In addition, Intel, based on the Meteor Lake design, has prepared a separate Core Ultra 200U series with a 2P+8E+2LE formula – such processors should become a more affordable option for thin laptops compared to Lunar Lake (but with a reduced GPU and NPU).

AMD, in response, is betting on expanding the Ryzen AI 300 lineup downward, and is also offering a fundamentally new class of Ryzen AI Max mobile processors, which will receive not only up to 16 Zen 5 cores, but also the most powerful integrated graphics based on 32-40 CUs, which corresponds to the level of the Radeon RX 7600. It is difficult to predict whether such solutions will gain popularity, but AMD itself presents them as the most advanced mobile processors modernity, head and shoulders superior not only to the products of the main competitor, but also, for example, to the Apple M4 Pro and M4 Max processors.

In addition, AMD is going to separately adapt its Ryzen 9000 processors for laptop applications. The company promised to release a 54-W analogue of the Ryzen 9 9950X and a mobile analogue of the Ryzen 9 9950X3D within a year. However, as experience with mobile variations of 16-core desktop processors of the previous generation shows, such CPUs do not find wide support among manufacturers and only end up in a few extremely expensive gaming laptops.

In addition to classic laptop processors, AMD has updated its line of solutions for portable gaming consoles, announcing the Strix Point-like Ryzen Z2 on Zen 5 architecture and with RDNA 3.5 graphics based on 16 CUs. Over the past two years, AMD has managed to firmly establish itself as a major supplier of chips for handheld consoles with its Z1 processors, and updating them should help the company not lose its dominant position in this new niche.

It is curious that Qualcomm, which is trying to compete with AMD and Intel in the mobile segment, reacted rather sluggishly to this entire cavalcade of announcements. It only expanded the Snapdragon X line with a younger eight-core model. But this seemingly minor announcement could play an important role in Qualcomm’s ability to attack the market position of x86 competitors. The fact is that the new chip will significantly reduce the cost of laptops on the Windows on Arm platform, and this could become a catalyst for its distribution.

As for the desktop segment, AMD confidently seized the initiative and left no chance for its competitor. While Intel only announced an addition to the Arrow Lake lineup, adding models not designed for overclocking to the already announced processors with the K index, AMD decided to put an end to the issue of dominance in the field of performance. The presented 12- and 16-core Ryzen 9000X3D with 3D cache promise not only a double-digit percentage advantage over the older Arrow Lake in games, but also a similar advantage in content creation and processing tasks. With this announcement, AMD literally knocks Intel down, from which it will obviously be very difficult for the “blues” to get out of it, at least until the release of desktop processors of the Nova Lake generation, scheduled only for 2026. This means that real competition in the desktop segment in the near future will only be possible in the middle and lower price segments.