Intel on fire: time to change technological meanings

A significant decline in the value of shares and total capitalization, large-scale layoffs, problems with the stability of processors, class action lawsuits, and delays in the release of key products. These are all pressing issues for the semiconductor giant Intel. One of the global leaders in the electronics industry is clearly going through a difficult period in its history. What factors led to the protracted crisis, and does the company have any prospects of overcoming the difficulties in the near future?

It seems as if the company has lost some of its visionary flair. In some cases, Intel has to play the role of catch-up, at least when it comes to new, promising segments that are just emerging and have the potential for rapid growth.

A large, inflexible structure reduces the speed of reaction and increases the time required to make important decisions. Having its own full-cycle production facilities is both an undeniable advantage and a burden for the company. The ability to control the entire development and production process – from the idea to the shipment of devices – has its obvious advantages.

However, when it comes to the highly complex segment of modern electronics, which requires constant technological progress, at some point the capabilities of the foundry and product divisions become out of sync. If production fails to keep up with the demands of the engineering corps, well-known problems begin to arise: annoying postponements of product launches, weakening of performance, and cost adjustments.

Intel processors for smartphones

Only the lazy have not criticized Intel for not having its own processors for smartphones. Indeed, this is a fact, but it is also a circumstance that requires a more detailed historical excursion. In 1997, Intel acquired part of the production facilities of Digital Equipment Corporation (DEC), as well as the StrongARM division. This deal included ARM technologies that were later used to create the Intel XScale processor line.

XScale chips were one of the most popular solutions in the heyday of handhelds. Processors of the Intel PXA line with ARM architecture could be found in the platforms of the best-selling models of that time, especially among Windows Mobile solutions. They were used in the HP iPAQ, Dell Axim, Fujitsi Siemens Pocket Loox, and some versions of PDAs from Palm and other manufacturers.

In the early 2000s, Intel XScale processors held a significant share of the PDA market, reaching up to 60% at certain times. However, even at the peak of their popularity, PDAs remained rather niche devices, and the number of PDAs sold was 10-15 million per year. This was probably one of the reasons why Intel agreed to sell its XScale division to Marvell Technology in 2006 and concentrate on its core business of x86 processors for PCs and servers. The $600 million received from the sale was obviously appropriate in those circumstances, but it was probably not Intel’s most successful deal.

In early 2007, Apple introduced the first iPhone, which revolutionized the way smart phones were used and their capabilities. After that, the term “smartphone” became commonplace for devices of this class, which began to develop rapidly.

At that time, Intel was actively developing the netbook business – affordable mobile systems for basic needs, as well as more powerful ultra-mobile PCs (UMPCs). For these categories of devices, Intel Atom family chips were used – inexpensive, cost-effective processors that were usually sufficient for simple user tasks.

So the idea of optimizing Intel Atom for smartphones seemed quite logical. Intel’s first attempts in this direction were made back in 2010, but the introduction of Atom Z2xxx chips at the end of 2011 can be considered a purposeful step. The processors had a fairly decent level of performance in their class, but the x86 architecture still imposed certain limitations.

The use of the then-current 32-nanometer process did not allow for high battery life for devices based on the first Intel Atom Z2xxx, and the software that was originally written for ARM required additional optimizations and the use of a binary translator. Although most applications from Google Play worked without any problems, the presence of exceptions did not add to the attractiveness of Intel Atom smartphones. That’s one of the reasons why the range of models based on these processors was rather modest, despite Intel’s strong support and incentive programs.

The company managed to attract several global manufacturers (Lenovo, ASUS, Acer) to cooperate, but mostly local brands with relatively small smartphone production volumes were ready to experiment with x86 chips.

In 2014, Intel updated its line of processors, introducing more productive Atom Z34xx (Merrifield) manufactured using 22-nanometer technology. A year later, the manufacturer offered Atom X3 (SoFIA) models with advanced functionality and 3G/LTE support. However, these chips also did not receive wide support from smartphone manufacturers. As a result, the ambitious project with x86 processors for mobile phones was completed. In 2016, Intel went through another restructuring with the dismissal of a large number of employees (~12,000). An attempt to expand into the smartphone market, despite the considerable costs of developing the ecosystem, supporting software developers and smartphone manufacturers, did not bring the desired result, so further actions in this direction were stopped.

In the course of further analysis of the situation, in discussions about “lost polymers,” Intel’s management has probably repeatedly mentioned the decision in 2006 to get rid of XScale and focus on “more promising areas.” Of course, we can’t say for sure how Intel’s history in the smartphone processor space would have developed if the XScale ARM chip division had remained part of the company.

In fact, Intel is now left out of the processor race in the dynamic and very large smartphone market. Given the current trends, we are unlikely to see another attempt by the company to enter this segment with modern chips based on the x86 architecture in the near future. As for the possible return of experiments with SoCs based on the ARM architecture, Intel is currently only open to contract manufacturing of such processors for external customers.

The recent news about Intel’s decision to divest its stake in Arm Holdings and the sale of 1.18 million shares (~$150 million), which, by the way, also fell by ~30% over the past month, is another confirmation of the unchanged priorities.

Refusal to develop the Larrabee project

According to the current Intel CEO Patrick Gelsinger, who does not resist openly criticizing not only competitors but also the decisions of his predecessors, another big mistake of Intel was the abandonment of the Larrabee project, the development of which could potentially allow the company to occupy a much better position in the market of discrete graphics and AI computers.

The Larrabee project was an in-house development of Intel and began in 2006 as an attempt to create a hybrid processor that would combine the features of a CPU and a GPU. The basic idea was to use many simplified x86 cores connected by a high-speed bus that could perform graphics and computing tasks in parallel. These cores were based on the Intel Pentium (P54C) processor architecture and supported extended SIMD instructions to improve graphics and parallel computing.

Larrabee remained at the conceptual level, without being commercialized. At that stage of development, the software realization of Larrabee’s potential was quite difficult, and the project had poor energy efficiency. As a result, in 2010, Intel decided to stop working on Larrabee, using certain developments for Xeon Phi accelerators, which remained relevant until 2018.

However, Gelsinger is confident that Larrabee’s original idea had much more potential and that further development of the area would have brought significant benefits to the company. Due to past “bad decisions” over the past 5-10 years, Intel had to spend a lot of money to acquire several HPC and AI companies. Moreover, not all acquisitions were successful. For example, the developments of Nervana Systems (2016, $350 million), which were embodied in Intel’s Nervana Neural Network Processor products (NNP-T and NNP-I), were curtailed in favor of Habana Labs projects in a few years. The acquisition of the latter cost Intel $2 billion in 2019. And now, it is Habana Labs’ developments embodied in the Gaudi accelerator that are the main products of Intel in the specialized AI computing segment.

Thus, in the absence of its own basic developments of projects with a stable and consistent history of development in a rapidly growing segment, Intel has to rely on the developments of third-party companies/startups, switching from one to another. Such a strategy also has the right to live, allowing to choose the most promising solutions, but it requires considerable financial efforts and additional time to implement real products.

Delays in production modernization

Probably the main problem for Intel was the difficulty in implementing new advanced technologies for manufacturing silicon crystals. The leader of the semiconductor industry 10-15 years ago used the most advanced technical processes for its products and had an impressive plan for gradual modernization. The famous Tick-Tock concept, introduced by Intel back in 2006, involved annual alternate process changes and updates to the internal architecture of the CPU.

At the first stage, it was proposed to switch to a “thinner” process, and a year later, to update the microarchitecture using the new process. Then the cycle would repeat again. Intel managed to maintain such a frantic pace for almost a decade when it came to tens of nanometers and 65 nm processes at the start of the Tick-Tock concept. Already at the 22 nm stage in 2014, an additional year’s optimization was required, which disrupted the tight schedule. But with the transition to 14 nm, it became clear that the seemingly trouble-free mechanism was failing, and the difficulties at the previous stage were not an accident.

The transition to 14 nm for desktop chips occurred with the production of very specific Broadwell processors in 2014, and the mass-produced Skylake processors (Core 6xxx) were introduced in 2015. After that, a long period of improvement of the existing process began, and we saw the first CPUs for desktop PCs based on 10-nanometer technology (Alder Lake, Core 12xxx) only in 2021. That is, six generations of Core processors later.

Intel experienced significant difficulties with the transition to 10 nm. At this time, the company was trying to keep its focus and declare its ambitions in new segments, such as autonomous cars, the Internet of Things (IoT), artificial intelligence, and others. Because of this, attention was somewhat scattered, while the direct production of crystals did not receive the necessary resources for development and active progress.

During this time, the main competitors of the “foundry front” – TSMC and Samsung – have significantly strengthened their positions, successfully mastering new technical processes. Therefore, Intel, in its traditional element, quite unexpectedly found itself in the role of a catch-up player. Delays in the development of new technical processes affect the timing of new products and their technical characteristics.

While Intel was only massively introducing 10 nm, Taiwan’s TSMC was already offering partners to order chip manufacturing using 8, 7, 6, and even 5 nanometer technologies (Apple M1 – TSMC 5 nm, 2020).

In 2021, Patrick Gelsinger took over the company and promised to restore Intel’s manufacturing power despite the lost time and missed opportunities. When it comes to such high-tech production facilities, their reorganization and modernization require bold decisions, a lot of time, and significant financial costs. Given that the processes take place in a very competitive environment, Intel decided to establish cooperation with “colleagues in the manufacturing business” by ordering the production of certain necessary components on a contract basis. It may not be the easiest solution for a silicon crystal manufacturer, but new products are a priority, and if it is necessary to temporarily seek help from competitors to partially manufacture them, it is rather a sign of Intel’s prudence.

By the way, during this year’s Computex exhibition, Patrick Gelsinger visited Taipei and personally thanked TSMC for its assistance in manufacturing Lunar Lake mobile chips at an official event. It is worth noting that this is not an isolated example of cooperation between manufacturers. Currently, the Taiwanese company manufactures chips for quite iconic and important Intel products, such as GPUs for Intel ARC graphics cards, components for Meteor Lake mobile processors, and Gaudi AI computers. TSMC will also contribute to the development of Arrow Lake desktop/mobile chips. Paradoxically, due to the aforementioned circumstances, Intel itself had to become a large contract customer. The timing of its own products still takes precedence over competition in the manufacturing field. This phenomenon is temporary and Intel will rely on its own production facilities in the future. There are all the prerequisites for this.

ТIntel’s technological leap

Back in 2021, Intel announced an ambitious plan to modernize its production facilities. The developed 5N4Y (Five Nodes in Four Years) strategy envisaged changing five process nodes in four years. Given the difficulties that the manufacturer had encountered during previous process thinning, such a challenge initially seemed unrealistic.

The starting point is the Intel 7 process, which is actually an improved version of 10 nm. However, these “nanometer” figures have not been directly related to the actual technical parameters of the chip for a long time, but rather are a certain marketing element to indicate the class of the process and reflect the progress of its improvement. In this case, each manufacturer actually has its own interpretation. Of course, this makes it difficult to compare the technical processes of different companies, so the term “nanometer class” is used to simplify.

Returning to Intel’s strategy, the next step is Intel 4 (4 nm class). Crystals based on this technology are already used in the Meteor Lake mobile processors that rolled off the assembly line in 2023. It is according to Intel 4 standards that the tiles with the processor’s computing cores are manufactured.

Intel process version 3 (class 3 nm) is being used to manufacture new Intel Xeon 6 server chips (Sierra Forest and Granite Rapids). The manufacturer will offer several modifications (Intel 3-T, 3-E, 3-PT) with certain features and optimizations for different applications. Moreover, Intel will actively offer these standards to third-party contract customers.

With the transition to Intel 20A (class 2 nm), the company is entering the “Angstrom era,” when further changes will be indicated not in units but in fractions of nanometers (Angstrom is 0.1 nm). In addition to directly improving the process itself, Intel 20A solutions will have several additional innovations. The company is starting to use RibbonFET transistors with wraparound gates (GAA, Gate-All-Around) located on four sides around the channels in the form of nanoribbons. Compared to FinFETs, the new elements have smaller physical dimensions, while significantly increasing the transistor switching speed and making it possible to use lower voltages. In fact, GAA is the largest structural modernization since the advent of “vertical” FinFETs in 2012, when the gate was placed on three sides of the channel. At the time, this was also a significant advance over the planar layout. We have another evolutionary round.

It should be noted that RibbonFET is not a unique Intel development. Samsung started using a similar GAA structure back in 2022 for its 3 nm process. In the Korean manufacturer’s variation, the technology is called MBCFET. In its turn, TSMC also plans to introduce GAA (TSMC’s Nanosheet) for its 2-nanometer N2 process in 2025.

Intel is ready to surprise in its efforts to catch up/get ahead of its main competitors. In addition to the structure of RibbonFET transistors, the Intel 20A process will also use the new PowerVia technology, which provides for the organization of power supply from the backside of the crystal (BSPDN, Backside Power Delivery Network).

This approach allows solving the important issue of interconnects by optimizing signal transmission on the front side and using the back side of the silicon wafer to switch the power circuit network. This also helps to reduce the voltage drop across the resistive resistance (IR drop), increasing performance at a given voltage level.

Foundry partners also have plans for BSPDN. TSMC is going to implement its Super Power Rail variant with backside power supply in the TSMC A16 process (class 1.6 nm), which will be preliminarily used in mass production in the second half of 2026. Super Power Rail is technically somewhat more complex and may be even more efficient, but by that time Intel will have already offered an improved version of PowerVia.

Samsung will also not stay away from the technical innovation, but will preliminarily offer its BSPDN option for the 2-nanometer SF2Z and SF1.4 (class 1.4 nm) process, which will be available only in 2027.

Thus, Intel has a certain window of opportunity. If the implementation of the planned transformations is successful and timely, the company will gain certain advantages in the manufacturability of silicon chips in the near future.

The first practical implementation of Intel 20A has already been determined. This process will be used to manufacture crystals (tiles) with computing cores for Arrow Lake mobile and desktop processors. The official announcement of these CPUs is expected in the near future, and the first deliveries are expected in the fall of this year (October/November 2024).

The next technological step is Intel 18A. It has further improved characteristics and is already in the 1.8 nm class. The process will also use RibbonFET and PowerVia technologies. The first products based on Intel 18A will be the next generation of Panther Lake mobile processors, as well as Clearwater Forest server chips with up to 288 computing cores.

Recently, processors of both lines passed an important technological milestone – the successful launch of the operating system. So, the development is going according to plan. Mass production of the chips is scheduled for 2025. At the same time, the company intends to offer Intel 18A to third-party customers.

Each subsequent technological stage involves the improvement of certain characteristics. During the experiments, manufacturers expect to improve the PPA (performance, power, area) indicators, i.e., to increase performance (transistor switching speed), reduce power consumption, or achieve higher performance at the same consumption level. The third criterion is the ability to increase the density of the elements, which will further reduce the crystal area, or to place more transistors on a platform of identical dimensions to implement additional functionality.

In the case of Intel, there is some unevenness in the technological steps, but in general, the 5N4Y priority plan is close to being fulfilled. Clear priorities allow for good results. Suddenly, Intel has turned from a catching-up company to a manufacturer with the most advanced semiconductor manufacturing processes. Although this is a painstaking process, the final results of which can be influenced by many factors, so let’s not jump to conclusions.

It is obvious that the company is trying to use the most advanced production, and not only for the manufacture of its own products. Intel plans to significantly increase its contract manufacturing capacity, offering third-party developers access to its most technologically advanced processes. The company expects to attract demanding customers who need chips with the best PPA performance, such as Apple, NVIDIA, Qualcomm, AMD, and others.

The existing ambitious 5N4Y plan is only the tip of the technological modernization of the manufacturer, Intel has a strategy for further development and improvement of production standards. In 2026-2027, the company will offer the Intel 14A process (1.4 nm class) and its improved version, and in the future it is planned to switch to Intel 10A (1 nm). Frankly speaking, this is a frightening acceleration, given how difficult and time-consuming Intel’s previous transition from 14nm to 10nm was. And here, in a few years, the prospect of “1 nm” looms on the horizon.

Is it even realistic?

To realize its ambitious plans, the company needs the right conditions and material and technical base. Intel plans to invest about $100 billion in the next five years to build new production sites and expand and modernize existing factories. This is a huge amount of money for any manufacturer, so it was very timely for the company to pass the CHIPS and Science Act, which allocates $57.2 billion in subsidies and investments in semiconductor manufacturing and research initiatives.

Given Intel’s position, it can be assumed that a significant portion of the allocated funds will be used to modernize the company’s production. In March of this year, the US government has already allocated $8.5 billion to finance research and development at sites in Arizona, New Mexico, Ohio, and Oregon.

Intel also has many production plans outside the US. Here we can mention the modern cluster in Ireland (Fab 34), which opened in 2023 in addition to the previous Fab 24 in this country, which is used to manufacture crystals using Intel 4 technology. The foundation has already been laid for Fab 29 near Magdeburg in Germany. The fab should start production in 2027 under Intel 14A standards. Intel has also chosen a site near Wroclaw, Poland, to build a chip assembly and testing cluster, which is also due to start operating in 2027.

Several fabs operate in Israel (Fab 28, 28a, 38), and Malaysia has a long and extensive production history, where Intel began to locate its facilities in 1972. There are technological centers and campuses even in China. Industrial facilities are also present in Vietnam and Costa Rica. In other words, Intel’s production geography is very diverse, although this imposes additional challenges during the reorganization.

To modernize production to compete in the semiconductor segment, expensive lithography equipment is needed. For example, an ultraviolet lithography system with a high digital aperture (High Numerical Aperture Extreme Ultraviolet, High NA EUV) will be used to produce chips using the Intel 14A process (class 1.4 nm). The cost of the newest complex ASML TWINSCAN EXE: 5000, which allows to achieve such standards, is about $380 million, which is about twice as expensive as the previous generation EUV installation.

Intel recently made an unboxing video of such a scanner from ASML, which is installed at the D1X factory in Oregon, where it is being debugged. Each such machine is to some extent a work of technological art. ASML will be able to produce only up to 20 such scanners per year. By the end of 2024, the Dutch manufacturer is ready to deliver 5-6 top-of-the-line TWINSCAN EXE: 5000, and all of them are already pre-ordered by Intel. Even the cost of High NA EUV equipment does not stop the company, which has serious plans and needs.

A pre-order for TWINSCAN EXE: 5000 has also been pre-ordered by SK hynix, although it is not yet clear when exactly it will be able to receive its complex. TSMC, on the other hand, has no intention of purchasing new scanners because of their high price. The Taiwanese company is actively purchasing the previous generation of ASML lithography units, assuring that they are quite satisfied with them and the manufacturer has no plans to switch to High NA EUV at least until 2030.

Some interesting facts about TWINSCAN EXE: 5000 – lithography complex weighs 165 tons, and it takes about six months to assemble, debug and finalize the production and involve 250 engineers in the process. Such indicators make it possible to remotely assess the complexity of both the equipment itself and the overall process of planning production modernization if, for example, a dozen such systems are to be implemented at different production sites. But if the end justifies the means and costs, then anything is possible. Obviously, Intel is ready to go this way.

In addition to preparing a strong production base for its own products, the company also plans to offer contract manufacturing services more actively. After modernizing its facilities, Intel will be able to provide access to the most advanced technical processes. There are plenty of potential customers. Probably most large technology companies would not mind having at least an alternative to TSMC, for a variety of reasons. This solves the issues of diversification, new product planning, and manufacturing costs. By the way, TSMC has recently increased the cost of crystal production services using 3/5 nm technologies by ~8%. In the presence of real competitors, the issue of price becomes debatable.

In the current environment, TSMC is the undisputed leader in contract manufacturing of semiconductors with a share of 62%. Samsung holds the second position with 13%, UMC/SMIC have 6% each, and GlobalFoundries has about 5% of this market. Intel’s current position here is insignificant, so for now the manufacturer falls into the general category of “Others” in the relevant rankings. But the company is just beginning to build its contract manufacturing model. It takes time and actual readiness to “bake” silicon wafers of the required quality in the right volumes and in compliance with the planned deadlines. As soon as these conditions are met, there will be a queue of eager customers.

Product set

Which of the recently announced Intel products are worth waiting for in the near future? During the Computex 2024 exhibition in June, the company presented Intel Xeon 6 processors of the 6000E (Sierra Forest) family based on energy-efficient Crestmont cores and 6000P (Granite Rapids) family based on productive P-Cores, which are not interesting for the average user but important for business.

Depending on the number of computing cores, the layout of the chips varies. It should be noted that Xeon 6000E will initially have up to 144 cores, and early next year Intel will offer a model with as many as 288 economical cores. As for the Xeon 6000P, the maximum number of P-Cores is 128. In both cases, the most equipped models (6900P/E) are offered for LGA7529, while 6700P/E and lower chips are offered for LGA4710.

Intel probably doesn’t like the fact that AMD is making significant inroads into the very conservative and profitable server processor segment. This market has traditionally been dominated by Intel, but with the help of EPYC chips, the competitor managed to attract both solution providers and customers of such systems. As a result, AMD has already taken 24% of the server x86 processor market. We will see whether the new Xeon 6 will help to at least partially regain the lost ground by the end of the year.

Intel continues to develop the Gaudi line of specialized AI accelerators. Back in April, the company announced the third generation – Gaudi 3, which offers a significant acceleration compared to the previous model and is able to compete with NVIDIA H100/H200.

Intel Gaudi 3 is manufactured using the TSMC 5 nm process technology and architecturally includes 64 tensor processor cores and 8 matrix mathematical engines. The accelerator is equipped with 128 GB of HBM2e memory (3.7 TB/s) and 96 MB of SRAM (12.8 TB/s). The declared performance is up to 1835 TFLOPs for FP8/BF16.

Despite the appearance of Gaudi 3, the developers are confident that the previous generation version, Gaudi 2, remains relevant. These solutions are much more affordable and are quite suitable for tasks with lower performance requirements. A set of eight Gaudi 2 modules complete with a universal platform (UBB) will cost $65,000. The same set for Gaudi 3 will be offered for $125,000. In both cases, the cost is significantly lower than comparable NVIDIA solutions.

According to Intel’s internal tests, when training a neural network on the LLAMA2-7B and LLAMA2-13B models, the performance of a node with 8/16 Gaudi 3 modules is 1.5-1.7 times higher than that of a platform with NVIDIA H100 accelerators. Comparing the capabilities of clusters with 8192 modules on the GPT 3-175B model, the Gaudi 3 assembly proved to be 40% faster.

On inference tasks, Intel compares the capabilities of a single Gaidi 3 module with the NVIDIA H200. Here, the Intel accelerator is ~10% inferior when using LLAMA-7B and LLAMA-70B, and surprises with a significant advantage when working with the Falcon 180B language model.

Despite the fact that Intel was somewhat late in launching relevant AI accelerators, the company intends to actively develop this area by offering attractive models in terms of price/performance ratio. There is a great demand for such solutions in the market, and many consumers are looking for alternatives to NVIDIA, which dominates and imposes conditions. So if Intel is counting on a significant share of this market, they should accelerate, as large customers such as Amazon, Meta, and Google often rely on their own developments without waiting for third-party solutions.

Intel expects to sell Gaudi for about $500 million by the end of the year. For comparison, NVIDIA will supply accelerators worth $40-45 billion in 2024. Would this ratio have been different if Intel hadn’t stopped its Larrabee project in 2010?

In the case of Intel’s new chips, it is generally encouraging to see how developers are free to experiment with heterogeneous crystals on a single substrate, densely placing several silicon wafers from different manufacturers, even those manufactured using different processes. This is where the experience of using Faveros’ proprietary packaging technologies and EMIB (Embedded Multi-Die Interconnect Bridge) high-speed interconnects comes into play.

The quintessence here is probably the specialized Ponte Vecio graphics computers, which contain 47 (!) functional crystals on a single substrate with a total area of 2330 mm2 and a total number of transistors of 100 billion. As for Ponte Vecchio in general, Intel was frankly delayed with its appearance (unexpectedly). The GPU accelerator with the Xe-HPC architecture was announced back in 2019 and went on sale only in early 2023. Moreover, Intel is already discontinuing these models, focusing on Gaudi 3 and working on a replacement for Ponte Vecchio – Falcon Shores, which was originally intended to be an XPU hybrid with CPU/GPU units, but later approved as a GPU accelerator only. The computer is scheduled to appear in 2025.

On September 3, Intel will officially unveil Lunar Lake mobile processors for the AI PC system concept. A series of cost-effective mobile chips with a functional formula of 4P+4E computing cores, a new graphics core with Xe2-LPG architecture, and an NPU4 unit to accelerate AI algorithm processing. The processors were designed before Intel’s production upgrades, so the chips were manufactured by TSMC using N3 and N6 technologies.

The Core Ultra 200V chips have an interesting layout with the Memory on Package (MoP) concept, with memory chips placed directly on the processor substrate. This structure is not typical for mass-produced x86 chips. The use of LPDDR5X-8533 will certainly allow for increased bandwidth and good overall latency, which are important for unlocking the potential of integrated graphics. This layout will also improve power efficiency and allow the SoC to be used in systems with smaller designs.

The power consumption of the Core Ultra 200V is 17-30 W, and the performance of the NPU unit is 48 TOPS. Thus, systems based on the SoC are fully compliant with the Microsoft Copilot+ PC concept. Intel doesn’t want to lose ground to Qualcomm’s Snapdragon X ARM chips, which are already making a name for themselves in “long-lasting” laptops.

Desktop users are probably already waiting for truly new processors from Intel. Of course, Core chips of the 13th/14th generations are impressive in terms of clock speeds, even despite the 10-nanometer process, but “we’ve been there before.” And sometimes you have to pay too high a price for trying to surpass 6 GHz. So, we have a new LGA1851 platform, new computing architectures, and new manufacturing processes. Arrow Lake-S will be the first chips to be manufactured according to Intel 20A (class 2 nm) standards. Although this technology will be used only for tiles with computing cores. But this is already a reason for jokes about the production of Ryzen 9000 on the “old” TSMC N4 (4 nm).

Regarding the latest CPUs, a more justified reason for reproach may be a rather modest performance increase (~5%) compared to the previous generation chips – Ryzen 7000. However, it is still worth waiting for the performance of Arrow Lake-S. It remains to be seen whether Intel will be able to adequately compensate for the refusal to support Hyper-Threading.

The exact timing of Arrow Lake-S has not yet been determined, but it is expected to be the last quarter of this year. There will probably be more details in the next month, but the top models should be presented in October, and a wider range of 2-nanometer CPUs will appear after CES 2025 early next year.

The Intel Innovation 2024 event, previously scheduled for September 24-25, has been postponed until next year. Perhaps Intel decided that holding a major event would be somewhat inappropriate amid a large-scale reorganization, numerous layoffs, and efforts to improve financial performance. However, the company claims that this will not affect the timing of the expected products – Lunar Lake mobile chips, as well as the new LGA1851 desktop platform and Arrow Lake-S processors. Moreover, Intel is intrigued by statements about the release of a new generation of Battlemag discrete graphics before the end of 2024.

Overall, this is a very good time for interesting solutions to emerge in this segment. The main competitors are not planning any activities and high-profile announcements for this period, preparing their new products for the next year. Unless NVIDIA decides to offer top-of-the-line RTX 50xx with Blackwell architecture, but this is a different class of devices. Therefore, additional activity in the mid-range graphics card niche may be quite appropriate here, especially if Intel manages to take into account certain nuances of the first generation of ARC graphics cards this time around, as well as offer interesting models with a competitive price/performance ratio.

Intel is actively working on optimizing graphics card software, fixing bugs and improving performance in many projects. Recently, new versions of Intel graphics drivers have been released much more frequently than AMD/NVIDIA ones. Perhaps this frequency is due to the fact that developers have something to fix, but it is good to know that work in this direction is constantly underway and, according to feedback, is yielding concrete results.

Conclusion

Looking at the decline in share price and the news of certain technical difficulties, it seems that the company is already on the edge of the abyss and only a miracle will save it. But we want to reassure those who are really worried about the situation. Intel’s story will continue. It should be understood that this is actually an industry-forming company that now owns 75-80% of the market for x86 processors for desktops/laptops/servers.

It is a tech giant with very large production facilities, tangible assets, and 100,000+ employees. In such a situation, even in the most difficult hour, the US government will come to the rescue and will not allow it to fall. Government programs to support technology manufacturers are already in place, and Intel is actively using them to accelerate its transformation. The current pace of production modernization is really impressive, and the company has a whole list of interesting and diverse products on the way. Shall we give them a chance to improve?