Network Security Internet Technology Development Database Servers Mobile Phone Android Software Apple Software Computer Software News IT Information

In addition to Weibo, there is also WeChat

Please pay attention

WeChat public account

Shulou

In just a week, Moore's Law died and came back to life.

2025-03-26 Update From: SLTechnology News&Howtos shulou NAV: SLTechnology News&Howtos > IT Information >

Share

Shulou(Shulou.com)11/24 Report--

In the past week, San Jose seems to have become a chip version of the instant universe. Moore's Law is dead in one universe, but still "alive and well" in the other.

The masters of these two universes are Nvidia and Intel respectively.

On September 28, Intel held an offline Innovation conference at the McNelli Convention and Exhibition Center, which is across from Nvidia's headquarters in San Jose airport. In the keynote speech, Intel CEO Pat Kissinger (Pat Gelsinger) released the 13th generation of Core processor Raptor Lake. Among them, i9-13900K products are 24 cores (8 performance cores, 16 energy efficiency cores) and 32 threads. They use a hybrid architecture of Intel 7 process process and x86 high performance. And, despite previous warnings of higher prices, the final price of the new generation is basically the same as that of the previous generation.

But the discussion of the day's speech was not limited to the new CPU itself. Kissinger first mentioned Moore's Law in his speech before releasing what Kissinger called "the fastest and best desktop processor in history."

"for the past few decades I have been in the midst of an argument that Moore's Law is dead." He said. "but the answer is, no!"

Kissinger opened his hands to the audience with a slightly exaggerated hand, and the PPT behind him showed a few lines of big words:

Moore's Law: Alive and Well (Moore's Law works well).

Image source: Intel "We expect the number of transistors integrated on a single chip package to reach 1 trillion by the end of the decade from 100 billion today." He said.

"We will not stop until the periodic table is exhausted, we will continue to be guardians of Moore's Law."

The speech was quoted by many media and was seen as a response to a speech made by Nvidia CEO Huang Renxun after the GTC conference a week ago.

At the previous 2022 GTC online conference on September 21, Nvidia CEO Huang Renxun released the RTX 40 series graphics cards and announced the new GPU architecture, Ada Lovelace and other product technologies and services. When asked about the controversial pricing issue in a media interview after the meeting, Huang Renxun attributed it to the death of Moore's Law.

"Today, the price of a 12-inch wafer is not a little more expensive, but a lot more expensive. Moore's Law is dead, it's over, it's over. The story that chip prices will fall over time is a thing of the past. Computing is no longer a chip problem, it's a common problem of software and chips." Huang Renxun said.

This is not the first time Huang Renxun has tried to put an end to Moore's Law. Huang Renxun believes that the alternative to Moore's Law is "accelerated computing". That is, rely on AI, service AI, through better AI technology and more customized design for specific scenarios to solve the problem of performance improvement.

Figure source: Invid Moore's law was proposed by Intel founder Gordon Moore in the 1960s. The basic idea is that the number of transistors that can be held on an integrated circuit doubles about every two years. However, with the continuous breakthroughs in the manufacturing process, related technologies begin to touch the physical limits, and whether Moore's Law is equally applicable to the future is also questioned by the chip industry.

People are looking for ways to "renew their lives", in which the views of Nvidia and Intel are distinct and opposing, and this different attitude is already familiar to the industry. Nvidia's chips are made by manufacturers, and it doesn't have to worry too much about manufacturing bottlenecks, while Intel insists on investing huge amounts of money on the manufacturing side, in addition to designing its own chips.

These different models determine the differences between the two, but this long-running struggle is not just a simple technical discussion between Huang Renxun and Kissinger, it is more important to contend for the right to define the development model of the next stage of the chip industry.

Moore's Law has never been a law of nature, not a law of physics, and even, in Moore's own words, when he first contributed to Electronics Magazine, he thought the magazine was a journal that no one read, and the law was just a trend he speculated casually based on his own experience.

Moore's Law really became the "law" of the chip industry because it defined the relationship between business success and technological progress in the industry-- it is not possible to take a huge technological lead and make money once and for all in this industry, because your products will be discounted every few years; only by constantly maintaining high-frequency technological breakthroughs and innovation can you survive. And such technological innovation is possible, who achieve the extreme, who will not only survive, but also get the most attractive return of human business society.

Therefore, the attempt to fundamentally negate Moore's Law is actually an attempt to change the rules of competition and cooperation in the whole industry.

Compared with constantly breaking through the process to transform the lowest chip process, the route of accelerating computing is obviously not so universal. Huang Renxun said that accelerated computing depends on AI, but today we do not have an omnipotent AI, only AI algorithms suitable for different scenarios, so the performance of accelerated computing is more specific. In Huang Renxun's words, "it looks like Nvidia's architecture accelerates everything, but in fact we just happen to accelerate 3000 things with the same architecture, because it took us 25 years to do it one by one."

This means that to get these performance improvements, you need to be able to support the corresponding AI algorithm model, and today, Nvidia's CUDA can serve more algorithms than anyone else, and it is also extremely closed and trying to become more dominant: "what we need to continue to do is to shove CUDA into every corner of the world in every possible way so that people can use it to develop applications."

According to Huang Renxun, Nvidia is the only full-stack accelerated computing company in the world: "an accelerated computing company must be full-stack, because accelerated computing requires you to understand the application, and then reprocess all the stacks to make it run faster."

To put it bluntly, Huang Renxun hopes that everyone will compete in its rules. As Intel has done over the past 50 years, he wants to change from Moore's law to Huang's law.

Source: it comes from the Internet, and Intel obviously doesn't want things to turn around.

Like Huang Renxun, who called himself a "full stack," Kissinger is trying to reshape Intel as the only orthodox Moore disciple, saying in a previous interview that Intel is the only company with all-round ability to keep Moore's Law alive.

In his speech, he also re-emphasized the new technology that made Intel so strong, and revealed some of the progress.

"technological breakthroughs in transistors-RibbonFET, breakthroughs in energy transfer technology-PowerVia, breakthroughs in lithography, the core technology of semiconductor manufacturing-High-NA lithography, and advanced packaging technologies. All these make us expect that a trillion transistors can be encapsulated on a single chip by 2030."

"We have a bold plan to deliver five process nodes in four years, which usually takes two years to break through one node, but we have to make five breakthroughs in four years." He said. "everything is going according to plan, even ahead of schedule."

These five nodes correspond to Intel 7, Intel 4, Intel 3, Intel 20A and Intel 18A. Intel 4, which uses EUV process for the first time next year, will enter the stage of mass production. Kissinger also revealed that the PDK version 0.3 of Intel 18A (equivalent to TSMC 1.8nm) process is already in the hands of developers.

"We expect the first batch of test chips to be completed by the end of the year."

In addition, Intel also introduced the technical progress in bypassing the physical limitations of electrons and continuing Moore's Law from an optical point of view, and briefly introduced a breakthrough in a solution called pluggable optoelectronic co-packaging (pluggable co-package photonics).

Optical interconnection can theoretically solve many problems that get stuck in Moore's Law, but the cost of manufacturing difficulty makes it impossible for industry to use it. Kissinger invited lab researchers to demonstrate "a sturdy, high yield (high-yielding), glass solution". It simplifies the manufacturing process through a pluggable connector and is expected to reduce costs.

Maintaining the position of Moore's Law in the industry is the most important task of his return. In recent years, there have been many divergent interpretations of Moore's Law, from wafers to the cooperation model of the whole industry, which have been framed into the framework of Moore's Law. But Kissinger is clearly a Moore's Law "fundamentalist". In a previous interview, he made it clear that what he said today was no different from Moore's Law:

"I define it by doubling the number of transistors."

Even for these new technologies that go beyond chip design itself, he stressed that these routes were foreseen by Moore that year.

"Moore foresaw this day of reckoning in his original paper, when we need to use a variety of heterogeneous and customized small technologies to build larger systems."

Extreme loyalty to the pioneer Moore should also be reflected in the response to "heresy". Kissinger, who was still squatting backstage before the press conference speech, seemed more aggressive than ever.

Touyuan: before Twitter took out the Core 13 series of chips at this launch, Kissinger first released the GPU. "not one, but a lot of GPU." He said. "graphical computing has always been one of my passions. When I left Intel 12 years ago, it was the only goal I didn't achieve, and now I'm back and we're going to finish it."

The GPU includes the previously released blade server of Intel data center GPU, code-named Ponte Vecchio, which Kissinger said has been shipped to Argonne National Laboratory to serve the auroral supercomputer. What attracts more attention is the Arc A770, the highest-end independent graphics card of the Arc series for the high-performance game market.

According to Intel's thinking, this GPU for gamers is aimed at the high pricing problem of GPU that Huang Renxun hopes to explain with Moore's Law is dead.

"GPU games have become super expensive in the past few years, but you, as gamers, should be troubled because the game community has suffered losses. today we are going to solve this problem."

As can be seen in Intel's presentation, this graphics card is priced at $329 for Standard Invidia's previous generation of RTX 3060 graphics cards.

In addition, changes in the data center market are also an important indicator of the struggle between life and death of Moore's Law.

Nvidia's data center business surpassed the gaming business for the first time in 2021, when revenue was $1.752 billion, but the share of data center revenue in the latest first quarter of fiscal 2023 has risen to 45.2%, with revenue of $3.75 billion, up 83% from a year earlier. Intel reported a 16% year-on-year decline in data center and artificial intelligence (DCAI) business in the second quarter of fiscal 2022, with revenue of $4.6 billion.

In the face of such catch-up, Intel also updated the progress of data center products and technologies at the conference, with a series of accelerators built into the fourth generation Intel Xeon scalable processors, which can open additional accelerator combinations for artificial intelligence, data analysis, networking, storage and other high-demand workloads beyond the original SKU basic configuration through on-demand activation mode. At the same time, Agilex FPGA's product line has also updated its roadmap, improving from logical structure performance to power efficiency. These updates are in response to Huang Renxun's "accelerated computing" route, which will be used to compete for the same group of users.

A more and more obvious trend is that the competition among chip manufacturers has become crisscross, extending to their respective hinterlands, and is becoming a new all-out war. In a fast-rolling battle, no one can be careless-even in the field of their own superior technology. At least on this level, Moore's Law still hangs over everyone.

This article comes from the official account of Wechat: ID:pinwancool, by Xuan Ning

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.

Views: 0

*The comments in the above article only represent the author's personal views and do not represent the views and positions of this website. If you have more insights, please feel free to contribute and share.

Share To

IT Information

Wechat

© 2024 shulou.com SLNews company. All rights reserved.

12
Report