Continuous mergers and acquisitions and frequent new product launches make the semiconductor chip industry in an unprecedented period of upheaval. On September 14, Nvidia announced the acquisition of Arm from SoftBank at a price of $40 billion to strengthen its weakness in CPU and formally enter the cloud data center by CPU+GPU. In early October, Intel sold its NAND storage business to SK Hynix for $9 billion to focus on data centers and PCS. AMD’s recent acquisition of Xilinx is also an attempt to capitalize on growth and put more resources into future data centers with more space. These deals have some integration and some cutting, but overall, the big three in the chip industry — Intel, Nvidia and AMD — have a very clear goal: their competitiveness in cloud data centers.
AMD, which has the lowest market cap of the big three, has been leading the pack in recent announcements. Nvidia’s 3000 series of graphics cards were released last month, but THE day before, AMD’s RX6000 series of graphics cards were no less impressive. We know that the biggest factor restricting the performance of the graphics card is bandwidth. RX6000 series is supported by rnDA2 unlimited cache technology. Compared with RDNA, bandwidth increases by 50%, frequency increases by 30%, that is to say, the silent frequency of RDNA2 can easily reach above 2G. This for performance increases will be enormous, coincidentally Microsoft officer yesterday declared their XBox game of X | S is the only carry rnda2 host platform, the author think that statement is an endorsement of Microsoft for AMD graphics performance.
Two weeks ago, AMD announced four Ryzen 5000 series cpus, which also attracted attention and broke Intel’s CPU myth. The Ryzen 5000 series processors use the new Zen 3 architecture and use a 7nm process. The Zen 3 delivers 2.4 times more performance per watt than the original Zen, and 2.8 times more energy efficiency than Intel’s 14NM MCPU 10gen I9. AMD also uses the new CCX technology, which allows each CORE to directly share access to 32MB L3 cache, a 19% improvement over the previous generation IPC technology.
Then, in fact, behind the seemingly complicated half-body market, a profound reflection of the game strategy between the chip three giants, the following author will take you to interpret. Advantage Strategy 1: We know that the IT industry produces a new ecosystem every ten years, such as Wintel’S PC ecosystem 20 years ago, and then the mobile terminal ecosystem formed by Android, IOS+ARM ten years ago. In the next one or two years, a new ecosystem is likely to be born, but no matter who the chip in this new dominant ecosystem will be won. Linux will most likely be the kernel of the new ecosystem.
Readers of the open source community may have noticed that in early October, Raymond, author of Church and Bazaar, announced that Microsoft was officially moving Windows to the Linux kernel. Although this was later denied by Microsoft, Microsoft’s contribution to Linux and the open source community is obvious.
At present, each Windows host can use WSL technology to start a Linux VM, and the Hongmeng operating system recently released by Huawei also includes the Linux kernel and Lite OS kernel in the system support list.
Most importantly, the Linux kernel operating system accounts for more than 95% of the cloud server market, which is a sign that the IT giants are getting ahead of Linux. Nvidia’s acquisition of ARM is a means to advance the layout of Linux. We know that Huawei has proved to the industry that ARM+Linux is a perfect match in cloud computing through kunpeng series of cloud servers, which can achieve a perfect balance between performance and energy consumption. And this aspect is especially bad for Intel, Linux founder Comrade Linus, for their attitude almost daily rage, interested students can baidu. The combination of Linux and ARM threatens the traditional X86 territory, but it also makes us see two interesting things. Intel is now contributing code to Linux like crazy. If we count the number of lines of code in version 5.8, Intel’s contribution to Linux version 5.8 has jumped to the top. Amd’s contribution to Linux has also entered the top 10. I’m sure Linus won’t be able to do any more after all that.
Another point of view is that AMD, the junior X86 player, has finally begun to blow up the market. Both ZEN3 and RX6000 have exceeded market expectations. The reason for this strategy is very simple: in the face of a possible marriage between ARM and Nvidia, if we don’t make a big move at this time, there may be no future opportunity. Competition is the best way to drive technological progress and the direct driving force of the open source community, and Linux will continue to be the hot property of IT giants in the future.
Advantage strategy 2: Layout around TSMC
According to the third quarter of this year’s financial, TSMC the chip manufacturing giant this year has reached a shocking 53% gross margin, which means they chip manufacturing costs less than half of the ex-factory price, this is among the world top 500 companies unique, even called IT far luxury brand of apple.
Building and maintaining fabs is too expensive and requires a steady stream of technological upgrades that only TSMC and Samsung can afford. Due to the division of explicit, TSMC has been focused on chip foundry from scratch, not only the process is very fast, and also on the yield is very satisfactory, and TSMC never involved chip design for years, for AMD, nvidia such customer chip IP protection work is very good, is also due to focus on chips, TSMC has the energy and financial resources to do a good job and expand its business, which is also in line with the future division of labor trend in the chip industry.
Due to financial reasons, AMD sold its fabs at the end of 2008 and focused on the chip design field, which is also the most upstream in the chip industry. AMD is mainly responsible for building a team to develop various chips such as CPUS. After the chip design is completed, AMD’s main wafer manufacturer is TSMC. The alliance allows AMD to beat Intel to the 7nm process. Although I do not subscribe to process determinism. But the 10% drop in Intel’s stock price two weeks ago was also a response to their long-term inability to break through the 14nm process.
With for the understanding of advanced process and advanced chip technology level, in recent years the nvidia will also belong to samsung’s orders before many turned to TSMC, but in this respect to the design, manufacture, packaging, testing one-stop Intel is obviously a little couldn’t keep up with the pace of The Times, for each link in the field of chip is so complicated, As long as there is a link to appear problem, so process upgrade will be a mirage, this is Intel to NAND business, it is in focus in the field of chip behind, and I believe that if a year didn’t make a breakthrough, Intel will probably give up their wafer manufacturing, defected to TSMC in the league.
Dominated strategy: Toothpaste type slow technology upgrading X86 architecture CPU relying on solid in the solid before the Wintel alliance, occupies a absolute market share, the desktop on X86 instruction set architecture chips chips, Intel master patent, less than seven but AMD control 3 into a number of patents, cross licensing two manufacturers, but there are few foreign authorization, Basic has become so X86 Intel and AMD sang together the errenzhuan, so although Intel r&d not low, but most of the results are used to reserve, just slow to promote their products, the evolvement of the technology stack this strategy not only can ensure its own profits, also can make the X86 architecture of the younger brother AMD not too far behind, So that they face monopoly lawsuits.
However, such a model has a fatal weakness. In the current explosive development of technology, the technological reserves of a few years ago may be worthless today. Although Intel has fully utilized the maximum capacity of 14nm process, the trend of diminishing marginal benefit of manufacturing will not appear until below 5nm. Even if Intel’s insistence on 14nm cpus remains ahead in terms of performance, they are no match for 7nm chips in terms of power consumption. And the cloud data center field attaches great importance to energy consumption, and we also mentioned above that the cloud data center field is the future of CPU chips, and in this respect, Intel is surrounded by Arm before, and AMD is catching up, so the future situation may not be good.
We should be well aware of the development trend of the chip industry and the competition pattern of the giants. At the same time, we must clearly see that the competition of the three giants will only accelerate the development of the semiconductor industry. So we have to get down into the game, not just watch the fun, slow down our pace of catching up,