If you’ve been following developments in the semiconductor sector, particularly in recent weeks, it would be easy to see how some people think Intel (INTC) is in serious trouble. After all, several signs are not promise.
First, although Intel announced strong revenue in the second quarter a few weeks back, it thanked the stock when the company still revealed a problem and delay for its 7nm manufacturing process. Second, related to that, Intel recently lost its crown as the largest and most valuable semiconductor company – a place it held for decades – to Nvidia (NVDA). In fact, at the close of trading on August 17, Nvidia’s hard charging passed a $ 300 billion market capitalization, against a $ 208 billion capitalization for Intel. Ultimately, it’s from a CPU performance perspective AMD Inc. (AMD) (which recently and briefly flooded the $ 100 billion capitalization mark) that Intel has handily slapped on the steering wheel on PC desktops, mobiles, and even server benchmarks. For a company that has traditionally dominated in all these areas, this has been quite a change.
Just when you start to think that Intel is gliding, the company may surprise you. And so it did, both at last week’s Architecture Day and at this week’s Hot Chips conference. The Architecture Day event in particular offered a comprehensive look at a number of innovations that enable Intel to work on prostate technology, chip packaging, core chip architecture, high-speed chiplet-to-chiplet interconnect, security and much more. Importantly, although some of these developments fall into the more theoretical realm, the vast majority will bring practical improvements from real performance that should translate into offerings that will be highly competitive with AMD’s latest.
Intel returned to its roots in semiconductor manufacturing to deliver what appear to be very promising improvements in basic transistor technology. Until just a year or two ago, Intel was able to use its long history of expertise in chipmaking to its clear advantage, bringing out regular improvements that other chipmakers would often take 18-24 months to capture. However, as process delays and other production problems began to pile up, it began to seem that the company’s own manufacturing became more of an albatross around its neck. With the debut of 10nm SuperFin technology, the company clearly reminds the rest of the sector that it has a number of different tricks up its sleeve when it comes to core semiconductor capabilities. More importantly, these advances can still be used to make significant advances in areas where companies that do not make their own chips cannot gain access until their production partners develop them as well.
By utilizing a new metal capacitor technology in combination with the company’s improved FinFET transistors, Intel claims that the 10 nm SuperFin technology makes it possible to drive higher currents, reduce resistance and thereby significantly improve performance over traditional transistors. designs. In practical terms, the first Intel part to use the new technology (a chip called Tiger Lake unveiled in early September) is expected to have a performance boost that the company claims is the largest “intranode” ( meaning on the same process large) improvement it has ever been. In fact, the company says it would be “comparable to a full-node transition”, which is a pretty big explanation – although it is also intended to counteract ongoing delays to the 7nm node. Real world performance remains to be seen, but initial comments and analysis from those who understand the details of chip architecture well are very promising.
At the same time, the company also recognizes that as transistors get smaller and smaller and process technologies get harder and harder, it makes sense to separate chip design, or “share” manufacturing enhancements, so that the two elements can advance separate timelines. Combining this with the fact that Intel recently openly discussed the idea of using outdoor plants for manufacturing chips, such as foundries, to build some of their parts, marks a very practical, probably even more modern, approach to the situation. It also reflects a very different attitude towards Intel – one that should help the company move forward on advanced chip designs, regardless of where (or by whom) they were eventually built.
Intel also provided a lot more information about its Xe GPU architecture, with a particular focus on how it will be able to scale from better integrated graphics – the first iteration will also be part of Tiger Lake – to several different types of discrete GPUs on different brands. The first standalone GPU, named DG1, is expected later this year, as well as a new server-focused GPU. Both of these chips, like the integrated GPU, will be based on what Intel calls the Xe-LP (low power) architecture. The company also unveiled a new Xe-HP (high-performance) GPU architecture designed for other discrete chips, aimed at more advanced data center applications, such as AI acceleration, as well as a gaming-optimized part that tracing will support.
Both at HotChips and on Architecture Day, the company also talked a lot about its software efforts, in particular the OneAPI architecture, which is designed to make the programming process for any type of non-CPU accelerator (from GPUs to FPGAs to other AI-focused accelerators and beyond) much easier to do. This is clearly a monumental goal and a challenging task. However, the company seems to be making solid progress and highlighted how software enhancements can only often lead to significant performance improvements.
The bottom line is that it is clearly too early to count out Intel. If nothing else, the significantly more competitive environment in which it has found itself seems to inspire it to do some of its best work in years. Final details have yet to come, but the good news is that the company is clearly aware of competitive challenges, and regardless of which suppliers you choose to support, we will all benefit in the end.
Disclaimer: Some of the author’s customers are vendors in the tech sector.
Announcement: None.
Editor’s note: The summary headers for this article were chosen by Seeking Alpha editors.
