For practical purposes, in M1 Ultra acts like one incredibly large piece of silicon that does it all. Apple is the most powerful chip today has 114 billion transistors packed into over a hundred processing cores dedicated to logic, graphics and artificial intelligence, all connected to 128 gigabytes of shared memory. But the M1 Ultra is actually a Frankenstein monster, consisting of two identical M1 Max chips bolted together with a silicon interface that acts as a bridge. This clever design gives the impression that the connected chips are actually one big entity.
As it becomes increasingly difficult to downsize transistors and it is impractical to make individual ICs much larger, chip makers are starting to stitch components together to increase computing power. LegoThis approach is a key way to develop the computer industry. And the Apple M1 Ultra shows that new technologies can lead to a significant jump in performance.
“This technology came at just the right time,” says Tim Millet, Apple’s vice president of hardware technology. “In a sense, it’s about Moore’s law,” he adds, referring to axiom ten years agonamed after Intel co-founder Gordon Moore, this chip’s performance, as measured by the number of transistors on a chip, doubles every 18 months.
It’s no secret that Moore’s Law, which has determined progress in the computer industry and the economy for decades, no longer applies. Some extremely complex and costly engineering tricks promise to help reduce size of components etched into silicon chips farther, but engineers are reaching the physical limits of how small these components can be, measured in billionths of a meter. Even if Moore’s law is obsolete, computer chips more important and ubiquitous than ever. Advanced silicon is critical to technologies like AI and 5G, and supply chain disruptions caused by the pandemic have highlighted how vital semiconductors are now for industries such as the automotive industry.
As each new generation of silicon takes a small step forward, more and more companies are turning to development of own chips to improve performance. Apple has been using custom silicon for its iPhones and iPads since 2010 and then in 2020 it announced that develop your own chips for Mac and macbooks, departure from Intel products. Apple used its work on smartphone chips to develop desktop PC chips using the same architecture, licensed from the British company ARM. By building its own silicon and integrating functions that would normally be performed by individual chips into one system-on-a-chip, Apple is in control of the entire product and it can customize software and hardware together. This level of control is key.
“I understood all [chipmaking] the world was turned upside down,” says Millais, a chip industry veteran who joined Apple from Brocade, a US networking company, in 2005. Unlike, say, Intel, which designs and manufactures chips that are then sold to computer manufacturers, Millais explains that Apple can work on chip design for a product at the same time as software, hardware, and industrial design.
When Apple announced M1 Maxits previous desktop chip, last October a few sharp-eyed observers noticed something strange: a long strip of silicon along one edge that seemed to do nothing at all. The mysterious silicon turned out to be part of a high-speed, dense array interconnect technology that Apple calls UltraFusion, which turns two M1 Max chips into one M1 Ultra.
When Apple started working on a new power user desktop, the product that would Studio Mac, the chip design team knew they couldn’t rely on Moore’s law alone for significant performance improvements. But TSMC, the Taiwanese chip maker behind Apple’s designs, has begun refining the technology for connecting two silicon cells with a high-speed interconnect. This idea has been around for many years, but previously it was mainly used to combine cores that perform different tasks. . Apple modified TSMC’s technique to create the cryptic interface seen on the Max by combining two very complex chips.
“UltraFusion gave us the tools we needed to fill that box with as much computing as possible,” Millet says of Mac Studio. Benchmarking M1 Ultra showed that it can compete with the fastest high performance computer chips and GPUs on the market. Millet says some of the chip’s capabilities, such as its potential to run AI applications, will become apparent over time as developers port the necessary software libraries.
The M1 Ultra is part of a wider industry shift towards more modular chips. Intel is developing a technology that allows various parts silicon, called “chipsets”, which must be stacked on top of each other to create custom designs that don’t have to be redone from scratch. CEO Pat Gelsinger, defined this “advanced packaging” as one of the pillars of a grand recovery plan. Intel competitor AMD is already using TSMC’s 3D stacking technology to build some processors for servers and high-end PCs. This month, Intel, AMD, Samsung, TSMC and ARM announced a consortium to work on new standard for chipset designs. In a more radical approach, the M1 Ultra uses the concept of chiplets to connect entire chips together.
Apple’s new chip is designed to increase overall computing power. “Depending on how you define Moore’s law, this approach allows you to create systems that use many more transistors than what fits on a single chip,” he says. Jesus del Alamo, professor at the Massachusetts Institute of Technology, researching new components of microcircuits. He adds that it’s important that TSMC, at the forefront of chip manufacturing, is looking for new ways to improve performance. “Obviously, chip manufacturers see that progress in the future will depend not only on Moore’s law, but also on creating systems that can be manufactured using various technologies that have yet to be combined,” he says.
“Others are doing the same and we are definitely seeing a trend towards more of these chipset designs,” adds Lynley Gwennapauthor microprocessor reportindustry newsletter.
The rise of modular chip manufacturing could help improve the performance of future devices, but could also change the economics of chip manufacturing. Without Moore’s Law, a chip with twice as many transistors could cost twice as much. “With chiplets, I can still sell you a basic chip for, say, $300, a dual chip for $600, and a super dual chip for $1,200,” he says. Todd Austin, electrical engineer at the University of Michigan. Basically, instead of chips getting faster every year for the same price, chiplets can mean that extra performance comes at a cost. Austin adds that the approach, which is still relatively new, will also make chip design more difficult, which can also add to the cost.
The M1 Ultra shows off the power of some of the most powerful chips on the market with a creative take on the chiplet approach. It also allows Apple to achieve a significant advantage for the Mac, just as it has done with the iPhone for years.
- 📩 Latest news about technology, science and more: Receive our newsletters!
- The largest Russian technology company too big to fail?
- That’s how global energy crisis ends
- We explain Casethe new standard for smart home
- Future of NFTs lie with the courts
- Chernobyl was a paradise of the wild. Then Russia invaded
- 👁️ Explore AI like never before with our new database
- 💻 Update your working game with our Gear team favorite laptops, keyboards, input alternativesand noise canceling headphones
Credit: www.wired.com /