The business has gotten a remarkable amount of performance out of silicon by merging two CPUs into one.
IN PRACTICE, THE M1 Ultra functions as a single, absurdly enormous slice of silicon that accomplishes everything. Apple’s most powerful CPU to date contains 114 billion transistors packed into over a hundred logic, graphics, and artificial intelligence processing cores, all of which are coupled to 128 gigabytes of shared memory.
The M1 Ultra, on the other hand, is a Frankenstein’s monster, made up of two identical M1 Max processors joined by a silicon interface that acts as a bridge. This innovative design gives the impression that the conjoined chips are one larger whole.
Chipmakers are stitching components together to enhance computing power as it becomes more difficult to decrease transistors and impracticable to make individual circuits much larger.
The computer industry’s goal is to progress in a Lego-like manner. And Apple’s M1 Ultra demonstrates how innovative methodologies can result in significant performance improvements.
“This technology arrived just in time,” says Tim Millet, Apple’s vice president of hardware technologies. “In a sense, it’s about Moore’s law,” he says, referring to the decades-old dictum that chip performance—measured by the number of transistors on a chip—doubles every 18 months, as named after Intel pioneer Gordon Moore.
Moore’s law, which has fueled advancement in the computer industry and the economy for decades, is no longer valid. Engineers are approaching the physical limits of how small these components, which have features measured in billionths of a meter, can be.
Some extremely complex and expensive engineering tricks promise to help shrink the size of components etched into silicon chips even further, but they are reaching the physical limits of how small these components, which have features measured in billionths of a meter, can be.
Even if Moore’s law is no longer valid, computer chips are more important—and pervasive—than they have ever been. Cutting-edge silicon is essential for technologies like AI and 5G, and supply chain delays caused by the epidemic have emphasized how important semiconductors have become to businesses like auto manufacturing.
A rising number of organizations are turning to create their chips for performance increases as each new generation of silicon takes a smaller stride ahead. Since 2010, Apple has employed proprietary silicon in its iPhones and iPads; however, in 2020, the company said that it will design its CPUs for Macs and MacBooks, moving away from Intel’s offerings.
Apple’s desktop chips, which employ the same architecture licensed by the British company ARM, are based on the work it accomplished on smartphone chips. Apple has complete control over a device by designing its silicon and merging functions that would ordinarily be performed by different processors into a single system-on-a-chip.
This allows Apple to change both software and hardware at the same time. That amount of control is the key
Millet, a semiconductor industry veteran who joined Apple from Brocade, a US networking startup, in 2005, says, “I saw the whole [chipmaking] world was upside down.”
Unlike, for instance, Intel, which designs and manufactures chips before selling them to computer manufacturers, Apple, according to Millet, may build a chip for a device at the same time as the software, hardware, and industrial design.
When Apple unveiled the M1 Max, the company’s previous desktop chip, in October of last year, a few keen observers spotted something strange: a large stretch of silicon along one side that appeared to do nothing.
The unknown silicon would turn out to be part of Apple’s UltraFusion high-speed interconnect technology, which combines two M1 Max chips into one M1 Ultra chip with a dense network of tiny connections.
When Apple began developing a new desktop computer for power users, the Mac Studio, the chip team recognized it wouldn’t be possible to rely solely on Moore’s law for a significant performance improvement.
However, TSMC, the Taiwanese chipmaker that manufactures Apple’s designs, had begun honing technology for linking two bits of silicon using a high-speed interconnect, an idea that has been available for years but has only been used to combine cores that do separate tasks.
Apple tweaked the TSMC technology to create Max’s mysterious interface, which is made up of two incredibly complicated semiconductors.
Millet says of the Mac Studio, “UltraFusion gave us the capabilities we needed to be able to fill up that box with as much compute as we could.”
The M1 Ultra has been benchmarked and found to be competitive with the fastest high-end computer CPUs and graphics processors available. Millet claims that once developers port over the relevant software libraries, some of the chip’s features, such as its ability to run AI applications, will become obvious.
The M1 Ultra is part of a larger trend toward increasingly modular processors in the semiconductor industry. Intel is working on a technology that allows “chiplets,” or little pieces of silicon, to be stacked on top of one another to build bespoke designs that don’t have to be redesigned from the ground up.
Pat Gelsinger, the company’s CEO, has highlighted “advanced packaging” as one of the company’s big turnaround plans. AMD, Intel’s main competitor, is already employing TSMC’s 3D stacking technique to make server and high-end PC processors.
Intel, AMD, Samsung, TSMC, and ARM formed a partnership this month to work on a new chipset design standard. The M1 Ultra takes a more extreme approach, employing the chipset concept to connect complete chips.
The goal of Apple’s new processor is to boost total processing power. “Depending on how Moore’s law is defined, this method allows you to design systems with many more transistors than what can fit on a single chip,” says Jess del Alamo, an MIT professor who studies novel chip components.
He goes on to say that the fact that TSMC, which is at the bleeding edge of chipmaking, is looking for new ways to keep performance growing is noteworthy.
“The chip industry believes that future growth will come not only from Moore’s law but also from developing systems that can be produced using many technologies that have yet to be combined,” he says.
“Others are attempting similar things, and we certainly see a trend toward more of these chipset designs,” says Linley Gwennap, author of the industry journal Microprocessor Report.
The rise of modular chipmaking may help future gadgets perform better, but it may also disrupt chip manufacturing economics. A chip with twice the transistors might cost twice as much if Moore’s law didn’t exist.
Todd Austin, an electrical engineer at the University of Michigan, adds, “With chiplets, I can still offer you the base chip for $300, the double chip for $600, and the uber-double chip for $1,200.” Instead of chips getting faster for the same price each year, chiplets could mean that extra performance comes at a cost. Austin adds that the strategy, which is still in its early stages, will add extra complexity to chip design, thereby increasing costs.
Thanks to a unique take on the chipset method, the M1 Ultra can match the performance of some of the market’s most powerful chips. It also positions Apple to gain a big advantage for Macs, just as it has for years with the iPhone.
Always be Updated with us visit GeeksULTD for Real-time Updates