Apple's chip designs for the M5 Pro and M5 Max have adjacent dies, using the same UltraFusion architecture that enabled the creation of Apple's Ultra chips. Here's how and why that's faster than before.

The M5 Pro and M5 Max are very powerful chips, but getting that high performance required the engineers to come up with new ways to design the chips in the first place. In the latest releases, Apple actually borrowed an idea from its Ultra line for its M5 chip packing.

Speaking to Heise Online in an article spotted by WCCFTech, Anand Shimpi of Apple's Hardware Technologies department explained that the UltraFusion architecture of the M2 Ultra and M3 Ultra helped create the M5 Pro and Max.

UltraFusion refers to the way that Apple used an interconnecting section to combine two Max chips into one larger Ultra chip. The technique allowed the two dies to communicate with each other at high speed, effectively working as one larger chip with double the core counts.

In the M5 Pro and M5 Max, the same technique is used. In effect, multiple dies, or parts of a chip, are placed very close to each other, reducing the physical footprint and enabling extremely fast communications between the elements.

Old idea, new application

In the interview, Shimpi confirms that Apple is effectively doing the UltraFusion chip-combining technique again, but in this new layout. The key difference here is that Apple's not simply marrying together two identical chip dies, but instead is marrying smaller dies to form the complete chip.

"We've actually split a number of functions across two different dies," Shimpi explains. "They aren't mirror images of each other."

The building blocks of the chip are stacked and connected together, certainly. However they are not identical nor complete chips in their own right.

With Apple's approach, it has a lot more control over what elements are adjacent Since they are not identical chiplets, you don't get the effect of the same heat-producing sections lining up.

This has already resulted in some benefits to the M5 series of chips. Under stress tests, the M5 Max was reportedly found to operate at a lower temperature than the M4 Max under load.

For Apple, the benefits go further than speed and performance. It also has the opportunity to combine different chiplets together, and to switch some out for others.

For example, while it may keep the CPU-related chiplet sections the same across both M5 Max versions, it can add a different GPU chiplet depending on which core count chip it's making.

This gives Apple the ability to create variants of chips, but doing so in a less expensive and wasteful way than by creating separate single-wafter designs.

Updated March 23, 1PM ET: Heise misinterpreted some of what Apple said. The dies are adjacent, and not stacked.