Apple goes to 2.5D for 114bn transistor M1 Ultra

Apple goes to 2.5D for 114bn transistor M1 Ultra

Technology News |
By Nick Flaherty

Apple is adopting a 2.5D approach to boost the performance of its latest M1 Ultra device that opens the door to future designs using chiplets.

A packaging architecture called UltraFusion that interconnects the die of two M1 Max chips on a siliocn interposer to create a system on a chip (SoC) with 114bn transistors.

This uses a silicon substrate and interposer that supports the two die with 10,000 interconnects with 2.5TB/s of low latency, inter-processor bandwidth between the die. This also links the die to 128GBytes of  low-latency unified memory using a 800GB/s interface.

This is comparable in size and performance to Intel’s chiplet-based Ponte Vecchio processor (see below).

The two die in the M1 Ultra are built in TSMC’s 5nm process and provide 16 high performance ‘Firestorm’ ARM cores and four high efficiency ‘Icestorm’ cores. However this means there are four clusters, each with four Firestorm cores clocked at 3228MHz and one Icestorm core clocked at 2064MHz. Each cluster shares 12MB of L2 cache, while the two high-efficiency cores in each die share 4MB of L2 cache.

The two die also bring 64 GPUs and two 16 core Neural Engines. The key is that by using the existing M1 Max chips the current software runs without modification.

The 32-core Neural Engine runs up to 22 trillion operations per second (TOPS) and the two media engines supports video encode and decode throughput using the ProRes format. These enable play back of up to 18 streams of 8K ProRes 422 video.

Related Apple M1 articles

The M1 Ultra also integrates custom Apple technologies, such as a display engine capable of driving multiple external displays, integrated Thunderbolt 4 controllers and Apple’s Secure Enclave, hardware-verified secure boot and runtime anti-exploitation technologies.

“M1 Ultra is another game-changer for Apple silicon that once again will shock the PC industry. By connecting two M1 Max die with our UltraFusion packaging architecture, we’re able to scale Apple silicon to unprecedented new heights,” said Johny Srouji, senior vice president of Hardware Technologies at Apple. “With its powerful CPU, massive GPU, incredible Neural Engine, ProRes hardware acceleration, and huge amount of unified memory, M1 Ultra completes the M1 family as the world’s most powerful and capable chip for a personal computer.”

Apple has been benchmarking the Ultra against 16 core x86 processors, rather than its M1 Max chip. It cites 90 percent higher multi-threaded performance than the fastest available 16-core PC desktop chip for the same power consumption, or 100W less for the same peak performance.

However the latest Ice Lake x86 desktop processors have now adopted a similar hybrid core approach with high performance and high efficiency cores, which will be a more appropriate comparison. Intel has also adopted a chiplet architecture for its Ponte Vecchio GPU, which is a comparable size of over 100bn transistors in TSMC’s 5nm process.

Related Intel and Ice Lake articles

The latest MacOS, Monterey, has been optimised for the increases in CPU, GPU, and memory bandwidth. Developer technologies like Metal let apps take full advantage of the new chip, and optimizations in Core ML utilise the full 32-core Neural Engine so machine learning models run faster than ever. 

Users have access to the largest collection of apps ever for Mac, including iPhone and iPad apps that can now run on Mac, and Universal apps that are optimised for the M1 family of chips. Apps that have not yet been updated to Universal will run seamlessly with Apple’s Rosetta 2 interpreter.

Other articles on eeNews Europe


If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles