The large image: Apple is pushing boundaries with its 114 billion transistor behemoth M1 Ultra. It makes use of state-of-the-art interconnection expertise to attach two disparate chips right into a single SoC. Fortunately, builders is not going to have to leap by hoops to make the most of the Ultra’s full potential as it can behave as a single unit on the system degree.
Apple not too long ago introduced the M1 Ultra, its new flagship SoC that may energy the all-new Mac Studio, a compact but high-performance desktop system. It claims the M1 Ultra powered Mac Studio would offer CPU efficiency as much as 3.8 occasions quicker than the 27-inch iMac with a 10-core processor.
“[The M1 Ultra is a] game-changer for Apple silicon that when once more will shock the PC trade,” mentioned Johny Srouji, Apple’s senior vice chairman of {hardware} applied sciences.
The M1 Ultra is undoubtedly a powerhouse. It combines two M1 Max chips over what Apple calls the UltraFusion interprocessor interconnection that provides 2.5 terabytes per second of low latency, inter-processor bandwidth.
As per Apple, UltraFusion makes use of a silicon interposer with twice the connection density & 4 occasions the bandwidth of competing interposer applied sciences. Since every M1 Max has a die space of 432 mm², the UltraFusion interposer itself needs to be over 864 mm². That is within the realm of AMD and Nvidia’s enterprise GPUs that includes HBM (High Bandwidth Memory).
Another benefit of Ultrafusion is that builders will not must rewrite their code, as on a system degree, the Mac will see the dual-chip SoC as a single processor.
Built on TSMC’s 5-nanometer course of, the M1 Ultra has 114 billion transistors, a 7x improve over the unique M1. It can assist as much as 128 GB of unified reminiscence with a reminiscence bandwidth of 800 GB/s, made attainable by its dual-chip design. It consists of 16 efficiency cores with 48MB L2 Cache and 4 effectivity cores with 4MB L2 cache, whereas the GPU can have as much as 64 GPU Cores. It additionally sports activities a 32 core Neural Engine that may execute as much as 22 trillion operations per second for accelerating machine-learning duties.
Digitimes experiences that Apple’s M1 Ultra SoCs use TSMC’s CoWoS-S (chip-on-wafer-on-substrate with silicon interposer) 2.5D interposer-based packaging course of. Nvidia, AMD, and Fujitsu have used related applied sciences to construct high-performance processors for datacenters and HPC (high-performance computing).
Taiwan chipmaker TSMC has a more recent various to CoWoS-S in its InFO_LSI (InFO with integration of an LSI) expertise for ultra-high bandwidth chiplet integration. It makes use of localized silicon interconnects as a substitute of enormous and costly interposers, just like Intel’s EMIB (embedded die interconnect bridge).
It is believed that Apple selected CoWoS-S over InFO_LSI because the latter was not prepared in time for the M1 Ultra. So Apple may need performed it secure by choosing a confirmed however costlier resolution over a less expensive, extra nascent expertise.
The Mac Studio might be out there beginning March 18, with a beginning worth of $3999, which incorporates 64GB of unified reminiscence and a 1TB SSD.
