Apple presented a new Macintosh, which could have been called in concept Mac Mini Pro, but in the end they named Mac Studio. At first glance, it’s a mini PC, but with the capabilities of a workstation. His particuliarity ? Apple has introduced a new processor of its own at the top of this computer’s range: the Apple M1 Ultra. How is?
We’ve been hearing for some time how the ARM ISA can be competitive with x86 PCs, but while at the rest of the manufacturers it’s a promise for the future, at Apple it’s already a reality and it’s is that its high-performance cores have the ability to trade hits with the most advanced architectures from Intel and AMD. Especially since they exploit a much larger decoder in the control unit, taking advantage of the fixed-size instructions of RISC-like instruction sets.
Although above all there is the fact that no one in the PostPC device market dares to go beyond certain sizes and consumptions for the products for which they are intended. Instead, those in Cupertino have full vertical integration in the form of an ecosystem based on their own operating system. While ARM itself, NVIDIA, Qualcomm and many others have the ability to develop something like what Apple does, they lack something compared to macOS that allows them to create a product for such processors.
The M1 Ultra is Apple’s most powerful processor
A few months ago, Tim Cook introduced us to the M1 Max, a SoC based on the same architecture as the M1 but with a large number of CPU cores, a much larger GPU and a greater number of memory controllers. A beast of 423mm2 of the area under the most expensive TSMC N5 node, and, therefore, something that no one would dare to do for a tablet or a mobile due to its large size. Among its benefits were a 10-core CPU with 2 for efficiency and 8 for performance, a 32-core GPU or Tile Rendering-like shader units, and a 512-bit LPDDR5 memory controller.
Well, the M1 Ultra doubles those specs. for what we are talking about 20 CPU cores with 4 efficiency and 16 performancesnail 64-core GPU for a total of 8192 ALU and a memory controller 1024-bit LPDDR5 with 128 GB of memory soldered to the board and 800 GB/s bandwidth. The problem is that in this case we are not talking about a SoC, but two of them that work in tandem and have been mounted on the same interposer, and it is that at these levels of complexity it It is not possible to connect such complex processors in any other way, to the huge number of cores they have.
What are your performances?
According to Apple’s marketing, the performance per watt is ten times better than an Intel Core i9-12900K PC, but keep in mind that this type of comparison is tricky, since Apple did not mention the conditions and subjects of comparison. He also claims that the combined power of the two GPUs is equivalent to or even better than an RTX 3090. Just taking into account the differences in architecture, it’s clear that this is a huge exaggeration of from Apple. As it lacks the integration of DirectX 12 Ultimate technologies, including Ray Tracing, and powerful tensor cores. However, its great advantage is that it consumes 200W less than the RTX 3090.
Although, to be honest, who is going to play with a Mac when there is a PC where there are games? We are rather dealing with a processor that is looking for a workstation, since it is the PC on which it is mounted, the Mac Studio, it is the replacement for the long-gone iMac Pro, yes, much more expensive this times because it needs an external display.
Is this an Apple innovation?
No, rather not, because those of Cupertino do not manufacture processors and depend on TSMC technologies to intercommunicate their two SoCs. What is clear is that the point has been reached where making very large chips is no longer cost effective, moreover, within specs a theoretical monolithic Apple M1 Ultra would exceed the size limit of what TSMC could make . Add to that that as the size increases, the percentage of seek faults that lead to chip scrapping increases and we already have the explanation for the need to split using a dual SoC as a processor.
If you separate a CPU with few cores nothing happens, the bandwidths used are low and this is what AMD has done with its Ryzen for example. But when we talk about high-caliber GPUs or CPUs for servers with dozens of cores that move large amounts of data, new structures are needed, such as Intel Foveros O SoIC TSMC. It is precisely the technology of the Taiwanese that allows designs like this Apple M1 Ultra and all this due to the fact that they reduce the enormous consumption resulting from the transfer of data over long distances.
Apple’s packaging architecture connects two M1 Max dies. pic.twitter.com/s1JRaSE8tq
—Andreas Schilling
(hunt) March 8, 2022
The key? Vertically connect each of the two M1 Max with the interposer they are connected to, then use a silicon bridge for mutual communication. Which allows shorter interconnects, and to be able to place them in large numbers, in total more than 10,000. Where each of them has a low clock speed and voltage. This allows 2.5 TB/s communication between the two SoCs that make up the Apple M1 Ultra without triggering consumption. These types of structures will be common for communication between different chiplets in future industrial designs.
The post Apple M1 Ultra Debunked: Why is it a Dual APU? appeared first on HardZone.
