|
-They are built by Apple for specific products, so the M1 for the MacBook Air and 13” MacBook Pro, the M1 Pro/Max for the MacBook Pros, the M1 Ultra for the Mac Studio;
-They use ARM64 architecture, and because every instruction is 4 bytes the computer doesn’t need to analyse it, while with X86/X64 it can be from 1 to 15 bytes so every chip needs to analyse it before running it. It also produces less heat.
-They are energy efficient, the high-efficiency cores help this process, and the Apple Silicons are a SoC, so no waste of energy to move data; for example, if an action takes 35 watts on an Intel, it’d take 10 watts on the M1.
-It’s a System on a chip, so the chip makes all the processing. It has CPU, GPU, an image processing unit (ISP), a digital signal processor, the neural engine, video encoders and decoders, a Secure Enclave, and the unified memory.
-Apple is adding more GPU cores (which are specialised in doing well specific tasks, faster and with less energy, which increases performance per watt) other than CPUs (which are for general tasks).
-It has unified memory, so all the components use the same memory, and it is why they are fast. Commonly, computers have a CPU memory and GPU one, when the GPU needs to do a task in the CPU’s memory, the data can’t be accessed by the GPU, and the CPU needs to copy this data to the GPU’s memory (this process is very slow). This because the CPU prefers to ingest little data faster, and the GPU likes to ingest more data slowly. Apple includes low latency and high throughput, perfect for both.
-The CPU is very fast, in order to make it faster, you can make it handle operation quickly, or make it handle more operations (parallel processing), and the M1 CPU, called firestorm, has both of them. It also uses out-of-order execution, so developers don’t need to write threaded code (which is very difficult). This type of execution analyses what data rely on what data (because the code is written is a sequence), so the CPU knows which operations can be done simultaneously. Out-of-order execution basically depends on how fast and how many micro operations can run in a buffer (basically how quickly the computer chops up the code into operations, using decoders). Intel and AMD can use a maximum of 4 decoders, while the basic M1 has 8, and the instruction buffer is 3x larger than a normal one
-Even if they use a different technology, they can run X86 apps with Rosetta, and most of the time they run faster than previous models
|