Friday, December 4, 2020

Deep Dive into why Apple M1 beat Intel

The key reasons Apple M1 has beaten the older Intel Mac by a large margin are->

1) Apple M1 is built on a 5nm process node (4 Performance Cores + 4 Low Power Cores) vs old Intel Macs (14nm. 4 Cores). The new 10nm Intel Tiger Lake comes much closer to the M1 than the old 14nm chips but still lags behind Apple. A smaller node allows them to fit more transistors, while also reducing heat evolved. Using 4+4 cores(M1) vs 4 cores (Intel) also provides a big advantage.

2) Since the M1 is actually an SoC (System on a Chip), they have included dedicated cores (highly power efficient) for image processing, neural processing and for video encoding/decoding, and have hardware acceleration. That's why M1 crushes Intel chips while exporting videos from Priemer Pro. (Intel does have a video encoder/decoder too, however it still gets beaten and everything else runs on the general CPU).

3) Unified memory. Both the CPU and GPU access the same pool of memory (there is no copying of data. No separate memory for the CPU or GPU). And this helps them to reduce the wattage of the GPU making it very power efficient (and less heat generated).

4) Number of decoders. Although M1 is a RISC chip and Intel is a CISC chip, both have decoders which break the RISC/CISC instruction into smaller micro-ops. The M1 has a staggering 8 decoders while Intel chips are stuck with a max of 4 to maintain backward compatibility.

Therefore, the Firestorm cores (3.2GHz with 8 decoders) in M1 are able to equal the AMD Zen 3 cores (5GHz with 4 decoders) performance wise. Running at a lower clock speed means higher perf/watt, lower power consumption and less heat generated.

(x64 (64-bit) instruction set is more cleaned up than the x86 (32-bit) instruction set. Intel can include 8 decoders but then the system won't be able to run 32-bit applications properly).

5) Apple has included special hardware to facilitate x86 Intel like memory ordering (which is usually absent in ARM), which has resulted in superior emulation of x86 apps through Rosetta 2. (whereas Microsoft's emulation of x86 apps on Windows ARM is just software).

Note: RISC or CISC chips are not natively superior to each other in high-performance scenarios like laptops/desktops. It all comes down to the chip microarchitecture and a variety of other factors. Since both use Micro-ops, CISC vs RISC hardly makes a difference.

Is it game over for Intel or AMD? It's too early to say so. Credit to Apple for creating a marvellous chip.

M1 performs so well because of Vertical Integration.

All in all, More Competition is better for Consumers and Apple's M1 chip will push the industry forward.

Maybe points have been oversimplified to ensure everybody can understand.



Submitted December 04, 2020 at 07:06PM by Tombstone2K https://ift.tt/2VzW5xd via TikTokTikk

No comments:

Post a Comment