Roger’s $0.02 - Apple’s M3: Not So Scary

Apple’s “Scary” event the day before Halloween was a chance to reveal the company’s latest iteration of the M-series SoC, the M3, M3 Pro, and M3 Max. Apple claims the new processor’s performance cores are 30 percent faster the efficiency cores are 50 percent and the Neural Engine 60% faster than their M1 equivalents. The improvements don’t stop there. The GPU has also been upgraded to offer up to 80% faster performance with increased power efficiency. What was most telling, though, was what wasn’t said.

Consider Apple’s press release for the M3. In it, the previous generation M2 was mentioned once, about the additional 5 billion transistors the M3 has over the M2. Other than that point, there was no direct performance comparison. In comparison, the M1 is referenced at least 20 times in the same press release.

While there’s no doubt that the M3 steamrolls the M1 in performance, when it comes to the M2 the answer isn't so clear-cut. Anandtech observes that while Apple positions the M3 as around 15% improvement over the M2 variants, benchmark tests, and settings weren’t disclosed to arrive at that conclusion. Adding to that many of the application benchmarks listed by Apple include mixed-together tests that also rely on GPU performance thus making any CPU-only gains less clear.

Also not mentioned was the slight downgrade shift of the M3 Pro from the previous M2 Pro SoC. The M2 Pro comes with 8 performance cores and 4 efficiency cores. The M3 Pro in contrast has 6 performance cores and 6 efficiency cores. In effect, the M3 Pro is losing two performance cores. Now this could be chalked up to several different factors. One TSMC’s new N3B (3nm) process might have some yield issues with outputting an 8 power/4 efficiency core layout. Two it’s cheaper to produce in this fashion. Because the M3 line is being made with TSMC’s new 3nm process, theoretically you get more chips per wafer. If there’s a price point that Apple is trying to meet per SoC it makes sense to tweak the design to meet it. Three Apple is trying to segment out the market for the products that use the chips and doesn’t want to cannibalize sales of the higher priced M3 Max or M2 Max offerings. Finally, it could be a combination of the three. The fact is that relative to the M2 Pro the performance benefit of moving to a M3 Pro is pretty modest.

Of the three the M3 Max is a cut above its predecessor. It adds 4 more performance cores, 2 more GPU cores, and a speedier neural engine combined with a 128GB memory capacity, 32GB more than the M2 Max. This adds 37% more transistors to the design than its predecessor resulting in a huge chip and consequently a huge price leap from the M3 Pro, around $800.

The most notable update that the M3 brings is the chip’s GPU. A new architecture that includes more cores, mesh shaders, and ray tracing means that Apple is trying to compete with PC GPU vendors on feature sets. But it doesn’t stop there. The M3 also introduces “Dynamic Caching” a new hardware-level memory management scheme aiming to make GPU memory allocation more efficient by using “… only the exact amount of memory needed…” Supposedly the system is transparent to developers, meaning they can’t control it, but because the M3 series relies on a unified memory architecture that isn’t user upgradeable this was something that was already in the works. Unfortunately like the general benchmark results touted by Apple granular details were absent.

From what I can see the M3 does offer a significant performance leap from those still using a M1 series machine, but for those using M2 series devices the benefit drops. The most compelling aspect of the M3 series is the GPU. By offering new hardware features in the form of mesh shaders and ray-tracing it brings Apple’s Mac SoC in line with its A17-based iPhone and iPad offerings. Coupled with the M3’s higher power limit it could mean the GPU has room for even more performance.

The conclusion from me. The M3 is another advancement for Apple’s in-house Mac silicon. However, the generational gains are smaller than previous iterations, which is to be expected. There’s a point where an architecture has matured to the point where any performance increases will be incremental. Unless you’re finding your M1 Mac to be a bit pokey, need the new GPU features, or need the horsepower of an M3 Max you’re better off saving your money until you have a pressing need for an M3 upgrade.