How “single memory” speeds up Apple M1 ARM Macs


Apple M1 chip
Apple

Apple is rethinking how components should exist and work inside a laptop. With M1 chips in new Macs, Apple has a new “Unified Memory Architecture” (UMA) technology that dramatically speeds up memory performance. This is how memory works on Apple Silicon.

How Apple Silicon handles RAM

In case you haven’t heard the news yet, Apple announced a new Mac tablet in November 2020. The new MacBook Air, MacBook Pro and Mac Mini models use an ARM-based processor dubbed the Apple M1. This change has long been anticipated, and the culmination of Apple’s decade was the design of ARM-based processors for the iPhone and iPad.

The M1 is a chip-based system (SoC), which means that there is not only a CPU inside the processor, but also other key components, including the GPU, I / O controllers, Apple’s Neural Engine, the AI and, most importantly, physical RAM is part of the same package. To be clear, RAM is not on the same silicon as the core components of SoC. Instead, it sits sideways as shown in the image above.

Adding RAM to the SoC is nothing new. Smartphone SoCs can contain RAM, and Apple’s decision to put RAM modules aside is something we’ve seen from the company since at least 2018. If you look at this iPad Pro 11 iFixit breakdown, you can see the RAM sitting on the Side with the A12X processor.

What’s different now is that this approach also gets to the Mac, a full-fledged computer designed for higher loads.

CONNECTED: What is the Apple M1 chip for Mac?

The Basics: What is RAM and Memory?

Two DDR4 ram sticks with black heat distributor.
Buccaneer

RAM stands for abbreviated access memory. This is the primary component of system memory, which is a temporary storage place for the data currently used by the computer. This can be anything from the files needed to run the operating system to the table you just edited to the contents of open browser tabs.

When you decide to open a text file, the CPU receives these instructions as well as which program to use. The CPU then collects the data required for all operations and loads the necessary information into memory. The CPU then manages the changes made to the file by accessing and manipulating the data in memory.

RAM usually exists in the form of these long, thin sticks that fit into special slots on your laptop or desktop motherboard, as shown in the image above. RAM can also be a simple square or rectangular module that is soldered to the motherboard. Either way, the RAM on PCs and Macs has traditionally been a discrete component that had its own place on the motherboard.

M1 RAM: The discreet roommate

Graphics showing the different parts of the M1 processor.
Apple

So the physical RAM modules are still separate entities, but they sit on the same green socket as the processor. “Big oops,” I hear you say. – What’s the matter? Well, first of all, it means faster access to memory, which inevitably improves performance. In addition, Apple is changing memory usage within the system.

Apple’s approach called “Unified Memory Architecture” (UMA). The basic idea is that the RAM of the M1 is a single set of memory that can be accessed by all parts of the processor. First, it means that if the GPU needs more system memory, it can increase usage while other parts of the SoC go down. Better yet, there is no need to carve memory portions to individual parts of the SoC and then transfer the data to different parts of the processor between the two locations. Instead, the GPU, CPU, and other parts of the processor can access the same data at the same memory address.

To understand why this is important, imagine the comprehensive features of how a video game works. The CPU first receives all the instructions for the game and then loads the data needed by the GPU onto the graphics card. The graphics card then takes all the data and runs it on its own processor (GPU) and built-in RAM.

Even if you have an integrated graphics processor, the GPU usually retains its own portion of memory, just like the processor. They both work independently on the same data and then commute the results back and forth between their memory locations. By rejecting the requirement to move data back and forth, you can easily see how keeping the same virtual filing cabinet can improve performance.

For example, Apple describes the unified memory architecture on the official M1 website as follows:

“M1 also includes our unified memory architecture, i.e. UMA. The M1 combines its high-bandwidth, low-latency memory within a single package within a single band. As a result, all SoC technologies can access the same data without copying it across multiple memory sets. This dramatically improves performance and energy efficiency. Video applications are more clickable. The games are richer and more detailed. Image processing is lightning fast. And your whole system is more sensitive. “

And it’s not just that all components can access the same memory in the same place. As Chris Mellor points out to The Register, Apple uses high-bandwidth memory here. Memory is closer to the CPU (and other components) and can only be accessed faster than a traditional RAM chip that plugs into a socket on a motherboard.

Apple is not the first company to test unified memory

Diagram showing how the CPU and GPU core can use Nvidia's unified memory feature.
NVIDIA diagram of the company’s Unified Memory function from the beginning. NVIDIA

Apple is not the first company to approach this problem. For example, NVIDIA began offering developers a hardware and software solution called Unified Memory about six years ago.

For NVIDIA, Unified Memory provides a single memory location that is “accessible from any processor in the system”. In the world of NVIDIA, when it comes to CPU and GPU, the same data goes to the same place. Behind the scenes, however, the system flips the required data between separate CPU and GPU memory.

To the best of our knowledge, Apple does not follow behind-the-scenes techniques. Instead, each part of the SoC accesses the data in memory in exactly the same location.

At the heart of Apple’s UMA is better performance through faster RAM access and a shared memory pool that eliminates performance penalties for moving data to different addresses.

How much RAM do you need?

The M1-based MacBook Pro

Apple’s solution is not both sunshine and happiness. Because the M1 has integrated RAM modules so deeply, you can’t upgrade after purchase. If you choose an 8GB MacBook Air, you won’t be able to increase the device’s RAM later. To be fair, updating the RAM for a while wasn’t something you could have done on a MacBook. Something the previous Mac Minis could do, but the new M1 versions didn’t.

The first M1 Macs are 16GB in size – you can get an M1 Mac with 8GB or 16GB of memory, but you can’t get more than that. It’s no longer just about plugging a RAM module into a slot.

So how much RAM is needed? When it comes to Windows PCs, the general advice is that 8GB is more than enough for basic computing tasks. Players are advised to extend this to 16GB, and “prosumer” activity may need to double again for tasks such as editing high-definition video files.

Similarly, with M1 Macs, an 8GB base model should suffice for most people. In fact, it can cover even the toughest everyday uses. It’s hard to say because most of the benchmarks we’ve seen carry M1 in synthetic benchmarks that push the CPU or GPU.

What really matters is how well an M1 Mac can handle keeping multiple programs and a range of browser tabs open. This not only tests the hardware, remember, as software optimization can go a long way in improving this kind of performance, so there has been an emphasis on benchmarks that are truly capable of pushing the hardware. In the end, though, we’d think most people just want to see how new Macs handle “real world” use.

Stephen Hall achieved impressive results on the 9to5 Mac with the M1 MacBook Air with 8GB of RAM. In order for the laptop to start rolling, it had to open a Safari window with 24 web tabs, six more Safari windows that played 2160p video, and Spotify running in the background. He also took a screenshot. “The computer eventually shut down completely only,” Hall said.

TechCrunchon’s Matthew Panazarino went even further with an M1 MacBook Pro rocking 16GB of RAM. It opened 400 tabs in Safari (plus a few more programs) and it ran fine without any problems. Interestingly, he tried the same experiment with Chrome, but Chrome caught fire. But he said the rest of the system performed well despite problems with the Google browser. In fact, during his tests, he noticed that the laptop had swapped places in an instant, with no noticeable loss of performance.

When your computer runs out of RAM, it uses up available SSD or hard disk storage as a temporary memory pool. This may reveal a noticeable slowdown in performance, although it doesn’t seem to be the case with M1 Macs.

These are just casual everyday experiences, unofficial tests. Nevertheless, they are likely to be representative of the expectations of intensive everyday use, and given the memory, 8 GB of RAM should be good for most people who don’t open hundreds of browser pages.

However, if you find that you are editing large, multi-gigabyte images or video files while browsing a few dozen pages and streaming a movie in the background on an external monitor, then perhaps choosing the 16GB model is the better choice.


This isn’t the first time Apple has rethought its Mac systems and moved to a new architecture.

CONNECTED: Deja Vu: A brief history of all Mac CPU architectures