In 2020 Apple made a bold move; they dumped Intel and shifted to their proprietary silicon for powering their MacBooks. Although the move to the ARM architecture from the x86 design language raised several eyebrows, Apple proved everyone wrong when the MacBooks powered by Apple silicon offered mind-blowing performance per watt.

According to several experts, the shift to the ARM architecture was a big reason for the boost in the performance/watt. However, the new Unified Memory Architecture also played a crucial role in improving the performance of the new generation MacBooks.

So, what is Apple's Unified Memory Architecture, and how does it work? Well, let's find out.

Why Does Your Computer Need Memory?

RAM sticks in a PC motherboard

Before getting into Apple's Unified Memory Architecture, it's essential to understand why primary storage systems like Random Access Memory (RAM) are needed in the first place.

You see, a traditional processor runs at a clock speed of 4 GHz during a turbo boost. At this clock speed, a processor can perform tasks in a quarter of a nanosecond. However, storage drives, like SSDs and HDDs, can only supply data to the CPU every ten milliseconds—that's 10 million nanoseconds. That means in the time between the CPU finishing processing the data it's working on and receiving the next batch of information, it's sitting idle.

This clearly shows that storage drives can't keep up with the processor's speed. Computers solve this problem by using primary storage systems like RAM. Although this memory system cannot store data permanently, it is much faster when compared to SSDs—it can send data in as little as 8.8 nanoseconds: infinitely faster than the quickest SSDs right now.

Two SSDs on a motherboard

This low access time enables the CPU to receive data faster, allowing it to continuously crunch through information instead of waiting for the SSD to send another batch for processing.

Due to this design architecture, programs in the storage drives are moved to the RAM and then accessed by the CPU through the CPU registers. Therefore, a faster primary storage system improves the performance of a computer, and that is precisely what Apple is doing with its Unified Memory Architecture.

Understanding How Traditional Memory Systems Work

A Gpu inside a Computer cabinet

Now that we know why RAM is needed, we need to understand how the GPU and the CPU utilize it. Although both the GPU and CPU are designed for data processing, the CPU is designed to perform general-purpose computations. On the contrary, the GPU is designed to perform the same task on different cores. Due to this difference in design, the GPU is highly efficient in image processing and rendering.

Although the CPU and GPU have different architectures, they depend on primary storage systems for getting data. There are two types of Random Access Memories on a traditional system with a dedicated GPU. This is the VRAM and the system RAM. Also known as Video RAM, the VRAM is responsible for sending data to the GPU, and the system RAM transfers data to the CPU.

A Desktop Gaming PC With an AIO Cooler for the Processor

But to better understand memory management systems, let's look at a real-life example of you playing a game.

When you open the game, the CPU comes into the picture, and the program data for the game is moved to the system RAM. After that, the CPU processes the data and sends it to the VRAM. The GPU then processes this data and sends it back to the RAM for the CPU to display the information on the screen. In cases of an integrated GPU system, both computing devices share the same RAM but access different spaces in the memory.

Data Transfer in Traditional Memory Systems

This traditional approach involves a lot of data movement making the system inefficient. To solve this problem, Apple uses the Unified Memory Architecture.

How Does the Unified Memory Architecture on Apple Silicon Work?

Apple does several things differently when it comes to memory systems.

In the case of generic systems, the RAM is connected to the CPU using a socket on the motherboard. This connection bottlenecks the amount of data sent to the CPU.

On the other hand, Apple silicon uses the same substrate for mounting the RAM and the SoC. Although the RAM is not part of the SoC in such an architecture, Apple uses an interposer substrate (Fabric) to connect the RAM to the SoC. The interposer is nothing but a layer of silicon between the SOC and the RAM.

Compared to traditional sockets, which rely on wires to transfer data, the interposer allows the RAM to connect to the chipset using silicon vias. That means the Apple silicon-powered MacBooks have their RAM baked into the package directly, making it faster to transfer data between the memory and the processor. The RAM is also physically closer to where the data is needed (the processors), thus allowing the data to get where it's needed sooner.

Due to this difference in connecting the RAM to the chipset, it can access high data bandwidths.

breakdown of apples architecture
Image Credit: Apple

In addition to the difference mentioned above, Apple also changed how the CPU and the GPU access the memory system.

As explained earlier, the GPU and the CPU have different memory pools in traditional settings. Apple, on the contrary, allows the GPU, CPU, and Neural Engine to access the same memory pool. Due to this, data does not need to be transferred from one memory system to another, improving the system's efficiency further.

Due to all these differences in the memory architecture, the Unified Memory System offers high data bandwidth to the SoC. In fact, the M1 Ultra provides a bandwidth of 800 GB/s. This bandwidth is substantially more when compared to high-performance GPUs like the AMD Radeon RX 6800 and 6800XT, which offer a bandwidth of 512 GB/s.

This high bandwidth enables the CPU, GPU, and Neural Engine to access vast data pools in nanoseconds. In addition, Apple uses LPDDR5 RAM modules clocked at 6400 MHz in the M2 series to supply data at astonishing speeds.

How Much Unified Memory Do You Need?

content creator video editing

Now that we have a basic understanding of the Unified Memory Architecture, we can look at how much of it you need.

Although the Unified Memory Architecture offers several advantages, it still has some flaws. Firstly, the RAM is connected to the SoC, so users can't upgrade the RAM on their system. Furthermore, the CPU, GPU, and Neural Engine access the same memory pool. Due to this, the amount of memory required by the system increases drastically.

Therefore, if you are someone who surfs the Internet and uses a ton of word processors, 8 GB of memory would be enough for you. But if you use Adobe Creative Cloud programs often, getting the 16 GB variant is a better option as you will have smoother experience editing photos, videos, and graphics on your machine.

You should also consider the M1 Ultra with 128 GB of RAM if you're training many deep learning models or working on video timelines with tons of layers and 4K footage.

Is the Unified Memory Architecture All for the Good?

The Unified Memory Architecture on Apple silicon makes several changes to the memory systems on a computer. From changing how the RAM is connected to the computational units to redefining the memory architecture, Apple is changing how memory systems are designed to improve the efficiency of their systems.

That said, the new architecture creates a race condition between the CPU, GPU, and the Neural Engine, increasing the amount of RAM the system needs.