What is Unified Memory?

When Apple moved the Mac to its own silicon in 2020, it also changed how Mac memory works. Unified memory is central to why Apple silicon Macs perform the way they do — and in 2026, it's the one specification you can least afford to get wrong. Here's how it works, why it matters more than ever, and how much you actually need.

Section 01What memory actually does

To understand unified memory, it helps to be clear on what memory does in the first place. RAM (random access memory) is your Mac's working space — the fast, temporary storage where everything you currently have open actually lives. When you launch an application, its data is loaded from storage into memory, because the connection between the processor and memory is far faster than the connection between the processor and storage.

More memory means more room to work: more applications open at once, more browser tabs, larger files, less time spent shuffling data back and forth to the slower SSD. When memory runs short, macOS starts using a slice of the SSD as overflow — called swap — and that's when a Mac starts to feel sluggish.

In Intel-era Macs, memory came as separate DDR3 or DDR4 modules mounted on (or plugged into) the logic board. That design served the Mac well for decades, and it had one genuine advantage the current design doesn't: in some models, such as the 27-inch iMac, you could add more memory yourself years after purchase.

The old way

Traditional memory modules

A DDR3 SODIMM memory module and a DDR4 memory module — the kind of removable memory found in Intel-era Macs.

Section 02What makes unified memory different

Traditional computers keep separate memory for separate processors. The CPU has its system RAM; a discrete graphics card carries its own video memory (VRAM); and the two pools sit in different places on the motherboard. When the CPU and GPU need to work on the same data — a photo you're editing, a frame of video — that data has to be copied between the two pools.

Apple silicon takes a different approach. The CPU, GPU and Neural Engine (the dedicated processor for machine-learning tasks) are all built into one chip, known as an SoC (system on a chip). The memory itself — high-speed LPDDR5X in current models — is mounted directly onto the same package, millimetres from the processor cores, and connected by a very wide memory bus.

Two things follow from this design. First, every part of the chip shares one pool of memory, so the CPU, GPU and Neural Engine can work on the same data without copying it back and forth. Second, the short distance and wide bus allow far more data to move per second — what's called memory bandwidth. For perspective: the top-of-the-range Intel MacBook Pro 16-inch of 2019 had a memory bandwidth of 41.6GB/s, whereas the M1 Max that replaced it two years later delivered 400GB/s.

The new way

The system on a chip

Visualisation of an Apple silicon SoC. The memory sits on the same package as the CPU, GPU and Neural Engine.

You can see the difference clearly on the logic boards themselves. On an Apple silicon Mac, the memory packages sit directly beside the processor die. On an Intel Mac, the memory is positioned further from the processor — in the MacBook Air below, on the opposite side of the logic board entirely.

MacBook Air M1 (A2337) logic board. Red: Apple silicon processor. Blue: unified memory, mounted on the same package.

MacBook Air (A2179, Intel) logic board, both sides. Red: processor. Blue: memory — on the opposite side of the board.

One honest caveat: because the pool is shared, everything draws from it. On a 16GB Apple silicon Mac, the graphics workload and the system workload are dividing the same 16GB. In practice the efficiency of sharing outweighs the cost of splitting for the vast majority of workloads — but it's worth knowing that "16GB unified" and "16GB of system RAM plus separate VRAM" are not describing the same thing.

Section 03The bandwidth ladder

Unified memory hasn't stood still since 2020. Each generation of Apple silicon has raised both the bandwidth and the maximum capacity, and the Pro, Max and Ultra tiers scale the memory bus up dramatically. The current base M5 chip moves 153GB/s — nearly four times the last Intel MacBook Pro. The M5 Max in the current MacBook Pro reaches 614GB/s, and the M3 Ultra in the Mac Studio reaches 819GB/s, roughly twenty times where Intel Macs left off.

Two rows deserve a closer look. The A18 Pro is the iPhone-derived chip inside the entry-level MacBook Neo, and its 60GB/s bandwidth sits below even the original M1 from 2020 — one of several trade-offs that make the Neo's price possible. And the Intel row is a reminder of how far the platform has come: DDR4 was perfectly good memory, but a laptop chip simply couldn't be fed data at these speeds when the memory sat inches away on the board.

It's also worth being fair to the older Apple silicon here. Bandwidth is not the same thing as overall speed — a newer chip does more with each byte it moves — but the M1 Pro (200GB/s) and M1 Max (400GB/s) still offer more memory bandwidth than today's base M5, which is part of why those 2021 machines remain so capable for memory-hungry creative work.

Section 04Why unified memory matters more in 2026

When we first published this article in 2024, unified memory was mostly a performance story. Two things have changed since then.

On-device AI runs in memory

The first is artificial intelligence. Apple Intelligence runs on every Apple silicon Mac, and the models behind it live in unified memory while they work. More significantly for professionals, the unified pool is what lets a Mac run large third-party AI models locally: because the GPU and Neural Engine can address the entire memory pool, a MacBook Pro M5 Max with 128GB can hold models in memory that would require a five-figure specialist graphics card on a PC. The rule of thumb is blunt — the bigger the model you want to run on-device, the more unified memory you need. Apple's current chips lean into this, with the M5 generation adding Neural Accelerators to every GPU core.

Memory is what the 2026 shortage is squeezing

The second is supply. A global memory shortage, driven by AI data-centre demand, has roughly quadrupled the price Apple pays for the LPDDR5X chips that unified memory requires. We covered the fallout in detail in The 2026 Mac Squeeze, but the short version is that Apple has quietly trimmed the memory configurations it sells: the Mac Studio's 512GB, 256GB and 128GB options have been withdrawn, the Mac mini Pro has lost its 64GB option, and UK prices rose across the lineup in June.

Put those two trends together and you get an uncomfortable squeeze: memory is becoming more useful at exactly the moment it's becoming harder to buy.

Section 05The catch: you can't change your mind later

Because unified memory is mounted on the processor package itself, it cannot be added to, swapped or upgraded — not by you, not by a repair shop, not by Apple. This is true of every Apple silicon Mac, from the MacBook Neo to the Mac Studio. The memory configuration you choose on the order page is the one the machine will have on the day it's recycled.

That makes the memory decision the most consequential one on the configurator. Storage can be extended externally over Thunderbolt; memory can't. And unlike the Intel era, there's no second chance two years in when your needs grow. Given that Apple silicon Macs are proving unusually long-lived, you should configure memory for the workload you'll have in year five, not the workload you have today.

Section 06How much unified memory do you need?

The old principle still applies: more memory means more headroom for multitasking and large files. But the sensible baseline has shifted since this article was first written. In 2024 we suggested 8GB to 16GB was fine for everyday use. In 2026, every new Mac except the MacBook Neo starts at 16GB, and 16GB is the floor we'd recommend for anyone planning to keep a machine for more than a couple of years.

Whatever you're considering, check the recommended system requirements of the software you actually use, and buy to meet or exceed them — future versions only ever get hungrier. And be honest about the other direction too: memory you never touch doesn't make a Mac faster. It buys headroom and longevity, nothing more.

A practical test on your current Mac

If you already own a Mac, it will tell you what you need. Open Activity Monitor, click the Memory tab, and watch the memory pressure graph at the bottom during a normal working day. If it stays green, your current amount of memory is coping and matching it in your next Mac is safe. If it's regularly yellow or red, or the swap figure is climbing into the gigabytes, step up a tier.