Can you combine or expand RAM on a soldered M1, M2, M3, or M4 Apple Silicon Mac for local LLMs?

While you cannot solder new physical memory chips inside an M1, M2, M3, or M4 Mac, you are NOT limited to that single Mac's built-in Unified Memory when running local LLMs. By connecting your Mac to a RAMDeck Hardware Hub ($199) over Ethernet, Wi-Fi, or built-in Tailscale, you can pool your Mac's Metal Unified Memory with a second Mac, a Windows or Linux PC, or even iPhones (~3.2 GB active unified memory on iPhone 16) to run 30B, 70B, and 120B+ models without buying a $3,500 Mac Studio.

What are the 3 ways to reuse and combine RAM you already own for local AI?

Scenario 1 is Hybrid Inference on one PC (combining GPU VRAM and motherboard System RAM inside one machine); Scenario 2 is physically installing spare RAM sticks into empty motherboard slots (which requires exact DDR generation and form-factor matches and downclocks to the slowest stick); and Scenario 3 is Network RAM Pooling with RAMDeck, which leaves every computer and phone intact and combines mismatched DDR3, DDR4, DDR5, Apple Silicon Unified Memory, and mobile RAM over your home network in 60 seconds.

Can you combine DDR4 and DDR5 computers—or laptop SO-DIMMs and desktop DIMMs—for local LLMs?

Physically on one motherboard: No—DDR3, DDR4, DDR5, laptop SO-DIMMs, and desktop DIMMs have incompatible pin layouts and voltages. Over your network with RAMDeck: Yes—because each computer manages its own physical memory locally and exchanges only compact activation signals over Ethernet or Wi-Fi, a DDR4 laptop, a DDR5 gaming PC, an M4 Mac mini, and an iPhone pool seamlessly into one shared local AI cluster.

How to Reuse & Combine Existing RAM for Local LLMs (3 Scenarios — Including Soldered M1, M2, M3 & M4 Macs)

How to Reuse & Combine Existing RAM for Local LLMs (All 3 Scenarios — Including Soldered M1, M2, M3 & M4 Macs)

By The TradeMAV Team • Memory Reuse & Apple Silicon Pooling Guide • Part of our Complete Local LLM Guide

Direct Answer: Yes — Depending on Which of the 3 Scenarios You Have

You can reuse and combine RAM you already own for local LLMs in three distinct ways: Scenario 1 combines GPU VRAM + motherboard System RAM inside one PC (built into local apps, though CPU RAM spillover slows to 1–7 tok/s); Scenario 2 physically adds spare RAM sticks into empty motherboard slots (only works if DDR generation and DIMM/SO-DIMM size match); and Scenario 3 leaves every device intact—including soldered M1, M2, M3, and M4 Apple Silicon Macs, Windows/Linux PCs, and phones—and pools their RAM over Wi-Fi, Ethernet, or Tailscale in 60 seconds with the RAMDeck Hardware Hub ($199).

Scenario 1: Combining GPU VRAM + System RAM Inside One PC (Hybrid Inference)

If your desktop or gaming laptop has both a dedicated graphics card (such as an 8 GB or 12 GB GPU) and 16 GB–32 GB of motherboard RAM, local runtimes can combine both pools inside that single machine:

How It Works: Model layers fill your high-speed GPU VRAM first, and any excess layers spill over into your PC’s regular CPU System RAM (see our guide to single-PC GPU Offload vs. multi-device splitting).
The Catch: Motherboard DDR4/DDR5 RAM (~30–80 GB/s) is roughly 5× to 15× slower than GPU VRAM (300–1,000+ GB/s). Once a 30B or 70B model spills heavily into single-PC CPU RAM, speed drops to 1–7 tokens/sec—and if total model size exceeds that one PC’s combined memory, it crashes or stalls on SSD swap (<0.1 tok/s).

Scenario 2: Physically Combining Old & New RAM Sticks in One Motherboard

If you have loose RAM sticks in a drawer—or want to pull sticks out of an older tower to upgrade your main PC—physical hardware rules apply:

Zero Cross-Generation Compatibility: DDR3, DDR4, and DDR5 sticks have different physical notch positions and voltages. You can never slot DDR4 sticks from a 2020 PC into a newer DDR5 motherboard.
Laptop vs. Desktop Form Factor Wall: Laptops use short SO-DIMM modules, whereas desktops use full-length DIMM sticks—they cannot be mixed without leaving one machine stripped.
Speed Downclocking & Stability: Mixing a slower 2400 MHz stick with a faster 3600 MHz stick on the same motherboard forces all installed RAM to downclock to the slowest stick’s speed.
Soldered Ultrabooks & Mini PCs: Most thin laptops and compact systems solder LPDDR memory permanently to the board, leaving zero upgrade slots.

Scenario 3: Combining RAM Across Separate Computers, Macs & Phones Over Your Network (RAMDeck)

Why Network RAM Pooling Beats Opening Computer Cases

Instead of pulling out screwdrivers or throwing away incompatible DDR4 laptops, leave every machine whole and connect them over your home Ethernet or Wi-Fi with the RAMDeck Hardware Hub ($199 • includes 1 Free $29 Official Zero-Install Node USB):

Mix Any Generation & OS Together: Combine a DDR4 Windows laptop + a DDR5 desktop + a soldered M-series MacBook + a Linux Mini PC + iPhones/Android phones into one unified 48 GB to 140 GB+ local AI memory pool.
Zero-Install “Unplug & Walk” USB: Plug the included Node USB into Computer #1, double-click once, physically unplug the thumb drive, and walk it over to activate Computer #2 and Computer #3 in seconds—leaving zero files installed on host drives.
No Downclocking Penalty: Each device keeps its slice of the model in its own native GPU VRAM, Apple Unified Memory, or DDR RAM and passes only tiny kilobyte-sized handoff signals per word.

What About Soldered Apple Silicon Macs (M1, M2, M3 & M4)? Debunking the “Fixed Memory” Myth

Many generic hardware guides tell Apple users: “If you own an M1, M2, M3, or M4 Mac, you cannot combine or expand your RAM—the 8 GB, 16 GB, or 24 GB your Mac shipped with is the hard limit for local LLMs.”

That statement is only true inside the aluminum chassis—and completely false across your network. While Apple solders Unified Memory directly onto the M-series chip package (preventing internal stick upgrades), Apple Silicon’s high-bandwidth Metal Unified Memory makes it one of the best distributed local AI building blocks in the world when pooled externally via RAMDeck:

M3 MacBook + Two iPhones Over Wi-Fi (28.0 tok/s Verified): A 16 GB MacBook that cannot fit a 30B MoE model + long context window alone can borrow unified memory from household iPhones and iPads via our official Apple App Store app (an iPhone 16 contributes ~3.2 GB of active unified memory; modern iPhones/iPads contribute ~2 GB to 4.5 GB)—streaming Qwen 30B MoE at 28.0 tokens/sec over ordinary home Wi-Fi!
M4 Mac mini + Windows RTX 3060 Mini PC (38–39 tok/s Verified): Pool a base-memory Mac mini with an existing Windows or Linux GPU PC to unite Apple Metal + NVIDIA CUDA in one plug-and-play cluster—running Qwen3-Coder 30B at 38–39 tokens/sec with multi-user PDF RAG.
16 GB MacBook on the Road + Home Desktop via Tailscale (31 tok/s Verified): With RAMDeck’s built-in 1-click Tailscale integration, your MacBook can pool RAM with your home desktop or second Mac across the internet from a coffee shop, campus, or client office.
Skip the $3,500+ 128 GB Mac Studio Upgrade Tax: Apple charges hundreds to thousands of dollars for upper Unified Memory tiers. Pooling two existing Macs and a spare PC/Mini PC unlocks 70B (~42 GB) and 120B (~60 GB+, 3.1 tok/s verified across 5 devices) models using hardware you already paid for.

Watch: Soldered M3 MacBook + 2 iPhones Streaming a 30B LLM at 28.0 tok/s Over Wi-Fi

Watch: Combining Mismatched DDR Generations, Macs & Phones Into One AI Pool

Side-by-Side Comparison: All 3 Ways to Combine Existing RAM

Requirement / Scenario	Scenario 1: Hybrid GPU + System RAM (1 PC)	Scenario 2: Mixing Physical RAM Sticks	Scenario 3: Network RAM Pooling (RAMDeck Hub)
Works on Soldered M1/M2/M3/M4 Macs?	Unified Memory already shared internally, capped at 1 Mac	No (Memory is soldered to Apple Silicon chip)	Yes — Pool Macs + PCs + iPhones Externally
Combines DDR4 + DDR5 or Laptop + Desktop?	Single PC only	No (Incompatible slots & voltages)	Yes — Any DDR Gen & Form Factor Together
Breaks Past 16 GB / 32 GB Single-PC Ceiling?	No (Capped at 1 PC’s total RAM)	Limited by empty slots (usually 2 or 4)	Yes — Scale to 48 GB, 96 GB, or 140 GB+
Tools or Technical Skills Needed	Software slider	Screwdrivers, BIOS XMP tuning, anti-static care	Zero Tools, Zero Coding (60-Second Router Hub + USB)
Runs 70B (42 GB) & 120B (65 GB+) Models?	Crashes unless single PC already has 64–128 GB	Rarely (requires costly high-density kits)	Yes — Pool 3 to 5 Existing Home/Office Devices

Want to calculate exactly how many gigabytes your current laptops, Macs, and phones unlock together? Try our Interactive Local LLM RAM Calculator, or explore 6 Practical Ways to Reuse Spare RAM & Computers.

Frequently Asked Questions

I bought a 16 GB M1, M2, M3, or M4 MacBook—am I permanently stuck with small 8B local LLMs?

Not anymore. While you cannot solder extra RAM inside your MacBook, plugging a RAMDeck Hardware Hub ($199) into your router lets you combine your 16 GB MacBook’s fast Metal Unified Memory with a second Mac, a spare Windows/Linux PC, or household iPhones/iPads—immediately unlocking 30B, 70B, and 120B+ local LLMs.

Can I combine RAM from a Windows desktop and a MacBook without using command-line tools?

Yes. Unlike developer-only terminal projects, RAMDeck is a 100% plug-and-play consumer appliance. Plug the Hub into your router and launch the one-click Windows/macOS apps—or use the included Free $29 Zero-Install Node USB—to unite both machines in under 60 seconds.

Does using an NVMe SSD as virtual swap work as a substitute for real RAM on 70B LLMs?

No. Because an LLM must read every active parameter on every single generated token, spilling weights onto NVMe SSD swap drops speed below 0.1 tokens/sec and wears out SSD flash cells. Pooling real physical RAM across 2 to 5 existing devices via RAMDeck keeps 100% of weights resident in fast memory.

Combine Every Gigabyte You Already Own — Macs, PCs & Phones

Order the plug-and-play RAMDeck Hardware Hub ($199 • 134% funded on Indiegogo • ships in January with 1 FREE $29 Official Zero-Install Node USB in every box), or read our full 10-chapter book Pool the RAM You Already Own (ISBN 979-8-16746-277-9) free at ramdeck.com/book.

Buy RAMDeck Hub Read the Free Online Book

— The TradeMAV Team • Trademark Notice: TradeMAV™ is a trademark of TradeMAV LLC. Apple, Mac, MacBook, Mac mini, Mac Studio, Apple Silicon, iPhone, iPad, Windows, Linux, Android, LM Studio, Ollama, and Tailscale are trademarks of their respective owners, which are not affiliated with, sponsoring, or endorsing RAMDeck or TradeMAV LLC.