There actually is a reason to make RAM non-modular - socketed RAM has latency overhead that would significantly reduce performance. Given Apple uses a Unified Memory Architecture, latency needs to be as low as possible.
Since I’m not a RAM/motherboard expert, I asked ChatGPT what the performance hit would be for switching to socketed RAM and this is what it came back with. It’s comparing the M3 series with top-of-the-line socketed RAM. If it’s making a mistake, I would love someone with more knowledge to chime in.
While the exact impact depends on workload, from soldered LPDDR5X to socketed DDR5 SO-DIMM could result in:
- General computing: 5-10% slower (due to latency and lower bandwidth)
- Creative workloads (video editing, 3D rendering): 15-30% slower (bandwidth-sensitive tasks)
- Machine learning & GPU-heavy tasks: 20-40% slower (due to GPU-memory bottlenecks)
- Battery life: 10-20% less (higher power draw)
Obviously, battery life doesn’t matter in a desktop, but I suspect those buying desktops are more likely to be using applications and workloads that require the highest performance.
HF electronics is certainly not trivial, and I understand that Apple did a great job in communicating that less and non-upgradable RAM would be better for their users.
There are two flaws here:
1. Non- upgradability means it has to be discharged after use. What's good for the manufacturer is not good for the environment or your wallet.
2. The incremental price the user needs to pay for larger memory are unproportionally higher compared to market price. Again, good for the manufacturer, bad for the user.
3. When you search the web on "unified memory", even the self-proclaimed tech sites mostly quote Apple's marketing messages, directly or in disguise.
If you are interested, here is my analysis as electronic engineer:
a) "unified" - for the user - mainly means "less". Before "unified", i.e. "separate", there was memory for CPU and memory for GPU. The GPU for example has massive memory for textures. So that's basically gone, respectively Apple never really had much of that anyway.
What was in 4GB sitting on Apple's GPU card is now a 4GB part of the precious unified memory. And if you had 16GB RAM for the CPU before and now have 16GB unified, you actually only have 12GB for the CPU as 4GB are used by the GPU (provided you needed/used 4GB in the first place).
b) The latency is mostly in the chips themselves, not in the connection. If you look at DDR RAM Chips, they require a massive number of clock cycles until they can provide the data. Clocking them with a 10 times higher clock speed mostly means they will require 10 times more clock cycles until they deliver the data. Proof: see DDR4 vs DDR5, nearly double the clock speed but a meager performance increase.
c) No doubt a connector will somewhat decrease the speed of the RAM. However, looking at the system performance, it may mean that a certain task may take 2:10 minutes instead of 2:05. In the larger picture, the performance claim you are making is simply far fetched. Looking at the M4 mini, a SSD speed of 3GBps is pathetic compared to what typical non-Apple SSDs can achieve. You want to make anyone believe that socketed RAM would make a difference?
