> It's sad that we have to resort to soldering components that used to be easily replaceable
Those memory components are now operating at much, much higher frequencies than in the good old days, and need to be mounted closer and closer to the compute units to reduce latency, making soldering them the easiest way for reliably achieving those frequencies and latencies for best performance.
Especially when you look at high end (g)DDR5 frequencies for APUs and what Apple is cooking with the Mx chips. There's no way you can easily get that using DIMM slots, so soldering it is.
Yeah we used to even have memory slots on GPUs back in the '90s, but those days are long gone now and it's not all due to planned obsolescence but simply due to the performance requirements whos goalposts have moved leagues and bounds since then.
> Those memory components are now operating at much, much higher frequencies than in the good old days,
Not arguing about the latency but the chips that have been soldered in that article are operating at much, much lower frequencies than some DDR5 DIMMs you can buy to slot on a desktop mainboard supporting an Intel Raptor Lake or Ryzen 7000 series CPU.
I was talking about modern day designs with high frequency and low latency memory, not that older laptop in particular.
Plus, isn't that laptop also thinner than models with slotted RAM? If you're chasing thinness then soldered is the only way to go even if you're not yet limited by frequency/latency.
> Especially when you look at high end (g)DDR5 frequencies for APUs and what Apple is cooking with the Mx chips. There's no way you can easily get that using DIMM slots, so soldering it is.
Apple's latest M3 processors use LPDDR5-6400. There are systems with socketed memory that do DDR5-6400, so I am not sure what you are talking about. In the server space, Intel's latest Xeons support MRDIMMs at 8800MT/s, which is even faster.
Yeah, this is where Apple has an advantage. It doesnt use special RAM, but what it does do is give you more than 2 channels in a laptop (on Pro/Max/Ultra).
LPDDR5 does use less power than DDR5, but what I was specifically referring to is the claim that Apple gets higher frequencies than you could get with slotted memories. This isnt true.
Interesting. Is there any comparisons on the difference in latencies at which those memories operate? I assume Apple's memory operates at lower latency due to it being directly on the CPU package rather than in a socket on the motherboard.
They are using slower ram. You're comparing apples to oranges. Look at the performance per watt of a modern laptop compared to a desktop or server with slotted components. Performance per watt and energy efficiency goes hand in hand with tighter integration of components.
Those memory components are now operating at much, much higher frequencies than in the good old days, and need to be mounted closer and closer to the compute units to reduce latency, making soldering them the easiest way for reliably achieving those frequencies and latencies for best performance.
Especially when you look at high end (g)DDR5 frequencies for APUs and what Apple is cooking with the Mx chips. There's no way you can easily get that using DIMM slots, so soldering it is.
Yeah we used to even have memory slots on GPUs back in the '90s, but those days are long gone now and it's not all due to planned obsolescence but simply due to the performance requirements whos goalposts have moved leagues and bounds since then.