https://arstechnica.com/science/2014/08/are-processors-pushi...
Light travels about a foot in 1 nanosecond so moving RAM from a foot away to touching the CPU saves 2 ns in round-trip latency. That's a big win on modern systems.
In reality it's a bigger effect since electric signals travel slower than light. And the effect is yet bigger because off-package memory must be accessed through buses with several levels of synchronization and buffer and gate delays.
Apple gets away with "unified memory" in the M1 because having the memory on-package means a good deal of the bus and sync and contention logic becomes unnecessary. So everything that touches RAM gets a lot faster. And almost everything touches RAM.
https://www.theregister.com/2020/11/19/apple_m1_high_bandwid...
https://www.macworld.com/article/3597569/m1-macs-memory-isnt...
We are finally moving to the conversion point between smartphone's and PCs. It has been in the making for a while.
Now, if that is a good thing for the consumers ( losing upgradability ) is another thing? But given how PC's have become less and less flexible anyway.
We used to have slots for everything ( GPU, Network, Audio, ...). Now we have ATX boards where you maximally put in a GPU. SATA is going the way of the dodo with m.2 becoming the new standard for anything data. I mean, what do we even change anymore on PC's... CPU, Memory, GPU.
GPU's will become a external device with USB4. Mass storage has been moving to NAS more and more or USB3 devices. The people that need 4+ HDD's are more exceptions. And you can get away with USB3 HUB's + external enclosures.
I questioned for a long time, why we still have Chipset, that artificially segment MB's when the difference has become very small between them anyway. You can easily move that last bit of IO into the CPU SOC.
The days that we buy AMD or Intel SOC's with some default CPU+RAM+IO is probably closer then what most think.
Separate hardware is probably going to become a Server / Workstation Pro only feature ( with big $$$ prices ).
The reality is that hardware has reached a point, that most people did not even upgrade for years anymore. And its more a smaller group / minority that really needs ultra fast hardware.
Flexibility is moved from big MB's to external devices, connected over high speed connections.
Sorry if i have gone a bit off topic but when you mentioned the onboard memory, it got me thinking about how we really are moving to a SOC/NUC/... future for even powerful hardware.
Apple clearly has a huge advantage here with a cutting edge process and for all practical purposes unlimited money to throw at this. Their cash on hand is more than AMD s full market cap.
From all benchmarks the Apple M1 laptops are around 1.8-2x more battery life than anything from AMD/Intel.
That includes compiling/running tests/and not just web browsing/checking mail.
And still, there are more catches: Apple is probably driving the screen at 60Hz instead of 120Hz for the G14, and the screen tends to be one of the main drivers of battery life.
What I'm trying to say is that the talks revolve around the M1 vs x86 for battery consumption, but the big savings can probably come from other components in the system.
You are running the newest of the newest.
My Lenovo Thinkpad X390 which i'm sitting on right now for a while, still has ~8h left to go.
Chrome with 30 tabs, intellij etc. open and running.
I do assume the M1 is doing a great job, don't get me wrong, but i would like to see a more objective comparison.
Personally speaking, the most interesting thing for me to see is that Apple now put a very strong SoC on all its models. From Air to Pro, with a relative good price point (as long as you don't upgrade anything...)
Had Apple switched to AMD Ryzen processors on TSMC's 7nm node the performance improvement will have been similarly stark.