True, thanks for pointing it out! Could it be used as a drop-in replacement for the Hynix chips Apple sues on M1? I can't find any official information on those parts. Maybe its some custom stuff made for Apple?
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Eta on 32g of Ram for M1/M2? When can we expect the new Mac Book Pro 16’?
- Thread starter jumpcutking
- Start date
- Sort by reaction score
True, thanks for pointing it out! Could it be used as a drop-in replacement for the Hynix chips Apple sues on M1? I can't find any official information on those parts. Maybe its some custom stuff made for Apple?
It depends on what Apple did with the memory controller. I believe that the LPDDR4X spec allows up to 16GB (128Gb) parts—these don’t currently exist. If Apple implemented the complete spec then it should be possible. But I can believe that Apple limited their controller and hardware to 64 Gb.
What I mean is not the logical interface but the physical one. On the M1 pictures I’ve seen the Hynix modules are these slightly elongated chips, but Samsung ones seem to be rectangles. I couldn’t find any information I could clearly interpret, but it seems to me that these different modules are not physically compatible with each other.
You’ve put more effort into this than I have 😀. I have no idea. All I did was search for Samsung LPDDR4X. It doesn’t matter much anyway unless Apple decides to do something.
Yeah, the lockdown was not kind to my mental health
It just would be interesting to know whether Apple could in principle ship an M1 with more RAM, or whether they are limited by the current supply. If M1 requires a specific RAM chip form/pinout and Samsung does not ship that, not much can happen.
How does the intel version of MBP 13" support 32GB LPDDR4X and wouldn't that be possible with M1 too if it the memorycontroller supports more than 16GB. I think it might, but Apple is saving 32GB for higher end machines 🤷♂️Yeah, the lockdown was not kind to my mental health
It just would be interesting to know whether Apple could in principle ship an M1 with more RAM, or whether they are limited by the current supply. If M1 requires a specific RAM chip form/pinout and Samsung does not ship that, not much can happen.
It is slower speed at 3733 MHz where the M1 uses 4266 MHz but I don’t know the physical layout. It is almost certainly 4 chips instead of 2.
Ok, is higher capacity LPDDR4X memory modules available at 3744MHz than 4266MHz? or maybe simply a consequence of using 4 chips and/or not on the same package as the memory-controller.
Sure, but that would have significantly increased the package cost and size (leading to indirect cost increases) as well as reduced power efficiency. Not a good tradeoff for the ability to offer a niche memory configuration. Instead, Apple decided to optimize the M1 for it's intended primary market: entry level consumer machines, where 16GB is more than plenty. As I wrote before, it's a business decision, and I think it makes a lot of sense. Prosumers will get heir own chip with more RAM later this year.
The Intel MBP does not put RAM on the processor package. There is much more space available on the logic board, and those RAM configuration use a whole bunch of RAM chips.
Sure they could definitely have designed the M1 to use more RAM but it might have slowed down their performance a bit. I don’t know enough to say.
It doesn’t appear so but these are not user available parts so it is difficult to find out from a search. If someone has access to various RAM manufacturers catalogs they would be able to tell. @leman says the M1 is using Skhynix parts but I can’t find anything by them over about 4GB online. So I have no idea.
It's not about performance (performance is more or less identical no matter how you wire these things up, Tiger Lake with LPDDR4X has the same aggregate bandwidth and RAM access latency as the M1). The problem is cost and power efficiency. Apple is utilizing RAM technology that has so far ben reserved for mobile phones and tables. This allows them to achieve incredible economy of scale as well as ridiculously low power consumption figures.
Very interesting conversation.
I’m not familiar with how m1 package/SOC works. I’d assume the system on a chip needs actual physical room for the chip allocations.
I haven’t the slightest idea.
I’m not familiar with how m1 package/SOC works. I’d assume the system on a chip needs actual physical room for the chip allocations.
I haven’t the slightest idea.
From what I read Apple also included the DRAM on the chip carrier to reduce latency a bit. But you can’t always believe what you read.
I know that Anandtech was surprised to see that Apple was hitting just about the theoretical LPDDR4X bandwidth limit on the M1.
I've read all kinds of things. There are claims that M1's RAM access latency is lower than in a more traditional memory setup, there are claims that it is significantly higher, and there is everything in between.
And then again, there are actual measurements (as done by Anandtech in their M1 and Tiger Lake reviews). And both M1 and Tiger Lake have the same ~100ns of RAM access latency using LPDDR4X-4266.
That's something else, this was about the bandwidth available to the single core. M1 is indeed fairly unique in this regard. In Intel or AMD designs, individual cores are not able to saturate the entire bandwidth provided by the memory controller. Tiger Lake for example only can achieve up to 21GB/s for a single core, and earlier generations were even lower. So if you want to take advantage of the full RAM bandwidth, you have to use multi-threaded algorithms. In contrast, a single M1's Firestorm core can almost fully saturate the available RAM bandwidth, which is one of the reasons why M1 is so extremely fast in single-threaded workloads that perform excessive memory copies.
You forget that the M1 MBP replaced a two-port Intel MBP that wasn't configurable with more than 16GB of RAM.
Only the four port models were capable of more.... and those haven't yet transitioned to Apple Silicon.
The MBA also was max 16GB on Intel.
Thus IMHO the 16GB limit for the current array of M1 systems is a market positioning decision rather than some sort of technology limitation.
You could spec previous 21” iMacs with more than 16gb couldn’t you? I think it was a bit of an eye opener when the 24” iMacs had the same limitations as the m1 MacBooks. The Mac mini was also previously configurable with more memory.
Driving 64 data lines at 3744MHz will either need higher voltage or very good trace design or both to reduce noise (if I still remember my circuit design classses from years ago.) That’s why the M1 SoC keeps the RAM ICs as close to the SoC as possible. Longer copper traces also will present timing problem the faster you clock your bus, likely resulting in adding additional clocks between address bus to data available duration. In other words, lower performance.
That's why Apple's M1 SOC is able to achieve faster performance even with 8GB of Unified Memory - larger bandwidth, faster instruction cycles, and less wasted clock cycles.....
No wonder my base M1 Mini and base M1 Air runs circles around my:
2019 16" MBP w/ 32GB (personal)
2020 Mini 6 core i7 w/ 64GB (work)
2019 4K iMac w/ 16GB (personal)
2020 13" MBP 4 core i7 w/ 16GB (work)
I think the iMacs could go up to 32 GB of DD4-2666. It also only came with 256 GB SSD or 1 TB Fusion drive. These machines aren't really very comparable. I'm sure iMacs with more RAM are coming sometime in the next year. If you need more than 16 GB of fast RAM, you will have to wait.
Edit: It looks like it could be updated to a 512GB or 1 TB SSD making them closer in capability than I originally thought.
Actual tests show that there is no RAM performance difference (in neither bandwidth nor latency) between M1 and Intel Tiger Lake using same 128-bit (dual-channel or quad-channel depending on how you count) LPDDR4X 4226, even though M1 has RAM integrated on package and Tiger Lake has RAM soldered to the mainboard. The advantages of integrating the RAM on M1 is not performance, it's cost and energy efficiency.
Of course, M1 ends up significantly faster overall, but that's because of the differences inside the SoC (cache hierarchy, interconnect, store/load buffers, memory controller parallelism etc.) not because of the differences in the RAM connection.
You haven't tried it on an M1. Nuff said.
I think the argument I was trying to make is that simply citing that the MacBooks they replaced didn’t have more than 16gb ram to evidence it’s a business decision doesn’t really wash. The iMacs they replaced did have options for greater ram capacity. And now they don’t.
I’m confident that machines are coming which will have more than 16gb of memory. I’m also confident they won’t be M1 chips.
Have you… however I’ve had tests conducted and while it performs faster… playback of compositions are still locked to ram
Utilization. I’m spoiled with my 128gb of ram… can load many minutes into it for playback.
