The Elimination of RAM

[MY_USERNAME]

I’m sure I’m a bit misguided in my thinking, since I can’t find a good discussion about it on the inter-webs.

Is Apple looking to completely eliminate the need for RAM with its Mx silicon?

Specifically, the end goal would be to produce a computer with 0bytes of RAM but every bit as efficient as its PC equivalent with RAM.

 

[sirio76]

AS have plenty of RAM, it’s just integrated in the SoC.

 

[cupcakes2000]

I’m sure I’m a bit misguided in my thinking, since I can’t find a good discussion about it on the inter-webs.

Is Apple looking to completely eliminate the need for RAM with its Mx silicon?

Specifically, the end goal would be to produce a computer with 0bytes of RAM but every bit as efficient as its PC equivalent with RAM.

Can’t see how it’s possible to eliminate RAM - it’s the memory for the computer to operate with. It’s somewhat reimagined by being on the SOC, not replaceable, using a unified system etc, but it’s as far as they can go without an entirely different form of computing. Even with some as yet not invented or not imagined form of computing- it will still need some place to collect it’s thoughts, so to speak. RAM essentially.

 

[MY_USERNAME]

AS have plenty of RAM, it’s just integrated in the SoC.

No because arguably that's already being done from my understanding. Essentially, it would be eliminating that process of storing temporary data in the RAM.

Sorry for potentially misusing vocabulary. My tech skills are weak.

 

[MY_USERNAME]

Can’t see how it’s possible to eliminate RAM - it’s the memory for the computer to operate with. It’s somewhat reimagined by being on the SOC, not replaceable, using a unified system etc, but it’s as far as they can go without an entirely different form of computing. Even with some as yet not invented or not imagined form of computing- it will still need some place to collect it’s thoughts, so to speak. RAM essentially.

You're probably right. I know that Apple hates RAM because of its power consumption (I don't remember where I read this.) I figure that would be a huge boost to battery life if it were possible to eliminate RAM.

 

[JPack]

There is no way to eliminate RAM. You can either store a lot of things or access it quickly, but not both.

Apple isn't integrating RAM in to the SoC either. Current memory is created using 10nm lithography vs. 5nm for logic. Apple is simply putting RAM closer to the SoC. This allows for higher bandwidth and higher profit margins.

When the M1 was released last year, there was a myth that RAM was somehow less important. Hopefully, that myth has been busted now that we're seeing 64GB M1 systems.

 

[Boil]

I’m sure I’m a bit misguided in my thinking, since I can’t find a good discussion about it on the inter-webs.

Is Apple looking to completely eliminate the need for RAM with its Mx silicon?

Specifically, the end goal would be to produce a computer with 0bytes of RAM but every bit as efficient as its PC equivalent with RAM.

Computers will always have RAM, all Apple is doing is eliminating removable / replaceable RAM.

 

[chabig]

You're probably right. I know that Apple hates RAM because of its power consumption (I don't remember where I read this.) I figure that would be a huge boost to battery life if it were possible to eliminate RAM.

It would be a bigger boost to battery life if Apple would eliminate the CPU/GPU. But since CPUs, GPUs, and RAM are all integral parts of a computer, you can be sure that none of them are going away. ?

 

[MY_USERNAME]

Computers will always have RAM, all Apple is doing is eliminating removable / replaceable RAM.

I would argue they are also increasing the efficiency of RAM and other systems so that less RAM is necessary to accomplish the same tasks.

 

[MK500]

I’m sure I’m a bit misguided in my thinking, since I can’t find a good discussion about it on the inter-webs.

Is Apple looking to completely eliminate the need for RAM with its Mx silicon?

Specifically, the end goal would be to produce a computer with 0bytes of RAM but every bit as efficient as its PC equivalent with RAM.

In a way, but not entirely. One way to look at it is that RAM is becoming more similar to what we thought of as cache and drive space is becoming more like RAM from a performance perspective, but each still has their roles.

UNIX operating systems like macOS have always done a good job of swapping parts of running software from RAM to disk extremely quickly so that only the most active processes are using valuable RAM. And cache inside the CPU does something similar with even smaller chunks of code.

In the days of spinning disks, swapping to disk was EXTREMELY costly in performance. We all remember pushing our old macs and hearing the disk go crazy and crunch away. As spinning disks turned into SSD, the penalty for swapping became WAY less. Now we are entering into a period that almost feels as big as the jump from spinning disk to SSD. These new SSDs are SO fast that they do indeed start to approach the realm of the performance of RAM (at least RAM from years ago).

But RAM has a much bigger pipeline to the CPU than disk. So even though swapping to the newest insanely fast SSD allows us to do a lot more with a lot less RAM, we still need RAM. Disk (SSD) without RAM would bottleneck the CPU as the number of small data requests is exponentially higher; so we need that RAM to keep things zipping along at full speed.

At this point in time, each level of storage has different costs associated with it. On chip cache is the most expensive. RAM is the next most expensive, and SSD is the least expensive.

Over time, as our mastery of silicon (and maybe other materials) grows, we may get to the point that all these things start to converge. I do believe the excellent chip engineers at Apple see this future and are incrementally moving us toward this. There are likely many years to go, but the performance improvements are going to continue to be mind blowing.

If you are curious about bleeding edge computing design, take a look at this video. I’ve linked to the part of the video that talks about the new computing architecture they have designed. Things are changing rapidly; and many computers in 5 or 10 years may look nothing like what they do today.

 

[MY_USERNAME]

In a way, but not entirely. One way to look at it is that RAM is becoming more similar to what we thought of as cache and drive space is becoming more like RAM from a performance perspective, but each still has their roles.

UNIX operating systems like macOS have always done a good job of swapping parts of running software from RAM to disk extremely quickly so that only the most active processes are using valuable RAM. And cache inside the CPU does something similar with even smaller chunks of code.

In the days of spinning disks, swapping to disk was EXTREMELY costly in performance. We all remember pushing our old macs and hearing the disk go crazy and crunch away. As spinning disks turned into SSD, the penalty for swapping became WAY less. Now we are entering into a period that almost feels as big as the jump from spinning disk to SSD. These new SSDs are SO fast that they do indeed start to approach the realm of the performance of RAM (at least RAM from years ago).

But RAM has a much bigger pipeline to the CPU than disk. So even though swapping to the newest insanely fast SSD allows us to do a lot more with a lot less RAM, we still need RAM. Disk (SSD) is still too slow to not bottleneck the CPU; so we need that RAM to keep things zipping along at full speed.

At this point in time, each level of storage has different costs associated with it. On chip cache is the most expensive. RAM is the next most expensive, and SSD is the least expensive.

Over time, as our mastery of silicon (and maybe other materials) grows, we may get to the point that all these things start to converge. I do believe the excellent chip engineers at Apple see this future and are incrementally moving us toward this. There are likely many years to go, but the performance improvements are going to continue to be mind blowing.

If you are curious about bleeding edge computing design, take a look at this video. I’ve linked to the part of the video that talks about the new computing architecture they have designed. Things are changing rapidly; and many computers in 5 or 10 years may look nothing like what they do today.

Awesome input!

 

[cupcakes2000]

Apple is simply putting RAM closer to the SoC

It’s more than that - it’s a SiP design, which is just a different form of SoC to all intents and purposes. They didn’t just ‘move the ram closer’.

 

[leman]

I’m sure I’m a bit misguided in my thinking, since I can’t find a good discussion about it on the inter-webs.

Is Apple looking to completely eliminate the need for RAM with its Mx silicon?

Specifically, the end goal would be to produce a computer with 0bytes of RAM but every bit as efficient as its PC equivalent with RAM.

Isn’t this the end goal of computing altogether? There is a lot of research being invested into fast persistent memory. If the SSD is as fast as RAM, why would you need RAM? In this regard Apple Silicon is no different to any other computer.

The reason we have all this complex memory hierarchy is because we don’t know how to build memory that can do it all.

 

[Krevnik]

Isn’t this the end goal of computing altogether? There is a lot of research being invested into fast persistent memory. If the SSD is as fast as RAM, why would you need RAM? In this regard Apple Silicon is no different to any other computer.

The reason we have all this complex memory hierarchy is because we don’t know how to build memory that can do it all.

Isn't that the truth though. But so long as RAM is keeping an order of magnitude (or two) advantage on SSDs, and so long as SSD write endurance is a thing, I don't see this dynamic changing except in specific scenarios where you can more fully control bandwidth requirements.

 

[bobcomer]

Isn't that the truth though. But so long as RAM is keeping an order of magnitude (or two) advantage on SSDs, and so long as SSD write endurance is a thing, I don't see this dynamic changing except in specific scenarios where you can more fully control bandwidth requirements.

But think of the power management savings if you could turn the machine off altogether instead of sleep, then no boot time to get back up and running from where you were.

I don't think SSD's will ever get to that level, but some other kind of static RAM, almost for sure, and probably sooner than we think.

 

[Krevnik]

But think of the power management savings if you could turn the machine off altogether instead of sleep, then no boot time to get back up and running from where you were.

I don't think SSD's will ever get to that level, but some other kind of static RAM, almost for sure, and probably sooner than we think.

If SRAM doesn’t need power, then it’s NVRAM.

Joking aside, it’s the ability to be non-volatile that eats transistors/time/etc. So I’m a little of the mind that without a wholly different approach, what has been happening for the last 40 years will continue: while NVRAM technology like NAND Flash will continue to evolve, those same changes will tend to benefit DRAM technology as well (i.e. process changes, data encodings, etc). Especially as current SSD technology has had to accept higher latencies and slower speeds in order to achieve higher densities for storage use (QLC vs TLC for example).

There were a couple esoteric forms of NVRAM that I’ve run across from the 90s that were interesting and closer to SRAM, like FeRAM: https://en.wikipedia.org/wiki/Ferroelectric_RAM, where the base technology dates back 70 years now. It just never scaled, but I think you can still get some of these chips.