Apple M: ​​are there any limits to the Miracle? – Consomac Rumors

Good morning all,

For a long time I have been skeptical of ARM processors. I will not come back to the issues of compatibility and standardization, but I will recall what I said about the competitors, Intel and AMD.

As much, facing the wall that Intel and AMD has been rubbing against for years and now you can almost say it, a decade, I could understand that ARM technology could lag, catch up and even overtake its competitors, as much I am very skeptical about what will happen “after”.

On the other hand, my experience of over 30 years today with Intel chips has shown me that they are pragmatic. Was NetBurst technology a siding? Never mind, they left with a clean sheet with Core technology. For example, I recently learned that within Intel & AMD CPUs, CISC instructions were and for a while already converted to RISC before being executed. Here too, for me, proof of a pragmatic approach.

So damn it, why, when they’ve been stuck for almost 10 years and ARM has been catching up for about that long, why aren’t they taking inspiration from ARM? And here, it is a question of Intel but also of AMD of course.

But back to ARM for a moment. Obviously the release of a second batch of Apple M machine (M1 + or M2, we will see) will be the opportunity to measure the progress margin of the M chips. ‘elsewhere subject to criticism on another forum.

To design it, I, among other things, started with the setbacks observed between the first M Macs and previous versions and I noticed, among other things, the disappearance of certain options: 10 Gbe LAN, eGPU support, etc. This drew my attention to a point that could be problematic in the SoC approach. : communication between the SoC and the rest of the world.

And here is the response that a stakeholder gave :

Citation

For me, the strength of Apple’s SoCs is an incredible integration, including unified memory (no copy or conversion, common L3 cache through the SLC), even more than the performance in absolute value of certain units.
Take a video stream from the webcam, have it processed by the image processor, analyze the result via the NPU then use the GPU to do a different and more complex processing of the image, so that it is accessible to the CPU : here we use the symbiosis of the SoC units and the SLC to obtain performance far above its category, far above.
This conception is antagonistic to that of the Mac Pro, made up of disparate elements from various sources and exchanged through a bus.

–

Before adding :

Citation

The corollary disadvantage is that beyond the limits of the scalability of the SoC, still quite extensible, it hits a wall, and where we add / upgrade GPUs or NPUs, we will not be able to increase performance.

–

In other words, we have summarized here both the strengths of the SoC approach as well as its weaknesses and limitations.

What I understand better then is that the performance gain observed between ARM chips and Intel chips is not only a result of the chip itself, but also of the SoC approach. Approach which eliminates the need for communication standards between the components of the PC itself.

There is obviously a very small downside to the example that the speaker has chosen, it is that it is well chosen so that all the communication takes place at the level of the SoC itself and take full advantage of it. ‘integration.

To put it simply, if Apple releases a component from the SoC, all the benefits of integrating it into the SoC will be lost.

It is therefore necessary to separate these two sources of performance gain when comparing ARM chips in an environment other than the one currently chosen by Apple.

I also understand why Intel & AMD cannot be inspired by this model since it de facto questions all the logic of the PC in question.

I had also, for a while, the fear that the 3D Point (Octane) memory would try to integrate this memory into the chip, just as it had already done for the Northbridge. I do not know if I had mentioned it at the time.

But one thing is clear : to keep the same performance as on the Apple M1, Apple must integrate as much in the SoC of these other machines as on the M1.

If I want to be ironic: Apple will no longer even need glue or solder to make closed machines.

There is still an open question: what are the possibilities for future development?

For the models to be released, they will benefit from the double effect ARM and SoC (with the addition of the possibility for Apple to multiply the cores especially since ARM consumes little). On the other hand, for what follows, I think that the gain of switching to the SoC model is a unique gain. Only the ARM gains will therefore remain from one generation to the next.

What to conclude?

That this approach is probably suitable for the vast majority of current Apple customers. They will benefit from a performance gain for which there are also two prices to pay.

For the first, namely the loss of the possibilities of evolution, they will probably be satisfied with it very well since already ajd everything is stuck on an MBP. It’s a safe bet that even high-end iMacs will follow this path, unless they stick with Intel processors (like one of the three current Mac Mini models).

For the second price to pay (the loss of x86 compatibility), it all depends on Apple in fact. If OSX support is extended over 8-10 years, that should be fine for them.

And for the other customers … either they will do like me, or they will pay double the price.

Source : Discussion

Amended 2 hours ago by Sethenès

–