x86 (5nm) vs M1 (10nm)

[ruslan120]

How much of the performance / power efficiency do you think is attributed to design (and vertical integration) versus being built on a smaller transistor size?

If Intel x86 chips were on 5nm today, what do you think the power efficiency/performance gap would be?

What if M1 chips were on 10nm and Intel on 5nm?

(I know this is an extremely complicated topic, involving microarchitecture, software optimization, use cases, etc.) Fun to poke around and speculate.

 

[hugodrax]

Design is the big factor.
Intel is stuck with one leg in the past due to compatibility to the year 1978 and then the Microsoft Windows backcompat ball and anchor holding the other leg back.

There are plenty of articles on this, Anandtech etc have good articles on it.
Bottom line a big contributor is a better 64bit clean design.
#1 twice the registers 32 vs 16
#2 Fixed length instructions which means faster decoding and execution.
#3 Wider decoding capability (8 vs 4 on intel)
#4 More execution units than x86
and on.... and on...

And this is just the first desktop CPU which is really just an A14 with some enhancements, wait till Apple starts dedicating more resources to making Desktop processors in addition to process shrinking.

Next 2 years will be very interesting. Intel has painted themselves into a corner and now the Balloon interest is due. I just don't know how much further they can push x86. They knew this time was going to come and was working on the failed EPIC architecture which was a real bad and foolish bet.

 

[Luposian]

How much of the performance / power efficiency do you think is attributed to design (and vertical integration) versus being built on a smaller transistor size?

If Intel x86 chips were on 5nm today, what do you think the power efficiency/performance gap would be?

What if M1 chips were on 10nm and Intel on 5nm?

(I know this is an extremely complicated topic, involving microarchitecture, software optimization, use cases, etc.) Fun to poke around and speculate.

You're traveling down a road that doesn't exist. M1 chips are 5nm, not 10nm, and they're not going to create a larger chip, as it would be unproductive in the extreme. Unless they could create a cheaper chip (and thus a cheaper computer), in doing so, there would be utterly no reason to go backwards like that. It wouldn't perform as well nor run as cool nor produce as many chips per wafer, etc. If you're going to speculate, speculate on things that will exist... like x86 getting to 5nm or more cores for the M1X or 4nm M2's or whatever.

As long as Apple continues to point the finger of comparison to Apple's own Intel computers, the advantage is obvious. Once they start pointing the finger at other Intel/AMD/nVidia/Radeon whatever, their advantage becomes far less prominent. My M1 Mac Mini blows away my Ryzen 7 laptop CPU (Cinebench R23), but the nVidia GPU in the same laptop blows my M1 Mac Mini apart like there is no tomorrow (Geekbench). So, keep Apples to Apples, and everything is good.

 

[bill-p]

How much of the performance / power efficiency do you think is attributed to design (and vertical integration) versus being built on a smaller transistor size?

If Intel x86 chips were on 5nm today, what do you think the power efficiency/performance gap would be?

What if M1 chips were on 10nm and Intel on 5nm?

(I know this is an extremely complicated topic, involving microarchitecture, software optimization, use cases, etc.) Fun to poke around and speculate.

We already have an Apple Silicon chip that's 10nm. That's the A11.

iPhone X - Geekbench

Benchmark results for an iPhone X with an Apple A11 Bionic processor.

browser.geekbench.com

And here's Intel "at its best" on 14nm, which is said to be "equivalent to TSMC 10nm":

Dell Inc. XPS 13 9365 - Geekbench

Benchmark results for a Dell Inc. XPS 13 9365 with an Intel Core i7-7Y75 processor.

browser.geekbench.com

So even if M1 and Intel chips were on the same (or equivalent) process node, performance difference may not be so pronounced anymore, but... M1 will still be vastly more efficient. So in the worst case, we'll have an Apple Silicon MacBook that performs basically about the same as the Intel counterpart, but with 20 hours of battery life, and still much cooler than the Intel counterpart.

And here's what people had to say about that particular laptop above:
https://www.reddit.com/r/Dell/comments/9vnvx5
Efficiency is a big problem for Intel chips, and I don't think Intel will be able to solve that any time soon.

But that's for efficiency. If efficiency is not such a big deal, and performance is all that matters, Intel and Apple Silicon on the same process will be an interesting matchup, I think.

Sadly, we probably won't see that any time soon. It's clear Intel will be far behind Apple in 2 years. TSMC is already going to move on to 3nm by the end of next year.

 

[jdb8167]

How much of the performance / power efficiency do you think is attributed to design (and vertical integration) versus being built on a smaller transistor size?

If Intel x86 chips were on 5nm today, what do you think the power efficiency/performance gap would be?

What if M1 chips were on 10nm and Intel on 5nm?

(I know this is an extremely complicated topic, involving microarchitecture, software optimization, use cases, etc.) Fun to poke around and speculate.

This really can't be answered because when you design a chip, you are designing to a particular node size and fabrication requirements. It isn't like you just take your 10 nm design and hand it TSMC and tell them to make it in 5 nm. It takes years of work and clever designers to take advantage of the new node size and the benefits of the fab process.

In general, smaller node sizes increase battery life and allow for more transistors for things like more CPU cores but it is way more complicated. There is a reason that the new Qualcomm chip using 5 nm is not going to be as fast as the A14. Design matters.