3 ms·

Only 18% CPU performance improvement.. looks like we are finally getting diminishing returns.

dougmwne 4y ago
This was on the same manufacturing process as the M1. There's generally a bigger boost when moving to a smaller process. But yes, there are diminishing returns, especially when you look at single core performance. Power efficiency and transistor counts are still increasing decently.
- sydthrowaway 4y ago
  18% is pretty good for the same process. So all these gains are from microarchitectural smarts only? Or a case of Apple presenting fab gains as their own?
  - adrian_b 4y ago
    The 18% is for multi-threaded benchmarks. They did not give any number for single-thread speed, so it can be expected that there the improvement is much less impressive. They said that the little cores have been improved substantially, so it is likely that a good part of the 18% comes from the faster little cores.
xiphias2 4y ago
3nm TSMC process isn't ready yet, we have to wait more for it. I agree that it's not a huge upgrade like what M1 was.
- onetom 4y ago
  I completely agree with "not a huge upgrade", but calling it "diminishing returns" is way too extreme.
onetom 4y ago
I would call 1.8% a diminishing return :) 18% is pretty decent, IMHO, for their bottom-most spec, especially given the RAM limit is being extended to 24GB, while the power consumption is still brutally good (18hrs video playback) and the screen is also a bit more spacious at 13.6"... and it's not like they don't have faster CPUs. it's just the cheapest you can get and even that is getting faster! tbh, im using an M1 Mac Mini since last December and it's sooo freaking fast and smooth, that I'm not even really motivated to upgrade at all, if it wouldn't be for the extra memory. though the 16GB is surprisingly sufficient too, if i have run browser tab suspender extensions and close my unused electron apps (notion, logseq, slack), while coding (in Clojure, using IntelliJ and Emacs).
onetom 4y ago
here are some real-world numbers for starting a Clojure process: 1. 2017 3.1GHz Quad i7 clj -M -e 1 1.73s user 0.12s system 232% cpu 0.795 total 2. 2020 M1 Mac Mini clj -M -e 1 0.65s user 0.05s system 157% cpu 0.445 total and I'm getting a very similar time-ratio, when I'm running our test suite for example. I find these speeds extremely satisfying already!

This was on the same manufacturing process as the M1. There's generally a bigger boost when moving to a smaller process. But yes, there are diminishing returns, especially when you look at single core performance. Power efficiency and transistor counts are still increasing decently.

sydthrowaway 4y ago
18% is pretty good for the same process. So all these gains are from microarchitectural smarts only? Or a case of Apple presenting fab gains as their own?
- adrian_b 4y ago
  The 18% is for multi-threaded benchmarks. They did not give any number for single-thread speed, so it can be expected that there the improvement is much less impressive. They said that the little cores have been improved substantially, so it is likely that a good part of the 18% comes from the faster little cores.

xiphias2 4y ago

3nm TSMC process isn't ready yet, we have to wait more for it. I agree that it's not a huge upgrade like what M1 was.

onetom 4y ago
I completely agree with "not a huge upgrade", but calling it "diminishing returns" is way too extreme.

onetom 4y ago

I would call 1.8% a diminishing return :) 18% is pretty decent, IMHO, for their bottom-most spec, especially given the RAM limit is being extended to 24GB, while the power consumption is still brutally good (18hrs video playback) and the screen is also a bit more spacious at 13.6"... and it's not like they don't have faster CPUs. it's just the cheapest you can get and even that is getting faster! tbh, im using an M1 Mac Mini since last December and it's sooo freaking fast and smooth, that I'm not even really motivated to upgrade at all, if it wouldn't be for the extra memory. though the 16GB is surprisingly sufficient too, if i have run browser tab suspender extensions and close my unused electron apps (notion, logseq, slack), while coding (in Clojure, using IntelliJ and Emacs).

onetom 4y ago

here are some real-world numbers for starting a Clojure process: 1. 2017 3.1GHz Quad i7 clj -M -e 1 1.73s user 0.12s system 232% cpu 0.795 total 2. 2020 M1 Mac Mini clj -M -e 1 0.65s user 0.05s system 157% cpu 0.445 total and I'm getting a very similar time-ratio, when I'm running our test suite for example. I find these speeds extremely satisfying already!