Benchmarking the M1

In my last post, I talked about [upgrading my laptop](https://www.kennyvh.com/posts/upgrading-my-laptop). I'm about a month in and, so far, I'm extremely satisfied with it and would recommend it to *most* people.

I used the qualifier "most", because although this machine has worked delightfully for me, there are a few things that could make it less than ideal for some use-cases that I won't fully go into detail here.

## My use-case

I'll be focusing on my use-case in this post, so what do I actually use my laptop for?

* web browsing
* checking emails
* discord
* recently, [TFT](https://teamfighttactics.leagueoflegends.com/)
* dj'ing
* blogging
* development & programming

Well, most of these things work out of the box. Some of these things work *to some extent*. Luckily for me, none of them don't work at all (although the process to get them set up is less than ideal for many people).

### Surfing the web

Web browsing for the most part, is *extremely* fast. My M1 MBA scores a higher [browser bench](https://browserbench.org/) than my desktop (which I will show in the benchmark section below). In fact, due to the M1 using the ARM architecture, there seems to be an instruction in the ARM ISA that [accelerates javascript](https://news.ycombinator.com/item?id=25233554). This is really great actually since a lot of tools these days offer an SPA version in the browser (not saying this is great, but this is the way things seem to be moving). MS office offers both a web version as well as an ARM native version, so no qualms there either.

### Gaming

I don't play too many games nowadays, I'm mostly in it so I can talk to my friends and hangout in the voice chat. Discord works without a hitch on my M1, which is mostly expected since it's built on top of [electron](https://www.electronjs.org/). TFT also runs quite well on the M1, which is a pleasant surprise as well as some other games like [tabletop simulator](https://store.steampowered.com/app/286160/Tabletop_Simulator/). I have some benchmarks posted in the [benchmarks](#benchmarks) section.

### DJ'ing

I've also recently picked up a [DDJ-400 controller](https://www.pioneerdj.com/en-us/product/controller/ddj-400/black/overview/) that has been a blast to play around with. It uses [rekordbox](https://rekordbox.com/en/), which also works without a hitch on my M1.

### Development

Lastly, programming and developing things is where the most friction is with the M1 (as expected when transitioning to a new architecture). So, depending on what kind of development you do, the M1 may or may not be ready for you yet. I do a broad range of development that is quite literally, "full stack" from embedded pcb design (using KiCAD) to writing embedded software (c/c++/rust) and a good amount of web development too (python/js).

Again, some of these things work out of the box. Some of them require a bit of effort to get setup. In my experience:

At the time, when I set up brew, they were recommending to run it through Rosetta. I didn't want to install and run all my packages in x86 only to reinstall them again when an ARM port came out, so I went ahead and went with the not-recommended native install. This went mostly smoothly, the main difference was manualling installing brew to `/opt/homebrew/` instead of `/usr/local` like on x86 mac os.

From there, when setting up a new machine, I normally just clone my [dotfiles](https://github.com/hkennyv/dotfiles) and run `brew bundle --file $HOME/.khuynh/Brewfile` and it installs all of my casks & formulae.

The rust toolchain didn't work out of the box, but it wasn't too difficult to get setup with. The "rustup-init" brew formulae failed to install, so I simply installed it manually and installed the nightly rust toolchain, which had M1 support. Easy!

I haven't done too much C/C++ development on my laptop yet, but I don't anticipate it to be too difficult. I've been mostly playing around with some ESP32 and STM32 dev boards I have. For these, there are many toolchain options, so I'm not too worried. I use PIC at work and expect there will be some difficulty using proprietary vendor toolchains though.

The javascript/node.js environment and toolchain installed without any issues through brew. I normally use VSCode insiders, so they had released an ARM native version already, so I was good to go!

Lastly, I do mostly data engineering stuff in Python, so the numpy/pandas/scipy stack is important to me. Python itself installed without an issue but problems arise when trying to install some of the scientific packages such as numpy and tensorflow. Since these heavily rely on clibs such as openblas, it took a lot of work to get these to install properly. It's still annoying to do since I make heavy use of virtualenvs, I have to build numpy from source and manually set the `OPENBLAS`env var to point to my installation each time and even then pandas and matplotlib have some issues working with it. I haven't even tried to do any deep learning stuff since I expect even more dependency issues with it. So, unfortunately, I would not recommend someone who uses the scientific python stack heavily to snag an M1 just yet. There are some key issues, however, that you can probably track to see when these issue are resolved: [numpy#18160](https://github.com/numpy/numpy/issues/18160), [numpy#17807](https://github.com/numpy/numpy/issues/17807), [scipy#13409](https://github.com/scipy/scipy/issues/13409).

**EDIT:** For anyone reading this currently, I've been using most of the python scientific stack natively with no issues (haven't tried the deep learning frameworks yet, mostly skl). What I've found to work is the following (assuming you have `brew` installed):

**Installing system dependencies**

```bash
brew install openblas
```

**Installing packages (order matters)**

```bash
pip install cython pybind11
OPENBLAS="$(brew --prefix openblas)" pip install --no-use-pep517 numpy
OPENBLAS=$(brew --prefix openblas) CFLAGS="-falign-functions=8 ${CFLAGS}" pip install --no-use-pep517 scipy
pip install matplotlib pandas scikit-learn
```

## Benchmarks

That being said, let's see some numbers. I'll be comparing 3 machines:

* 2020 M1 Macbook Air 13" 16GB
* mid-2012 Macbook Pro 13" i5 16GB
* Ryzen 2600X, RX580, 24GB (hackintosh + windows)

The benchmarks I'll be running are designed to be more-or-less things that I would *actually* do. So, with that being said, the benchmarks will be:

* cinebench r23 (single core and multi core)
* compiling sqlite (single-job and multi-job)
* running one of my crappy advent of code (AOC) rust programs
* TFT FPS

In addition, I'm publishing all of my benchmarking data in a [gist](https://gist.github.com/hkennyv/0f041cb93552d74a5a546ceb585e1bc7). You can also view the analysis for this [here](https://github.com/hkennyv/kennyvh.com/blob/main/content/notebooks/benchmarking-the-m1.ipynb).

### Cinebench R23 benchmark

![benchmark-cinebench-sc](/images/uploads/benchmarks-cinebench-sc.png "cinebench single core benchmark")

![benchmark-cinebench-mc](/images/uploads/benchmarks-cinebench-mc.png "cinebench multi core benchmark")

### Compiling sqlite benchmark

![benchmark-sqlite1](/images/uploads/benchmarks-sqlite-1.png "sqlite single-job benchmark")

![benchmark-sqlite2](/images/uploads/benchmarks-sqlite-2.png "sqlite multi-job benchmark")

### Rust AOC benchmark

`TODO`

### TFT benchmark

`TODO`

Upgrading my laptop


#### Out with the old...

The time has finally come to retire my beloved daily driver laptop, my
**Mid-2012 Apple Macbook Pro i5 13"**. This laptop was by far the best laptop I've
owned and the best part is, I bought it in 2016 for \$300 from an old coworker!
This was a popular model since it was the last model that was (mostly)
user-servicable. I took full advantage of that in upgrading the RAM to 16GB,
swapping out the 500GB HDD for a 500GB SSD, and even removing the SuperDrive
for this handy [SATA HDD tray](https://smile.amazon.com/gp/product/B0058AH2US/ref=ppx_yo_dt_b_search_asin_title?ie=UTF8&psc=1)
with another 500GB HDD. This made for a fairly snappy machine, at least for a
laptop from 2012.

Sadly, after years of use, some parts of the laptop were wearing and its
internals were starting to show their age. I started to keep an eye out for a
new laptop, some things that I was particularly interested in were:

- USB-C - an absolute must for me. I have two [Lenovo P27h-10 27" monitors](lenovo.com/us/en/accessories-and-monitors/monitors/professional/P27h-10A16270QP127inch-Monitor-HDMI/p/61AFGAR1US)
  at home and I would _love_ to utilize the unified displayport/usb-c pd
  feature. Plus, less cables is always a win for me.
- Higher resolution screen - the Mid-2012 MBP 13" is rocking a 1280x800
  resolution, and while that was probably pretty good in 2012, not having a
  1080p screen (at minimum) in 2020 kind of sucks. Not to mention, my screen
  was actually starting to give out and the left fifth of the screen would
  occasionally go all black w/ green lines
- More powerful processor - the Mid-2012 MBP boasts a dual core, 2nd gen core
  i5 processor. For most things I do, it's not even that big of a deal, but
  having a more modern CPU would be nice :p

I had also considered a [Lenovo Thinkpad T480](https://www.lenovo.com/us/en/laptops/thinkpad/thinkpad-t-series/ThinkPad-T480/p/22TP2TT4800)
for the excellent linux support, decent build quality, and RAM and storage
expandability. However, I decided to stick inside the Apple ecosystem in the
end.

#### In with the new...

The new kid on the block is the Apple's [M1](https://en.wikipedia.org/wiki/Apple_M1)
SOC. Funny enough, I actually had _just_ purchased an Intel 2020 Macbook Pro
i5 13 at the beginning of November, fully aware the M1 was coming out. I had
opted for the Intel 2020 MBP because I was expecting the M1 to be fairly
lackluster and didn't want to deal with any of the first-gen transitional
issues that would come with switching to ARM.

However, after watching the keynote and seeing Apple make some extremely
bold claims in their ["One More Thing" keynote](https://www.youtube.com/watch?v=5AwdkGKmZ0I),
I was both skeptical and very interested at the same time. As an embedded
systems engineer, I'm always super excited to see some innovation in the
hardware space (one of the reasons I've disliked intel's anti-consumer practices
and stagnating products for a while and been super bullish on AMD since zen1
first released in 2016). However, I still wanted to be very cautious with
first-party benchmarks, especially such bold ones as Apple had claimed.

Not too long after, third-party benchmarks flooded youtube & I watched almost
all of them before finally returning my 2020 Intel MBP and pulling the trigger
on a 2020 M1 MBA (with the help of a _very generous_ employee discount from one
of my friends, thanks SM!).

#### Comparison

Here's a table comparing the specs and costs between the two machines:

|               | 2020 Intel MBP 13"                                                 | 2020 M1 MBA 13"                       |
| ------------- | ------------------------------------------------------------------ | ------------------------------------- |
| **Processor** | 10th gen Intel Core i5 (4c)                                        | Apple M1 CPU (8c)                     |
| **Graphics**  | Intel Iris Plus integrated graphics                                | Apple M1 SOC 8-Core GPU               |
| **Storage**   | 512GB SSD                                                          | 512GB SSD                             |
| **RAM**       | 16GB 3733MHz LPDDR4X                                               | 16GB Unified Memory                   |
| **I/O**       | 4 Thunderbolt ports                                                | 2 Thunderbolt / USB 4 ports           |
| **Price\***   | **\$1799.00** (although I purchased an open-box one for \$1349.00) | **\$1449.00** (\$1091.00 w/ discount) |

\* before tax prices

So, in the end, I spent ~\$300 less than I intended to, yay! And for what, I
would argue, is a superior machine (for my needs). So far though, I am happy
with my purchase and will try to continue to write about my experiences with
it.

I think I will do another post with some benchmarks of this new machine and
compare it to what I was getting with my old Mid-2012 MBP for funsies.

Cheers!
