The chips are down

Foreword

I recently chanced upon some benchmark numbers of the new(ish) Apple M1 chip, and the results were rather surprising. Like all good little engineers, I started digging deeper, and frankly, what I found WILL BLOW YOUR MIND! (1)

Note that this post is mostly a nerd-out. But there is a finance bit to it. If you are only interested in the finance bit, skip to the Wild guess section below.

As usual, a reminder that I am not a financial professional by training — I am a software engineer by training, and by trade. The following is based on my personal understanding, which is gained through self-study and working in finance for a few years.

If you find anything that you feel is incorrect, please feel free to leave a comment, and discuss your thoughts.

Disclaimer

Before we begin, I want to note that I am not an Apple fan. At least, not a traditional one. I do have iPads (multiple over the years), iPods (multiple over the years, and technically, my wife’s), and various other Apple services/products.

But for computers, I’ve always been a DOS/Windows/Linux kind of guy. I’ve never owned a Mac of any form (except assigned by work), and I’ve never owned an iPhone.

An impressive little package

I stumbled across a bunch of blog posts and videos and whatnot’s, and it’s probably not very useful to link all of them here. So I’ll just link this one video (2), start at the linked timestamp (around 8m30s). There are a lot of issues with the methodology (3), but it presents a fairly accurate picture based on my understanding of the other blog posts/videos.

In short — the new M1 chip from Apple seems to be outperforming both Intel and AMD chips by a fairly large margin. This particular video is against mobile chips from Intel and AMD, though other sources suggest that general trend carries across to desktop chips. This out performance is also generally true for both CPU and GPU intensive benchmarks. Most importantly, the Intel and AMD chips, are sometimes packaged in machines multiple times more expensive.

The only times the M1 has conclusively lost are generally in one of these categories:

• Benchmarks where more than 4 CPU cores are involved — the M1 only has 4 “performance” cores, so loses out in these benchmarks.
• Benchmarks where an external GPU is used — M1 machines don’t seem to work with eGPUs (at least, none of the comparisons I’ve seen included M1 machine+eGPU).
  • In GPU-intensive benchmarks, and depending on the eGPU, the M1 generally loses by a fair margin.

You do not make potato chips with apples

Apple is not a traditional manufacturer of chips. Yes, they’ve dabbled more and more in recent years, mostly for their iPhones/iPads, but most people think of Apple as more of a purveyor of “final packaged products”, instead of a player in the chips space, like Intel or AMD, both of whom have decades more experience.

There have been many, many other players who have tried to take on the chips space. For the most part, they are either gone or have gone niche — Intel, AMD, nVidia (for GPUs) remain the undisputed heavyweights in this space.

So what is Apple doing, challenging the champions… and seemingly winning?

Golden Apple

On a different note, Apple has also been, traditionally, a manufacturer of higher end products (“lifestyle” products), and the prices they charge tend to reflect that. While they have made forays into the lower end of the consumer market in recent years, people still generally associate Apple with “expensive but good”.

And then the M1 comes along, plonked into relatively affordable machines like the Mac mini, new iPads, MacBook Air, etc., many of which start below the psychological $1,000 barrier. For reference, my first ever computer (in the 80’s) was a $2,000 beast (in size) that probably has less firepower than the modern day handheld calculator.

So it is doubly confounding, that these relatively cheap Apple machines are beating the traditional champions housed in machines that are much more expensive.

Wild guess

A long time ago, chip manufacturing was tightly coupled with chip design — Intel (or AMD, etc.) designs a new chip, and then manufactures that chip in their own foundries. So a large part of a chip manufacturer’s ability to deliver superior products, was based on their ability to research and come up with better chip manufacturing (as opposed to design) techniques (4).

In more modern times, this has changed somewhat. Actual chip manufacturing tends to be outsourced to 3rd party manufacturer like TSMC, Samsung, Qualcomm, etc. The traditional manufacturers also do some manufacturing of their own, but my understanding is that the 3rd parties generally have better technology that can produce better chips (4).

As a somewhat simplistic mental model — Intel/AMD/Apple draws on a piece of paper what they want built (i.e: the chip design), flies it by carrier pigeon to TSMC/Samsung, and it is TSMC/Samsung that does the actual manufacturing.

In this new paradigm of chip manufacturing, it becomes easier to see how it is that a relative upstart (admittedly a very well funded one) is able to produce the M1, and apparently, at a lower price point.

So what?

The traditional strength of the OG chip manufacturers used to come, largely, from their ability to manufacture better chips. That strength apparently has been eroded by this new “Manufacturing Chips as a Service” (MCaaS?). This seems to be best shown by the M1 chip — the first, in a very long time, (that I know of) chip from a non-OG manufacturer that unambiguously beats out chips from the OG group, at least in the mobile space.

Intel recently made announcements that it’s planning on building a bunch of new factories. Part of that seems to be motivated by national security concerns (the new factories are entirely in the US), part of it seems to be motivated by the recent chips shortage. But I can’t help but wonder, if part of it is also they recognize that their traditional strength and thus competitiveness has been eroding for years?

Roger that – short Intel, AMD, nVidia!

No. That’s not what I’m saying at all (though I am seriously rethinking my model of AAPL valuation).

No doubt Intel, AMD, nVidia have leadership that aren’t idiots, and will likely try to turn their ships around. AMD and nVidia, in particular, are also doing pretty well financially (in part due to Intel’s recent stumbles).

Also, remember that hardware is not like software. The additional complexity required to add 1 more bit of data is not always linear — in some cases, adding 1 additional bit of data doubles the complexity of the circuit. So the fact that M1 only has 4 performance cores, while top end OG chips have as high as 64 (hyper threaded) cores suggest that at least at the very high end of performance, the OGs still have a trick or two up their sleeves.

The OGs can also learn from the M1. Traditionally, they’ve made chips with scaling power/performance tradeoffs. Maybe the better way is to just build chips with a fixed number of “performance” cores, and a fixed number of “efficiency” cores. My very naive (and likely outdated) thinking suggests this would make the circuitry quite a bit simpler to design.

It’s still not a foregone conclusion that the M1 heralds a new age of Apple chips everywhere, but it is, at least, a fairly resounding shot across the bow. And maybe, the OGs will take notice and up their game.

Footnotes

1. I’ve also recently read about click baits and the tricks they use. Did it work? 😉
2. I am not endorsing the author of the video — I’ve seen his videos before, but I don’t subscribe to them, I chanced upon this one. That said, his videos are of reasonable quality, though with some caveats (see below).
3. Benchmarks are not very scientific — manufacturers have been known to game them with specialized hardware/code that makes their products seem better than they really are. Also, the benchmarks necessarily must run atop software (the OS, various drivers, etc.), as well as other non-chip hardware (motherboard, RAM, the various buses, etc.). The fan-speed/noise complaints and the thermal imaging bits are also rather iffy — that’s very heavily dependent on the material of the casing, layout of the board, etc.
4. Technically, I’m formally trained to design my own chips. Practically, I go to NewEgg/BestBuy just like everyone else (not endorsement). Which is to say, if I’m misusing some terms, I apologize, please let me know and I’ll correct.