[lhl]

Tuxedo Computing and Slimbook both sell Ryzen 4800H computers that will outperform the M1 in heavy multithreaded workloads and come with Linux preinstalled. These laptops aren’t quite as slick as the MBP but weigh in at 1.5kg, have huge 91Wh batteries, and have a better keyboard (I have one from a different OEM, but same ODM design). They also have user upgradable memory and storage - I am running with 64GB RAM and 2TB SSD at a total cost (with upgrades) of less than what Apple is charging for their base 8GB/256GB MacBook Pro.

I expect a future “M2” to maybe take the performance crown, but AMD isn’t standing still. Cezanne has Zen 3 cores, which should boost IPC by about 20%, and Rembrandt should get to 5nm and have RDNA2 graphics.

[alwillis]

Tuxedo Computing and Slimbook both sell Ryzen 4800H computers that will outperform the M1 in heavy multithreaded workloads and come with Linux preinstalled. These laptops aren’t quite as slick as the MBP but weigh in at 1.5kg, have huge 91Wh batteries…

1. You're not going to get 20 hours of battery life.

2. Don't forget it's not just the M1—it's the unified memory, the 8 GPU cores and the 16-core Neural Engine. Most CPU and GPU-intensive apps are going to run faster on the M1 than on your machine. Even x86-64 apps using Rosetta 2 on an M1 Mac may run faster, since those apps are translated to native code on the M1.

3. Mac's SSD is probably faster; it's essentially a 256GB cache for the processor.

4. The Mac can run iOS/iPadOS apps too.

5. If done right, Linux compiled for the M1 will likely run faster on an M1 Mac than it does on a machine like yours, especially if Apple provides a way to access certain hardware features.

We’ll have to see what happens but expect these machines to be pretty popular with users, even those who need to run Linux when that the distros are updated.

We shouldn't forget that the underpinnings to all of this is Darwin, the BSD-derived Unix layer which is already running natively on M1, including the compiler and the rest of the toolchain.

[lhl]

> 1. You're not going to get 20 hours of battery life.

Sorry to burst your bubble, but you're not going to get 20 hours of battery life in real world usage on the M1 either. The early tests show about 10-12h, which is the same as my (and many other) laptops under regular usage.

> 2. Don't forget it's not just the M1—it's the unified memory, the 8 GPU cores and the 16-core Neural Engine. Most CPU and GPU-intensive apps are going to run faster on the M1 than on your machine. Even x86-64 apps using Rosetta 2 on an M1 Mac may run faster, since those apps are translated to native code on the M1.

Now it feels like you're just regurgitating marketing talking points. Can you tell me what "unified memory" even is exactly? Is it zero-copy support, because AMD has had that on its APUs since... 2013 or thereabouts. Is it LPDDR4 on a pop package, because all that means to me as an end user is I can never upgrade my memory and that I'm limited 16GB of memory (which I regularly go over - I am using 19GB of RAM right now just with browser tabs open). As for performance, we already know from the early testing that the M1 under-performs 8C Zen2 for heavy MT workloads like compiles and renders, so ... what are you saying exactly, somehow running software via emulation/translation will magically make that faster?

> 3. Mac's SSD is probably faster; it's essentially a 256GB cache for the processor.

Again would you simply assume that a Mac's SSD is "probably faster"? It in fact is not. The 256GB SSD on the M1 MBA was tested at 2676MB/s reads, my value NVMe SSD, a $200 2TB ADATA SX8200PNP does 2917 MB/s on my laptop. As for SSD as cache - what are you talking about? L2/L3 latency is typically about 10ns latency. NVMe latency is typically on the order of hundreds of microseconds, roughly 10,000X slower.

> 4. The Mac can run iOS/iPadOS apps too.

Poorly, but I mean, but surely this irrelevant to Linux performance?

> 5. If done right, Linux compiled for the M1 will likely run faster on an M1 Mac than it does on a machine like yours, especially if Apple provides a way to access certain hardware features.

Which hardware features? This is rhetorical. I know this is just hand-waving.

> We’ll have to see what happens but expect these machines to be pretty popular with users, even those who need to run Linux when that the distros are updated.

We'll see what happens. You can track the state of Docker here, for example: https://news.ycombinator.com/item?id=25119396

> We shouldn't forget that the underpinnings to all of this is Darwin, the BSD-derived Unix layer which is already running natively on M1, including the compiler and the rest of the toolchain.

Darwin/macOS may be POSIX compatible, but it is not production compatible with Linux. Like lots of other devs, I've used Macs in the past (for many years) and you always run into compatibility issues small and not so small until you're either running either a completely parallel devchain via Homebrew or MacPorts, or in a VM. Honestly, WSL these days is a more Linux-friendly dev environment than macOS. But then again, it's even easier/better to run Linux and Docker these days.

[alwillis]

The early tests show about 10-12h, which is the same as my (and many other) laptops under regular usage.

Here's an early test that’s quite different from what you described. I’d bet dollars to doughnuts your laptop can't play fullscreen, 4k/60fps video for 20 hours using only the battery:

In fullscreen 4k/60 video playback, the M1 fares even better, clocking an easy 20 hours with fixed 50% brightness. On an earlier test, I left the auto-adjust on and it crossed the 24 hour mark easily. Yeah, a full day. That’s an iOS-like milestone.

Another one: Just 17% of the battery to output an 81GB 8k render.

These are just a couple of highlights from the article "Yeah, Apple’s M1 MacBook Pro is powerful, but it’s the battery life that will blow you away": https://techcrunch.com/2020/11/17/yeah-apples-m1-macbook-pro...

You have to look at the totality of the what's going on.

In short, the M1 Macs are right up there with the fastest machines available at reasonable prices and at a fraction of the power consumption.

The machines set a new level of performance per watt and there's no disputing that. That's pretty good for their first attempt at Apple Silicon Macs.

[saagarjha]

> The early tests show about 10-12h, which is the same as my (and many other) laptops under regular usage.

Yeah, laptops that look like bricks.

> Can you tell me what "unified memory" even is exactly?

GPU shares memory with the AP

> Is it LPDDR4 on a pop package

On SoC, no PoP

> I'm limited 16GB of memory

Wait for new hardware

> which I regularly go over - I am using 19GB of RAM right now just with browser tabs open

This isn’t how memory works :/

[lhl]

> Yeah, laptops that look like bricks.

Eh, the laptop I'm using at the moment has a 15.6" display and 91Wh battery and is less than 100g heavier and about 1mm thicker than the 13" MBP. It's also 500g lighter than the 16" MBP. Lots of other properly tuned modern x86 laptops can perform similarly. For example the 14" 1.48kg 18mm 56Wh battery HP EliteBook 845 G7 manages >12h on NBC's wifi websurfing test: https://www.notebookcheck.net/HP-EliteBook-845-G7-review-AMD...

> This isn’t how memory works :/

Fair point that free might not be the best way to measure things, I have more tabs open now but still not doing work (obviously), so let's compare:

                total        used        free      shared  buff/cache   available
  Mem:       65328424    26679236    31393956     1165140     7255232    36284812
  Swap:      67108860     3197072    63911788

With totaling per-process shared/private memory (I uses memstat.sh for this). And the total I get is: 18.37 GiB - lower, but actually not so far off.

This is only with a two browsers (a few hundred tabs) and some resident electron apps open, mind you. Before upgrading (w/ 16GB memory) I was often hitting swap, and now I'm not. But if you don't ever need >16GB of RAM, lucky for you I guess.