661–670 of 1001 posts
They push these new generations out so quick, and with crazy performance boosts. Impressive
Now if only Apple would sell these for use outside of their walled garden.
Base models only: - M1 | 5 nm | 8 (4P+4E) | GPU 7–8 | 16-core Neural | Memory Bandwidth: 68.25 GB/s | Unified Memory: 16 GB | Geekbench6 ~2346 / 8346 - M2 | 5 nm (G2) | 8 (4P+4E) | GPU 8–10 | 16-core Neural | Memory Bandwidth: 100 GB/s | Unified Memory: 24 GB | Geekbench6 ~2586 / 9672 - M3 | 3 nm (first-gen) | 8 (4P+4E) | GPU 8–10 | 16-core Neural | Memory Bandwidth: 100 GB/s | Unified Memory: 24 GB | Geekbench6 ~296…
+------+------------------+--------------+----------+----------------+-------------------+-------------------+---------------------------+
| Chip | Process | CPU Cores | GPU | Neural Engine | Memory Bandwidth | Unified Memory | Geekbench6 (Single/Multi) |
+------+------------------+--------------+----------+----------------+-------------------+-------------------+---------------------------+
| M1 | 5 nm | 8 (4P+4E) | 7–8 | 16-core Neural | 68.25 GB/s | 16 GB | ~2346 / 8346 |
| M2 | 5 nm (G2) | 8 (4P+4E) | 8–10 | 16-core Neural | 100 GB/s | 24 GB | ~2586 / 9672 |
| M3 | 3 nm (first-gen) | 8 (4P+4E) | 8–10 | 16-core Neural | 100 GB/s | 24 GB | ~2965 / 11565 |
| M4 | 3 nm (second-gen)| 10 (4P+6E) | 8–10 | 16-core Neural | 120 GB/s | 32 GB | ~3822 / 15031 |
| M5 | 3 nm (third-gen) | 10 (4P+6E) | 10 | 16-core Neural | 153 GB/s | up to 32 GB | ~4133 / 15437 (9-core) |
+------+------------------+--------------+----------+----------------+-------------------+-------------------+---------------------------+Base models only: - M1 | 5 nm | 8 (4P+4E) | GPU 7–8 | 16-core Neural | Memory Bandwidth: 68.25 GB/s | Unified Memory: 16 GB | Geekbench6 ~2346 / 8346 - M2 | 5 nm (G2) | 8 (4P+4E) | GPU 8–10 | 16-core Neural | Memory Bandwidth: 100 GB/s | Unified Memory: 24 GB | Geekbench6 ~2586 / 9672 - M3 | 3 nm (first-gen) | 8 (4P+4E) | GPU 8–10 | 16-core Neural | Memory Bandwidth: 100 GB/s | Unified Memory: 24 GB | Geekbench6 ~296…
Earlier quoted context omitted.
> * Why cannot I not punch in a password while Face ID is working? If I'm skiing, I know Face ID isn't gong to work, stop making me wait. Funny, a similar thing has been driving me crazy on my Ubuntu 20.04 laptop with fingerprint login. When unlocking, I can either enter a password or use fingerprint. On boot, I am not allowed to enter a password until I fail with fingerprint. If I use fingerprint to log in on boot,…
I think this is a GNOME thing...the keychain by default has the same password as the login password, so logging in with the password unlocks it too. fingerprint login doesn't unlock it: https://news.ycombinator.com/item?id=38527876 , https://wiki.archlinux.org/title/GNOME/Keyring
Base models only: - M1 | 5 nm | 8 (4P+4E) | GPU 7–8 | 16-core Neural | Memory Bandwidth: 68.25 GB/s | Unified Memory: 16 GB | Geekbench6 ~2346 / 8346 - M2 | 5 nm (G2) | 8 (4P+4E) | GPU 8–10 | 16-core Neural | Memory Bandwidth: 100 GB/s | Unified Memory: 24 GB | Geekbench6 ~2586 / 9672 - M3 | 3 nm (first-gen) | 8 (4P+4E) | GPU 8–10 | 16-core Neural | Memory Bandwidth: 100 GB/s | Unified Memory: 24 GB | Geekbench6 ~296…
Serious questions. How is Asahi these days? Is it ready as a daily driver? Is it getting support from Apple or are they hostile to it? Are there missing features? And can I run KDE on it?
Though either Fedora itself, how it built with Asahi or just running it with little disk space end up with freeze on boot after random updates. Twice, once without even rpmfusion enabled. Either some weird btrfs issue or I dont know what.
Like I'm Linux dude for two decades and dont do anything fancy, so this is weird. Switched to Asahi Ubuntu on ext4 and it working great so far.
Earlier quoted context omitted.
It's probably due to the Electron bug[1]. A lot of common apps haven't patched up yet. I also have an M2 Pro with 32GB of memory. When I A/B test with Electron apps running vs without, the lag disappears when all the unpatched Electron apps are closed out. 1. https://avarayr.github.io/shamelectron/ Here's a script I got from somewhere that shows unpatched Electron apps on your system: Edit: HN nerfed the script. Foun…
unpatched include Asana, Bitwarden, Dropbox… some pretty high-profile apps
Base models only: - M1 | 5 nm | 8 (4P+4E) | GPU 7–8 | 16-core Neural | Memory Bandwidth: 68.25 GB/s | Unified Memory: 16 GB | Geekbench6 ~2346 / 8346 - M2 | 5 nm (G2) | 8 (4P+4E) | GPU 8–10 | 16-core Neural | Memory Bandwidth: 100 GB/s | Unified Memory: 24 GB | Geekbench6 ~2586 / 9672 - M3 | 3 nm (first-gen) | 8 (4P+4E) | GPU 8–10 | 16-core Neural | Memory Bandwidth: 100 GB/s | Unified Memory: 24 GB | Geekbench6 ~296…
And the fastest M4 max was already fastest single and multicore CPU by a decent margin, while the fastest non-Apple CPU was only specialized for single or multi.
The fastest multicore CPUs are the ones with a lot of cores, e.g. 64+ core Threadrippers. These have approximately the same single-core performance as everything else from the same generation because single-core performance isn't affected much by number of cores or TDP, and they use the same cores.
Everyone also uses Geekbench to compare things to Apple CPUs but the latest Geekbench multi-core is trash: https://dev.to/dkechag/how-geekbench-6-multicore-is-broken-b...
This is only the base model, no upgrades yet for the Pro/Max version. The memory bandwidth is 153GB/s which is not enough to run viable open source LLM models properly.
It would take 48 channels of DDR5x-9600 to match a 3090's memory bandwidth, so the situation is unlikely to change for a couple of years when DDR6 arrives I guess