Longhorn

Twitter: https://twitter.com/never_released

Public Key

npub1p2yl7tnq9q3ps8a7wk23shrv97uwfgmt5953rt35rrsev8evm0nqgme7pz

NIP-05 Address

Profile Code

nprofile1qqsq4z0l9eszsgscr7l8t9gct3kzlw8y5d46z6g34c6p3cvkrukdhespqqq3samnwvaz7tmjv4kxz7fwd4hk6mmnw3ezuurfde4szxrhwden5te0wfjkccte9eekummjwsh8xmmrd9skc09usv8

Publishing to

relay.momostr.pink

relay.snort.social

Author Public Key

npub1p2yl7tnq9q3ps8a7wk23shrv97uwfgmt5953rt35rrsev8evm0nqgme7pz

Show more details

Last Notes

2026-07-15 15:41:53 CEST

I love monolithic kernels
https://www.phoronix.com/news/Apple-Magic-Bluetooth-Battery
https://files.mastodon.social/media_attachments/files/116/924/277/550/121/047/original/225f2aaab181be3a.png

2026-07-09 03:06:43 CEST

> take care of your body, your future doctor is using ChatGPT to pass med school

2026-06-16 18:27:01 CEST

- reply

it's all equity, and $60b is the price assuming that SpaceX is actually worth $2.8t which it very much is not

2026-06-10 01:21:03 CEST

macOS 27 has two separate dyld shared caches on the x86 side of things.
One "complete" one and one which is _much_ smaller and has some stubbed out frameworks as part of the "x86Support" component.
game-test-tool sets the x86Support=1 boot argument to flip to that reduced env.
Ecosystem.framework does a bit of the work needed behind the scenes to flag which scenarios are supported.

2026-06-08 23:39:57 CEST

> Apps built with the latest SDK must adopt the scene-based life cycle or they fail to launch. For migration guidance, see Transitioning to the UIKit scene-based life cycle. (141837548)

2026-05-10 22:09:50 CEST

> The SPSel feature was designed to allow software running at privilege levels (EL1, EL2, or EL3), such as an OS kernel, to use the stack of a user application, thus avoiding kernel stack overflow. However, this usage is not present in Linux kernel.
???
https://developer.arm.com/documentation/ka005621/latest

2026-04-06 17:38:36 CEST

wait but
https://files.mastodon.social/media_attachments/files/116/358/503/970/586/865/original/3588d73daf4214e5.png

2026-03-22 17:02:30 CET

Google Search using AI to rewrite headlines
https://www.theverge.com/tech/896490/google-replace-news-headlines-in-search-canary-coal-mine-experiment

2026-03-17 12:58:46 CET

advertising bounce buffers :D
https://community.intel.com/t5/Blogs/Tech-Innovation/Artificial-Intelligence-AI/Confidential-AI-with-GPU-Acceleration-Bounce-Buffers-Offer-a/post/1740417

2026-03-11 17:40:27 CET

RE: https://ec.social-network.europa.eu/@EUCommission/116211334179600320
this is a meme and not in the good meaning of the word
even less than GAIA-X, just rofl :/
#note1zlr…azy5

2025-12-24 19:04:05 CET

huh Xqci RISC-V extensions by Qualcomm for microcontrollers: https://github.com/quic/riscv-unified-db/releases/tag/Xqci-0.13.0
https://files.mastodon.social/media_attachments/files/115/775/858/305/063/818/original/4097eb3dce883aad.png

2025-12-09 22:47:17 CET

Tenstorrent learnt the lesson (?) that December 8/9 are the best part of the year to do layoffs evidently

2025-12-05 00:24:33 CET

Huh. Splitting the GPU into two
https://files.mastodon.social/media_attachments/files/115/663/874/651/312/070/original/661fbeaf98222ad1.png

2025-11-30 18:49:38 CET

Luxury GPU features that (at least some) AI accelerator providers tried to get rid of:
- having DRAM controllers
- if it has the above, having a way to do load/store to DRAM
- having coherency
- having MMUs
- having page faults be recoverable (well, AMD GPUs too...)

2025-11-30 18:02:26 CET

You can really call anything RISC-V
(This is what Tenstorrent ships)
https://files.mastodon.social/media_attachments/files/115/639/722/684/516/137/original/c4073af165e5350c.png

2025-11-02 18:50:09 CET

- reply

they even intriguingly had/have LiDAR on the iPad Pro camera bump

2025-10-20 10:17:21 CEST

since when did Word become like this
https://files.mastodon.social/media_attachments/files/115/405/503/234/829/568/original/a1e19d30070d7442.png

2025-10-14 05:17:49 CEST

- reply

disaggregated prefill...

2025-10-14 04:46:12 CEST

- reply

and the Mac Studio has twice the memory bandwidth to ~3 times the mem bw (Max/Ultra)
but a lot less flops...

2025-10-03 23:31:48 CEST

Hmmm...
And
> Automatic protection can ensure that your app is only made available on Google Play to users on devices that pass device integrity checks and, for apps targeting a minimum API level of 28 or higher, can add device integrity checks and an hardware-backed encryption layer to your app’s runtime anti-tamper protection.
Slowly but surely locking down the whole ecosystem with remote attestation across the board
https://files.mastodon.social/media_attachments/files/115/312/364/245/247/638/original/f5d2f6bfd87604ac.png

2025-10-01 23:06:19 CEST

Linus Torvalds' bingo on what to not ship in hardware:
https://lore.kernel.org/lkml/CAHk-%3DwgYcOiFvsJzFb%[email protected]/t/#mf1b6116ecfd963353312d26f61b8b49ee409fd1f

2025-09-30 00:03:43 CEST

BMC security is aaaa

2025-09-28 13:24:16 CEST

a mix of cocaine and ChatGPT to raise your "productivity"

2025-09-21 10:17:09 CEST

Something quite interesting: https://fil-c.org/
tldr: CHERI guarantees (purecap only) on existing hardware
The catch is as always with purecap models is that you have a whole new ABI.
Interestingly pointers stay 8 bytes here, with the capability being invisible to the running program (but instead maintained invisibly by the runtime).

2025-09-20 01:02:53 CEST

unbound levels of stupidity

2025-09-13 10:17:04 CEST

FEAT_MTE4 notably decouples MTE and TBI.
This allows to have a PAC 4 bits bigger than the prior MTE configuration when using the new MTE with TBI off configuration.
That's especially relevant when using big virtual addresses as that can make the number of PAC bits quite small.
In the typical 48-bit VA configuration on desktop OSes the MTE without TBI configuration allows to have an 11 bits wide PAC instead of 7 bits.

2025-08-17 18:28:04 CEST

staying a whole life in a do-nothing office job to then also do nothing useful and end up having had a negative impact on the world. having taken far more than value given...

2025-08-08 22:34:26 CEST

is it worth it to submit HVF _x86_ nested virt patches for qemu
or too late for it to matter as x86_64 macOS is on the way out?

2025-06-29 23:18:34 CEST

- reply

followed by attempting to replace (some of) those colleagues with an LLM apparently.....

2025-06-29 23:00:31 CEST

Regular layoffs and 10% of employees in performance improvement programs yearly to increase morale

2025-06-25 19:30:19 CEST

lmao @ the LLM experience
https://files.mastodon.social/media_attachments/files/114/745/185/518/215/851/original/c04a36190b9f4788.png

2025-05-13 20:11:41 CEST

doing a job interview in this state was probably not wise... and I inevitably screwed it up tbh
expecting a rejection email any day now lmao

2025-05-08 17:25:33 CEST

- reply

but from where? can always have a robot that clicks keyboard buttons...

2025-04-01 18:32:36 CEST

Intel marketing
https://files.mastodon.social/media_attachments/files/114/263/664/231/574/443/original/4da6f7039fd3fea8.png

2025-04-01 02:30:21 CEST

so most local Apple Intelligence models have 3b parameters

2025-03-12 01:37:08 CET

10x the speed and half the memory usage by migrating the TypeScript compiler from TS to Go isn't exactly a ringing TypeScript endorsement.
https://devblogs.microsoft.com/typescript/typescript-native-port/

2025-03-05 20:25:47 CET

- reply

the base parts don't expose vm.ecc* at all so wonder what the difference is between tiers

2025-03-05 19:57:56 CET

- reply

% sysctl -a | grep ecc
vm.vm_ecc_corrected_count: 0
vm.vm_ecc_uncorrected_count: 0
vm.vm_ecc_db_pages: 0
vm.vm_ecc_zero_pages: 0
vm.vm_ecc_panic_pages: 0
vm.vm_ecc_max_db_pages: 8388
on an M4 Max ootb

2025-03-05 19:56:47 CET

- reply

don't apple Mx Pro/Max parts have ECC already

2025-03-03 14:59:23 CET

Adapt to removal of Windows Arm32 .NET debugging
> .NET support for Windows on Arm32 has ended. Debugging support for this platform will be removed from Visual Studio 2022 starting with the 17.14 update.
https://learn.microsoft.com/en-us/visualstudio/debugger/adapt-to-removal-of-windows-arm32-dotnet-debugging?view=vs-2022

2025-02-28 19:41:12 CET

May you live in interesting times...

2025-02-23 06:53:36 CET

wonder if anybody would care about a Linux binary compat layer for macOS (w/o a Linux VM) or if VMs are good enough for that nowadays

2025-02-17 19:02:49 CET

Broadcom Core Ultra 9 385K

2025-01-24 20:33:07 CET

- reply

Quite a lot of changes overall
https://files.mastodon.social/media_attachments/files/113/884/999/537/556/851/original/bcabf6c06a5d90a3.png

2025-01-24 20:31:12 CET

- reply

> This feature can be enabled by loading nvidia.ko with the
`NVreg_RegistryDwords=RMIntrLockingMode=1` kernel module parameter.

2025-01-24 20:30:24 CET

in NVIDIA R570 Linux changelog:
> * Added support for VRR on systems with multiple displays.
> * Added support for the systemd suspend-then-hibernate method of system
sleep. This feature requires systemd version 248 or newer.
> * Implemented a feature that allows low latency display interrupts to
be serviced even when the system is under heavy contention. This
is especially useful for reducing stutter when using virtual reality.
> This feature is experimental and disabled by default.

2025-01-08 00:32:36 CET

Nvidia CEO says company has plans for desktop chip designed with MediaTek
https://www.reuters.com/technology/nvidia-ceo-says-mediatek-will-be-able-sell-nvidias-desktop-cpus-2025-01-07/
> "We're going to make that a mainstream product," Huang said. "We'll support it with all the things that we do to support professional and high-quality software, and the PC (manufacturers) will make it available to end users."

2025-01-04 19:27:08 CET

btw (on arm64)
Hyper-V from Windows 11 version 22H2/23H2 works without (and does not use) VHE
From version 24H2 onwards VHE is mandatory. Those releases also have ARMv8.1-A atomics and RCpc from ARMv8.3-A as required.

2024-08-26 06:32:53 CEST

- reply

now that's funny - u-boot on X Elite should be funny too

2024-08-26 06:31:31 CEST

how often do you all rely on nested virtualisation? Is it to make things easier or is it an actual dependency for your workflow?
And is it for client/development machine usage or for some production use cases too?

2024-08-25 01:53:32 CEST

hot take: I think launching the original Snapdragon 7c on a Windows platform was a bad idea, especially with 4GB of RAM devices

2024-08-21 04:56:35 CEST

question: is an ecore clocked at 5GHz still an ecore

2024-08-19 21:36:47 CEST

- reply

if you had to choose between edk2 or RTKit what would you pick

2024-08-19 21:13:52 CEST

uefi as an operating system on cortex-m with preemptive scheduling

2024-08-11 17:39:47 CEST

so
is 50% return rates good
for snapdragon windows laptops

2024-08-10 18:46:37 CEST

Making RSA faster on Graviton2: https://www.amazon.science/blog/formal-verification-makes-rsa-faster-and-faster-to-deploy

2024-06-02 22:13:50 CEST

gathered from x.com:
nvidia actually exposes a GPU upload heap (mapped BAR to user-mode) in Vulkan on Ampere Altra in their drivers.
This caused app breakage for Tellusim as Ampere Altra forces mapping BARs as device mem, instead of normal NC, for things to work properly. Device mem doesn't allow unaligned accesses.
It's a quirk that Apple Silicon Macs also have, and I'm tempted to say that perhaps NV should just not expose a GPU upload heap on Altra. After all, D3D12 didn't need it for ages.