@dhbrojas @__tinygrad__ was thinking of getting the p150a to tinker with myself - curious how it goes!
English
Jakub
26 posts
forgot to drink coffee this morning and now ive had an awful headache all damn day
English
@joannejang one of the only times i had to search it up was when using lto with nvcc and i can't say i had fun...
English
but i did learn that they call some kernels cute which is pretty cute 🎀💕
Joanne Jang@joannejang
think i flew too close to the sun and now i have to know about ptx it's so ood for me that google doesn't believe it either so it's telling me this
English
@jarredsumner @DanielLockyer I'm curious about the use of #[cold]. Are PGO/AutoFDO/Propeller/BOLT part of the public releases?
English
@DanielLockyer The sad thing is the wall clock time on this benchmark is very very slightly slower still, but the cpu time drops by 41%. So it uses significantly less CPU and in end to end time, it did not get faster.
English
@SaschaWillems2 @code Highly recommend giving @zeddotdev a try. Performance is great, and if you prefer there is a disable AI toggle.
English
Dear devs behind @code: Can you you please stop randomly adding AI related buttons, menu entries, panes etc. or at least make these opt-in by default. Clicking in a spot you're used to only to accidently open some "AI agent" window is pretty annoying...
English
@matt_dz It's so new that it's almost 7 years old, there's a legacy v1, stable v2 and recommended v3... Paper for v1 from 2019 btw microsoft.com/en-us/research…
English
mimalloc: A new, high-performance, scalable memory allocator for the modern era
microsoft.com/en-us/research…
May 13, 2026
by Daan Leijen
English
Jakub retweetledi
Attention @arxiv authors: Our Code of Conduct states that by signing your name as an author of a paper, each author takes full responsibility for all its contents, irrespective of how the contents were generated. 1/
English
@jarredsumner @theo Counting `unsafe` blocks seems kinda strange to me - sure, the quantity gives you some info, but there are valid and invalid ways of using unsafe. Also in other languages you get no info about the `unsafe` areas - so being able to pinpoint this behaviour is a huge win imo
English
@theo `cargo geiger` counts unsafe usage in the compiled output, which is more accurate than grepping source code because it includes dependencies.
I don't think counting `unsafe` usage is a very useful metric, but note that the count for uv is technically higher than Bun.
English
uv has 350k lines of Rust, and 73 "unsafe" calls.
The Bun Rust port is already 681k lines of Rust, and has over 13,000 "unsafe" calls.
English
@carllerche From what I understand: bun (now with anthropic) forked zig, made it faster with LLM, zig has strict ban on LLM, so official zig compiler is slower
English
The most surprising part of this to me is that the Rust version compiles faster than the (official) Zig version. I thought Zig prioritized fast compilation times. What am I missing?
Jarred Sumner@jarredsumner
99.8% of bun’s pre-existing test suite passes on Linux x64 glibc in the rust rewrite
English
i’d like to report a murder
bashbunni@sudobunni
@trashh_dev bet ur wife wishes she didn't know how to commit
English
4000 commits and 960,000 LOC. 3 compiler errors until maybe it compiles & panics immediately
English
1. /grill-with-docs
2. "Oh, I need to prototype some UI"
3. /handoff to /prototype
4. Create prototype, /handoff back to grilling session
5. /to-prd, /to-issues
6. npm run sandcastle
7. /improve-codebase-architecture
I love this shit
English
@notpuang for me, doing a lot of C and C++ and experiencing the different foot guns first hand helped. especially wrt mapping concepts across to rust and what it solves.
i often constantly ask myself "how can i offload this contract to the compiler" (lifetimes, ownership)
English
guys i think yall should support the podcast, it's not going well...
Theo - t3.gg@theo
I'm going to be so real with you guys. The podcast is performing 10x better than I expected...
English