NUM1NEX
@numinex.blog
Oh perfect! I am currently going through LL hell and was actually taking a look inside of ergot internals for usage of cordyceps :D
October 8, 2025 at 5:21 PM
Oh perfect! I am currently going through LL hell and was actually taking a look inside of ergot internals for usage of cordyceps :D
This video has some pretty interesting explanation of lifetimes. www.youtube.com/watch?v=gRAV...
but what is 'a lifetime?
YouTube video by leddoo
www.youtube.com
September 25, 2025 at 6:34 PM
This video has some pretty interesting explanation of lifetimes. www.youtube.com/watch?v=gRAV...
@nikomatsakis.com blog-post is the gold mine if you want the nerdiest stuff. Other than that I dunno, I've learned a lot about Rust by watching C++ conference talks (mostly around async). Maybe Niko can point to some other resources smallcultfollowing.com/babysteps/
baby steps
smallcultfollowing.com
September 25, 2025 at 6:30 PM
@nikomatsakis.com blog-post is the gold mine if you want the nerdiest stuff. Other than that I dunno, I've learned a lot about Rust by watching C++ conference talks (mostly around async). Maybe Niko can point to some other resources smallcultfollowing.com/babysteps/
If you combine it together with the reduction in open entry roles thanks to AInfluencers. I wonder - does nobody see it ? It's a disaster in the making.
If you factor in the planned obsolescence (both for hardware and software) it looks even more grim.
If you factor in the planned obsolescence (both for hardware and software) it looks even more grim.
September 17, 2025 at 10:00 AM
If you combine it together with the reduction in open entry roles thanks to AInfluencers. I wonder - does nobody see it ? It's a disaster in the making.
If you factor in the planned obsolescence (both for hardware and software) it looks even more grim.
If you factor in the planned obsolescence (both for hardware and software) it looks even more grim.
I was thinking about something similar recently. Given that tech, and SWE in particular has one of the highest burnout rates, while it being one of the best-paying jobs out there, which means that people working in it can afford earlier retirement or career switches.
September 17, 2025 at 10:00 AM
I was thinking about something similar recently. Given that tech, and SWE in particular has one of the highest burnout rates, while it being one of the best-paying jobs out there, which means that people working in it can afford earlier retirement or career switches.
But you need self-driving cars, because AI is the future. Don't look at the graveyard of startups that tried doing it.
September 5, 2025 at 4:34 PM
But you need self-driving cars, because AI is the future. Don't look at the graveyard of startups that tried doing it.
If max(N) concurrent requests is what you optimize for, maybe moving to HTTP3 /w QUIC could help, but last time I tried to tweak QUIC I just gave up (googlers have unique talent to create single most overcomplicated designs known to human-kind).
September 5, 2025 at 6:55 AM
If max(N) concurrent requests is what you optimize for, maybe moving to HTTP3 /w QUIC could help, but last time I tried to tweak QUIC I just gave up (googlers have unique talent to create single most overcomplicated designs known to human-kind).
even further (none of it is trivial work in Rust ecosystem, since it so heavily overindexed on tokio, that everything is bound to it).
September 5, 2025 at 6:55 AM
even further (none of it is trivial work in Rust ecosystem, since it so heavily overindexed on tokio, that everything is bound to it).
But other than that, most of those services are I/O bound, which means that that throwing more compute at it doesn't do much since NIC is the bottleneck here. Batching of syscalls could definitely help (io_uring) and given that the HTTP connections are ephemeral multi-shot ops would reduce syscalls
September 5, 2025 at 6:55 AM
But other than that, most of those services are I/O bound, which means that that throwing more compute at it doesn't do much since NIC is the bottleneck here. Batching of syscalls could definitely help (io_uring) and given that the HTTP connections are ephemeral multi-shot ops would reduce syscalls
So from quick googling, looks like this chip uses ARM v8 arch which has SIMD registers, so from the point of view of the CPU bound work, one of the low hanging fruits is prob using SIMD for (de)serialization. You can try building with -march=native flag if you already did not.
September 5, 2025 at 6:55 AM
So from quick googling, looks like this chip uses ARM v8 arch which has SIMD registers, so from the point of view of the CPU bound work, one of the low hanging fruits is prob using SIMD for (de)serialization. You can try building with -march=native flag if you already did not.
I thought the typo is the otherway around and the avg is 32ms lmao.
September 5, 2025 at 6:13 AM
I thought the typo is the otherway around and the avg is 32ms lmao.
How come avg is higher than p99 ?
September 5, 2025 at 6:09 AM
How come avg is higher than p99 ?
I recall tinygrad having a few folks writing GPU drivers in rust, so was my wild guess. Id be curious how they compare to candle, I believe its the most popular DL framework written in rust.
August 23, 2025 at 5:59 PM
I recall tinygrad having a few folks writing GPU drivers in rust, so was my wild guess. Id be curious how they compare to candle, I believe its the most popular DL framework written in rust.
After reading the ideology paragraph, it seems like they are heavily inspired by tinygrad. Wouldnt be suprised if its being developed by people that come from tinygrad, but they had enough of working with geohot lmao.
August 23, 2025 at 5:25 PM
After reading the ideology paragraph, it seems like they are heavily inspired by tinygrad. Wouldnt be suprised if its being developed by people that come from tinygrad, but they had enough of working with geohot lmao.
Wdym is this legit ? From quick look they seem to be doing things similar to what tinygrad does, especially those lazy tensors.
August 23, 2025 at 5:04 PM
Wdym is this legit ? From quick look they seem to be doing things similar to what tinygrad does, especially those lazy tensors.
Back in the days, we used to call those people script kiddies, today they go by as vibe coders. The more things change the more they stay the same.
August 19, 2025 at 3:22 PM
Back in the days, we used to call those people script kiddies, today they go by as vibe coders. The more things change the more they stay the same.
I've read the docs, the description makes perfect sense, but as you've mentioned the `std_tcp` example uses `tokio` tcp stack, so I would probably rename it to make it clearer.
August 18, 2025 at 5:13 PM
I've read the docs, the description makes perfect sense, but as you've mentioned the `std_tcp` example uses `tokio` tcp stack, so I would probably rename it to make it clearer.
I've seen it done in the `opentelemetry_sdk` crate, not sure how good is that trait tho:
docs.rs/opentelemetr...
docs.rs/opentelemetr...
Runtime in opentelemetry_sdk::runtime - Rust
A runtime is an abstraction of an async runtime like Tokio. It allows OpenTelemetry to work with any current and hopefully future runtime implementations.
docs.rs
August 18, 2025 at 5:02 PM
I've seen it done in the `opentelemetry_sdk` crate, not sure how good is that trait tho:
docs.rs/opentelemetr...
docs.rs/opentelemetr...
So what I am trying to get at is, that it could be worth exploring making the target interface registration be runtime agnostic. I know that IO with io_uring is it's own pack of worms due to the interface differences, but the task spawning I think could be abstracted.
August 18, 2025 at 5:02 PM
So what I am trying to get at is, that it could be worth exploring making the target interface registration be runtime agnostic. I know that IO with io_uring is it's own pack of worms due to the interface differences, but the task spawning I think could be abstracted.