Especially once you start selfhosting anything large.
Using an NPU / APU to accelerate something small isn't a terrible tradeoff, as soon as you deploy some real accelerators you're burning so much energy. Can't disaggregate, won't be able to keep batches full, will have idle time ...
oh dear, you've imported some of the horrors of GPU shared memory/LDS swizzling to CPU land.
yeah.
Who says store based can't be FHS compliant anyway?
gitlab.freedesktop.org/FHS/fhs-spec...
on crates.io, I want a way to commit irrevocably that all future versions will go through trusted publishing
Your best hope is that in the future amdgcn-spirv target usage shows up widely and the spirv path happens to work for 803.
If not that, it's possible to try to build with 803 as your only supported target for the entire package set but I'd expect big trouble.
my shader export compression is mightily broken and the effect is eerily reminiscent of trying to make out details in a night sky obscured by light pollution, except the sky now has intermittent complex transforms darting across its surface
GeekWire excerpt: "In addition, while all Ai2 programs for 2026 are fully funded, these people said, FFST is moving from providing Ai2 with overall annual funding to a proposal-based process, with future support expected to favor real-world applications of AI over building open-source foundation models. The shift helps explain the departures of researchers focused on model development." The words 'over building open-source foundation' are highlighted in green.
The OLMo fully open foundation models are probably not going to get new revisions.
Pretty sad as they were one of the few true open source options - mostly 'open weights' models don't share training recipes or data and are more like freeware.
definitely
define your own macro that also checks a crate feature to override to take the fallback path, run tests with and without that feature?
Need to finis this project GF - Needs to finish this project - "Oh yeah I haven't finished this" - "Haha woops it's not done yet" - "Right I forgot about this project"
swe confession: i keep trying to learn jj but always end up back in my idiosyncratic perma detached head git workflow
equality saturation is basically if starlight glimmer was datalog
Continuwuity v0.5.6 is hot off the press! Continuwuity is a chat server for communities that speaks Matrix - and our community has grown a lot recently. We've got security, performance & reliability improvements aross the board.
continuwuity.org
#Matrix #SelfHosted #OpenSource #FOSS #Privacy
Inject cats chart from the parent post with :3 crudely drawn in on the cat-ear shaped hump
github.com/triton-lang/...
Got gfx906 (Radeon VII, MI50/MI60) support upstreamed in triton this week!
Very happy to help prevent cards that are under a decade old and on the official support roadmap being ewaste due to poor software support.
just yesterday rocmPackages.hipblaslt.src failed on hydra because the source tree has 5GiB of yaml in it.
🫠
The implication that someone has to hate writing code to be happy to be able to automate some of it doesn't make much sense to me.
I got into programming as a child because I like making things run better or making new things, not because I intensely value typing in specific lines of code. No financial pressure, definitely for the love of the game.
i built an entire x86 CPU emulator in CSS (no javascript)
you can write programs in C, compile them to x86 machine code with GCC, and run them inside CSS
lyra.horse/x86css/
Embed image for a lunnova.dev blog post about ROCm 7.1.1 (AMD's GPU compute stack), styled entirely as a Neon Genesis Evangelion title card homage. The bold, squashed, serif typography, the red-and-white-on-black color scheme, the embossed red "m" in "ROCm" (mirroring the iconic red "N" in EVANGELION), and the tagline "You can (not) build." with "(not)" in contrasting replicate the Evangelion titling aesthetic. The background shows htop CPU %age bars in a state of overload and distress, fading towards the bottom of the screen. The date (left) and lunnova.dev (right) are displayed at the base of the image.
By popular (two people with strong opinions) demand, embed image updated.
You want to build the ROCm package set.
Please insert $5,000 to download more RAM.
lunnova.dev/articles/roc...
bsky.app/profile/norv...
I'm excited to potentially get a copy!
soatok.blog/2026/02/17/c... #Matrix #security #cryptography
acausal blog post sending somehow. too powerful.
