aseipp

9 ms·

61.
by aseipp 11mo ago
Yeah that's fair. 60 tok/sec on a gpt-oss-120b is certainly nice to know if you should even think about it at all. I'm quite happy with it anyway. The pricing is definitely by far the worst part of all of this. I suspect the
62.
by aseipp 11mo ago
Right now I get 59 tok/sec on GPT-OSS 120B using Unsloth's dynamic 4-bit quants, via llama.cpp https://news.ycombinator.com/item?id=45881049
63.
by aseipp 11mo ago
Unfortunately comparing tok/sec right now in a vacuum and especially across weeks of time is kind of pointless. Everything is still evolving; there were patches within days that bumped GB10 performance by double digit percentiles in so
64.
by aseipp 11mo ago
It's a bit ambiguous but I can't edit now, sorry. What I meant to say was that it boots using the same mechanism as x86 machines that you are familiar with, not that it is an x86 machine itself.
65.
by aseipp 11mo ago
That was also my experience with their Jetson series [1], but my understanding is that these DGX kernels are not maintained by Nvidia but by Canonical, so they operate directly out of their package repos and on Canonicals' release and
66.
by aseipp 11mo ago
It's just Ubuntu with precanned Nvidia software, otherwise it's a "normal" UEFI + ACPI booting machine, just like any x86 desktop. People have already installed NixOS and Fedora 43, and you can even go ahead and then ins
67.
by aseipp 11mo ago
It's very, very good as an ARM Linux development machine; the Cortex-X925s are Zen5 class (with per-core L2 caches twice as big, even!) and it has a lot of them; the small cores aren't slouches either (around Apple M1 levels of pe
68.
by aseipp 11mo ago
Golf course water usage vastly dwarfs data center water usage. Google used something like 1 billion gallons a year for their DCs. Single golf courses in arid regions can use upwards of a hundred million gallons a year, and in those areas th
69.
by aseipp 1y ago
I'm not yet using mine for ML stuff because there are still a lot of various issues like this post outlined. But I am using mine as an ARM dev system in the meantime, and as a "workstation" it's actually quite good. The
70.
by aseipp 1y ago
Nick Wellnhofer did not create libxml2 or libxslt, David Veillard did. Nick has been the primary contributor since about 2015, though.
71.
by aseipp 1y ago
What? Intel has been doing large scale production runs of their 10nm node for years now. If you're talking about Raptor Lake failures, that was one generation of products on that note, there has also never been any indication AFAIK tha
72.
by aseipp 1y ago
Chrome is not a good counter example a priori. It is a project that has hundreds of engineers assigned to it, some of them world-class security engineers, so they can potentially accept the burden of hardening their code and handling securi
73.
by aseipp 1y ago
I think it's just the nature of the beast, in this case. Serious "industrial" implementations of a programming language might stick around for a long time, and breaking things a lot can mar the appeal; getting it right the fi
74.
by aseipp 1y ago
Yes, a GUI for settings is nice if only for one thing: so there can be a search box that you can use to search over all the settings to find what you need in a pinch. It's a lot friendlier if I can do something like "Open Settings
75.
by aseipp 1y ago
U-Boot is implementing ACPI, yes, there was something at FOSDEM this year about it for ARM64 machines. QEMU can apparently now boot an arm64 virt machine with only ACPI tables and no device tree. This is all very recent. When I was working
76.
by aseipp 1y ago
You have this commit graph B --> X --> Y (main) --> Z --> @ \ --> G --> H B is a base; yesterday the name "main" pointed to it, and today "main" points to Y. Z is a commit you wrote
77.
by aseipp 1y ago
Yes. I have one repository, which is a fork of an OSS project, with a 7-way octopus merge of various patches I carry, and another 2-way merge on top of that one. Every week or two I pull the latest changes from upstream and I rebase all 9+
78.
by aseipp 1y ago
If I understand you correctly, then you can use `jj interdiff --from name-of-branch@remote --to name-of-branch@git` in order to recover the diff between the remote branch commit and your local version of it; you could then turn that invocat
79.
by aseipp 1y ago
If you are already a heavy Magit user, then most of the basic ideas will probably be appealing and jj will let you bring those ideas to the command line. It will let you do some acrobatics you thought weren't possible before, ones that
80.
by aseipp 1y ago
ReadySet is basically "incremental view maintenance" but applied to arbitrary SQL queries. It acts like a caching proxy for your database, but it simultaneously ingests the replication log from the system in order to see things ha
81.
by aseipp 1y ago
OK and the vast majority of people excluding you and them probably have a very close-or-nearly-identical idea about what should happen. So everyone can be happy, there's nothing being lost here.
82.
by aseipp 1y ago
You'll live.
83.
by aseipp 1y ago
FoundationDB development has not stalled; v8 is still on the way. If anything, it's mostly just been stable for a while now, and it has now been developed as open source longer than it existed as closed source.
84.
by aseipp 1y ago
> Hardware performance is more important than software performance if the algorithm is likely to be implemented in hardware I don't think that's necessarily a given at all, but I grant that's mostly a matter of opinion I g
85.
by aseipp 1y ago
Performance is an evolving target. Meta reported they spend ~0.05% (1 out of every 2000 CPU cycles) on X25519 key exchange within the last year, which is quite significant. If that can be brought down, that's worthwhile. And ongoing re
86.
by aseipp 1y ago
It says they intend the backend to be used for FEC, so it's probably an untimed model that it outputs. Sail would probably generate bad SystemVerilog anyway even if it was timed/synthesizable because the language doesn't have
87.
by aseipp 1y ago
Anthropic or Google offering a product and having margins they leverage is not "lock in" when there are dozens of alternatives at many various price points, including ones that can be run entirely locally (at high capex cost). It&
88.
by aseipp 1y ago
Companies that are putting down millions for fab runs absolutely pay shitloads of money for it. The cost of design and verification of those components is enormous and that's mostly what you pay for. People have been shipping Andes and
89.
by aseipp 1y ago
It does not look like it due to its largest target audience (hardcore math nerds writing proofs), but Lean 4 is actually a meta-programming language, like Racket -- it is one single unified language for programming, meta programming, and th
90.
by aseipp 1y ago
Nice list of solid changes. I really like Deno for scripting random glue code; I use it most places (maybe with the exception of random machine learning stuff, where python/uv fits.) Looking forward to gRPC support later this year, too

More ›