JonChesterfield

6 ms·

31.
by JonChesterfield 7mo ago
"Fair" or "insane" ideas on price vary a lot between people. See also "competitive" salary on job posts. You might think $10 an hour is fair. Or you might think $1000 an hour is fair. If the developers you&#x27
32.
by JonChesterfield 7mo ago
Nope. A direct connection from residential British Telecom line is fine.
33.
by JonChesterfield 7mo ago
_Lots_ of traffic. It's going to end up being the global Internet backbone.
34.
by JonChesterfield 7mo ago
Something similar here but aggregated instead of failover, see openmptcprouter or diy equivalent.
35.
Show HN: Clangd for CUDA Device Code(docs.scale-lang.com)
8 points by JonChesterfield 7mo ago|0 comments
36.
by JonChesterfield 7mo ago
Bad idea. I would be very angry to discover I've bought this. Customer support are going to get shouted at and products returned.
37.
by JonChesterfield 7mo ago
An alternative is people install the software they choose to on the machines they're using. Optionally write a list of suggested programs down somewhere. In that world, there is no central IT team pushing changes to machines and arguin
38.
by JonChesterfield 7mo ago
Well, all the machines in the current outfit are Linux as far as I know. Services are self hosted. Seems to be fine, teams et al run adequately in a browser for talking to people on other stacks. Previous place had a corporate controlled wi
39.
by JonChesterfield 7mo ago
So gain access to a machine that can ask microsoft intune to eviscerate the company, ask it to do so, done. Bit of a shame all the machines had that installed really. Reminds me of crowdstrike.
40.
by JonChesterfield 7mo ago
This is a trap for engineers. If you don't worry about the returns, you won't get any. There are circumstances where that is fine. Be sure you're in one of them first.
41.
by JonChesterfield 7mo ago
A square foot is bigger than the area used by a person standing and people mostly weigh more than 40 pounds so that seems unlikely to be the design criteria for places people walk.
42.
by JonChesterfield 7mo ago
That would be unsound? Travel to Europe _before_ giving your assets away so you can stick the landing and work on building useful stuff there instead.
43.
by JonChesterfield 8mo ago
The game is deeper than that. Your model is probably about right for the compiler you're using. It shouldn't be - compilers can do better - but it's all a work in progress. Small scale stuff is you don't usually spill ar
44.
by JonChesterfield 8mo ago
I've seen claims that it's different for different architectural registers, e.g. _lots_ of backing store for rax, less for rbx. It's likely to be significant for the vector registers too which could plausibly have features li
45.
by JonChesterfield 8mo ago
Good post! Stuff I didn't know x64 has. Sadly doesn't answer the "how many registers are behind rax" question I was hoping for, I'd love to know how many outstanding writes one can have to the various architectural
46.
by JonChesterfield 8mo ago
It'll go much faster if you give each process a warp instead of a thread. That means each process has its own IP and set of vector registers, and when your editor takes a different branch to your browser, no cost.
47.
by JonChesterfield 8mo ago
Merely mislead by marketing. The x64 arch has 512bit registers and a hundred or so cores. The gpu arch has 1024bit registers and a few hundred SMs or CUs, being the thing equivalent to an x64 core. The software stacks running on them are ve
48.
by JonChesterfield 9mo ago
Relatively successful was slotting the cat5 jacket, cutting off two or three of the pairs, twisting/tying the remaining pair to the old wire, then sliding the jacket back over the join before wrapping in a conservative amount of electr
49.
by JonChesterfield 9mo ago
Pulling cables through walls is really easy for some construction styles and really difficult for others. Can involve taking up floorboards and drilling horizontally through beams, plumber style. Or cutting slots in masonry with angle grind
50.
by JonChesterfield 9mo ago
Lots of sympathy with this plight. Great to hear that someone has done the needful and rendered MoCA style modems over pairs of copper. I'm probably a customer for that. I'm currently running MoCA over spliced coax as part of the
51.
by JonChesterfield 9mo ago
I see a lot of references to `device_map="cuda:0"` but no cuda in the github repo, is the complete stack flash attention plus this python plus the weights file, or does one need vLLM running as well?
52.
by JonChesterfield 9mo ago
I think if you hit full path coverage in each of them independently and run all the cases through both and check they're consistent you're still done. Or branch coverage for the lesser version, the idea is still to generate intere
53.
by JonChesterfield 9mo ago
You run equivalent or equal calculations simultaneously on N computers and take majority wins, aircraft control or distributed filesystem style.
54.
by JonChesterfield 9mo ago
There's an Ubuntu box here to run things under cuda because two days trying to get cuda to run properly on Debian turned out to be the limit of my patience. For something that should be familiar it's intensely irritating as a dev
55.
by JonChesterfield 10mo ago
That one is called scalar evolution, llvm abbreviates it as SCEV. The implementation is relatively complicated.
56.
by JonChesterfield 10mo ago
I think this works. A subset of S3's API does look like a CRDT. Metadata can go in sqlite. Compiles to a static binary easily. I've spent a mostly pleasant day seeing whether I can reasonably use garage + rclone as a replacement f
57.
by JonChesterfield 10mo ago
I think this is one where the behaviour is obvious to you but not to people first running across the project. In particular, whether power loss could do any of: - you lose whatever writes to s3 haven't finished yet, if any - the local
58.
by JonChesterfield 10mo ago
Deleted comment pointed out that LICENSE.pdf is a screenshot from hashicorp. That's pretty weird, raised an issue for it.
59.
by JonChesterfield 10mo ago
That is a much stronger guarantee than your documentation currently claims. One site falling over and being rebuilt without loss is great. One site losing power, corrupting the local state, then propagating that corruption to the rest of th
60.
CUDA Tile Open Sourced(github.com)
201 points by JonChesterfield 10mo ago|106 comments

More ›