majke

9 ms·

61.
by majke 1y ago
I didn’t realize Martin is blogging again! Hurray!
62.
by majke 2y ago
In the context of DNSSEC dns zones are very much enumerable. Cloudflare does amazing tricks to avoid this https://blog.cloudflare.com/black-lies/
63.
by majke 2y ago
The linked CV document is marked private.
64.
Derek Lowe: Revised and Extended: What's Happening Inside the NIH and NSF(science.org)
10 points by majke 2y ago|1 comments
65.
by majke 2y ago
> He was using his computer in a much, much different way than I’d seen someone use their computer before, and it made me feel old and alarmed by the number of new tools at our disposal and how HudZah intuitively knew how to tame them.
66.
by majke 2y ago
Correct. No mptcp on windows and android as of 2024
67.
by majke 2y ago
Author here. It was fun trying to actually use MPTCP. There aren't many resources on how to actually run it and what to expect. There are many unpolished aspects, like I was able to find (and report) a bug in MacOS (which has been fixe
68.
Mptcp: Revolutionizing connectivity, one path at a time(blog.cloudflare.com)
7 points by majke 2y ago|4 comments
69.
by majke 2y ago
The firmware would need to be certified. They mentioned that in the q&a couple of minutes ago. Apparently its a critical component.
70.
by majke 2y ago
What software stack you use for training?
71.
by majke 2y ago
Quite a strong cast of presenters back in Jan 2024 https://cs153.stanford.edu/syllabus.html
72.
by majke 2y ago
Ah! Now i remember. This was on the flooded tracks! So: tight train loop, and then flood tracks and then the game borked. Probably flooding came before train movement and it caused bad deref.
73.
by majke 2y ago
The original TTD was super stable though. I remember once I was able to crash it by building a very tight loop for a train, that it crashed with its own tail. This was the only time I saw the red font with debugging info.
74.
by majke 2y ago
> From a single core perspective, meaning from a single software thread, things are quite impressive for the chip, as it’s able to stress the memory fabric to up to 102GB/s. This is extremely impressive and outperforms any other des
75.
by majke 2y ago
I don't think single M1 cpu can do 100GB/s. This source says 68GB/s peak: https://www.anandtech.com/show/16252/mac-mini-apple-m1-teste...
76.
by majke 2y ago
This has puzzled me for a while. The cited system has 2x89.6 GB/s bandwidth. But a single CCD can do at most 64GB/s of sequential reads. Are claims like "Apple Silicon having 400GB/s" meaningless? I understand a typ
77.
by majke 2y ago
The best explanation of classic positional encoding I found in this blog https://medium.com/mantisnlp/positional-encodings-i-main-app...
78.
by majke 2y ago
Right. Not tlb, but more of memory subsystem stress. Typically changes to vm layout cant be done concurrently. Also, if the vm area is large, pages might be thrashing, requiring the kernel to do loads of work swapping things in and out and
79.
by majke 2y ago
> A legitimate problem here is the case where you need to read from or write to many locations at once. With the memory-mapped scheme as described, you can only issue as many concurrent page faults as you have threads in the program. &gt
80.
by majke 2y ago
So… eBPF is now BPF and old BPF is now cBPF or classic bpf.
81.
by majke 2y ago
Piotr has an amazing video on the inner workings of his model https://www.youtube.com/watch?v=bzyG4Wf1Nkc
82.
by majke 2y ago
I wonder if there is a token/watt metric. Afaiu cerebras uses plenty of power/cooling.
83.
by majke 2y ago
The secrecy of the inner tech is intentional.
84.
by majke 2y ago
It's a first time I hear about /dev/nvidia-uvm. Is there any documentation on how nvidia API works? Especially, how strong is the multi-tenancy story. Can two users use one GPU and expect reasonable security? Last time I chec
85.
Fast Kinodynamic Planning on the Constraint Manifold with Deep Neural Networks(sites.google.com)
2 points by majke 2y ago|0 comments
86.
by majke 2y ago
ESP32 uses the Xtensa instruction set. As far as I understand this is a custom instruction set, which is surprisingly popular in low power applications. Investigating this lead me to an interesting polish company called semihalf who were re
87.
by majke 2y ago
Pardon my ignorance, but how do matrix operations on quantized data work? Is hardware support needed? AFAIU int4 matrix multiplication is supported by cuda, but I'm not sure about other operations. The blog post mentioned fp6, and I do
88.
by majke 2y ago
https://hnrankings.info/41655954/ Looks like it was flagged, before being brought back by an admin. This happens quite often.
89.
by majke 2y ago
What error code specifically you get?
90.
by majke 2y ago
Dont forget the invention of open spaces! Open spaces is proven to be inferior to closed offices, but cost cutting and flexibility in layout is most important for HR.

More ›