00:29:38 you can also just use zephyr as a light weight kernel 00:39:41 super computers are often used by scientists who actually have no idea how computers work lol 😂 . They want to drop in python code and expect it to automatically scale across a thousand nodes. That means a robust network stack, network storage, fault tolerance, automatic backups. You have so many nodes that at any one time some machines crash (whether it's due to softwa 00:39:41 bug or cosmic ray), the system needs to move work to another node seamlessly etc. It's a lot easier with an OS, not all about raw performance, at least on the CPU side. 00:40:40 Quarky93[m]: surely they don't run python on them, come on 00:40:46 maybe one of those networked numpy backends 00:40:57 but I can't imagine them spinning up a trillion instances of cpython 00:41:07 This is one of the reasons SmartNICs are hot right now (NVIDIA's DPU etc.), moving the load of orchestrating the datacenter to the NIC, freeing up the CPU 00:44:34 sure the backend is most likely a C++ library, but the easiest way to invoke that is cpython :) 00:46:47 oh, not to mention R code... 00:46:54 Right, but if that caused perf issues they'd change it in a heartbeat 00:47:18 just have the scientists use some wrapper library that sends over asm to execute on the cluster 00:48:11 how do you pass messages between nodes? 00:48:43 you'll have to bake in a TCP stack in the binary? 00:49:36 at some point you'll just be shipping an OS that only runs one program lol 01:05:15 that woud be the ideal, perf-wise ;) 01:15:57 The ideal solution is to treat the CPU as an accelerator with a lightweight ARM host 01:50:48 Was randomx chosen specifically due to difficulties running it in a browser (preventing coinhive et al)? 01:51:16 Or rather designed with that intention 06:30:39 pretty sure that was one of the goals. It uses a lot of compute so your fans would ramp up alerting you that something's going with the browser session. 09:16:49 osineofine: I don't believe that was a goal. The goal was to make it uneconomically viable to create ASICs that would outcompete consumer hardware in efficiency. 11:52:07 yanmaani, unfortunately, from my experience, Quarky93[m] is correct. you can insert really any scripting langauge in place of python... i've seen perl, R, python ... 11:53:04 sometimes, as a software user, i'll get lucky and they've stitched together some c++ with various scripts, but ... yeah 11:55:39 Quarky93[m]: no, you could have an OS with absolutely no security features too 11:55:44 like TempleOS 11:55:56 gingeropolous: surely the supercomputer folks are better tho 11:56:50 yeah, i guess a university cluster isn't technically a super computer ... 12:12:05 I've seen code being run on a national supercomputer and it's not pretty XD 12:53:23 found my power meter. M1 poower supply drawing 19.6W while xmrig is running 12:53:56 0.7W idle 12:59:54 120 h/s/watt, better than Ryzen 13:01:03 as expected I guess 13:01:06 yeah 13:01:15 5nm process, ARM arch is just cleaner 13:01:42 too bad, this guy won make a very effective space heater in the winter :P 13:01:49 won't* 13:04:11 so sad 13:07:49 Wonder how Neoverse will do 13:08:31 they puttin that bad boy into a multi CPU server format? 13:09:19 apple got out of the server biz a long time ago. would be interesting to see them re-enter 13:09:45 well there's rumoured to be a 32-core mac pro mini coming 14:33:50 more CI/CD abuse https://twitter.com/AaronErickson/status/1386684421617557509 15:42:24 <\x> why dont they just block stratum traffic 15:42:27 <\x> but hmmm 15:42:41 <\x> yeah pools offer tls/ssl now 15:42:43 <\x> so idk 16:14:21 packet sniffer? 16:26:44 <\x> even iptables will do it if its not encrypted 16:26:54 <\x> iptables string regex