Jump to content
Search In
  • More options...
Find results that contain...
Find results in...


  • Content Count

  • Joined

  • Last visited


This user doesn't have any awards


About igormp

  • Title

Contact Methods

Profile Information

  • Location
  • Gender
  • Interests
    Embedded systems, computer architecture, machine learning
  • Occupation
    Data Engineer


  • CPU
    Ryzen 3700x
  • Motherboard
    AsRock B550 Steel Legend
  • RAM
    2x32GB Corsair LPX 3200MHz
  • GPU
    Gigabyte 2060 Super Windforce
  • Case
    Sun Ultra
  • Storage
    WD Black SN750 500GB + XPG S11 Pro 512GB + random 2.5" SSDs and HDDs
  • PSU
    Corsair CX650m
  • Display(s)
    LG UK6520PSA, 43" 4k60
  • Cooling
    Deepcool Gammax 400 V2
  • Operating System
    Arch Linux

Recent Profile Visitors

2,141 profile views
  1. Idk exactly went wrong for sublime, or what went so right for VSCode. Around 2012~2013, you pretty much only had Sublime and TextMate as fancy text editors available, then atom came in 2014 (but was slow af and a resource hog), and only then came VSCode and became extremely famous. Sounds like the Windows CE or Blackberry history against modern mobile OSes.
  2. AFAIK, it's related to how you deal with memory/cache coherency between registers/cores/caches/memory. Tbf, you didn't specify any instructions Anyway, TIL and found those on the ARM docs: https://developer.arm.com/documentation/den0024/a/Memory-Ordering/Memory-types Apparently the equivalent to x86's TSO is called "Device-nGnRnE" Yeah, but it lacks the whole ecosystem whilst being closed source and kinda expensive. For something fast that can manage multi-GB files with ease and has a nice ecosystem I justo resort to vim nowadays (I do still use VSCode 90% of
  3. Oh, then ECC is not really needed. From what I've tested with my 64gb of ram, you usually see random bitflips once a month, so it's more relevant when you have large amounts of ram and run 24/7 (which increase the probabilities of bitlips a lot) IMO, you should try to go with 2x32gb sticks
  4. It's just just "regular arm instructions", those are only used to toggle the memory model. The TSO emulation that Apple implemented, and which can be toggled on or off during runtime, is what makes rosetta so fast. It could be implemented, sure, but then it would require some major design changes to cache coherency, buffers and overall memory management inside a core and goes beyond the ISA itself. VSCode is one of the most (if not THE most) performant electron programs out there, doesn't change the fact that electron is shit, but it's really easy to extend and that's why it has s
  5. Yeah, that's still hard to manage in a single module. Cost and magnetics is another issue, someone on the framework forum tried to do some sketches for a rj45 module and it neared ~$60 just in components.
  6. Yep, it's pretty easy on windows with hyper-v and GPU-P: https://forum.level1techs.com/t/2-gamers-1-gpu-with-hyper-v-gpu-p-gpu-partitioning-finally-made-possible-with-hyperv/172234
  7. I found this link: https://www.microcenter.com/product/613487/gskill-ripjaws-v-32gb-(2-x-16gb)-ddr4-3600-pc4-28800-cl16-dual-channel-desktop-memory-kit-f4-3600c16d-32gvkc-black If you're paying $170 for that memory kit, I'd try to see if it's doable to get a 2x32gb 3200mhz kit for not much more. Or, as others mentioned, go for ECC, just remember to look for UNBUFFERED/UNREGISTERED DIMMS, otherwise it won't work with your CPU.
  8. I guess it really depends on your usage, it may not be important for you, but portability sure is important for me. I'd rather have a 14" LG gram with nice battery life, decent performance for a web browser (I can always remote into a powerful PC if needed anyway) and portability for when I'm at my uni, coworking space or at someone else's house than a monstrous gamer laptop that looks weird af, is heavy, loud and (usually) has shit battery life due to all of the "high performance" components, but some do need high performance while being just portable enough, such as mobile workst
  9. Samsung could also be a player in this scenario, since their exynos CPUs are okaish, and they already do laptops (even some ARM-based ones).
  10. What's the size of your current MP510, and how full is it? It's already a top notch NVMe drive and you likely won't get anything noticeably better than it, your problem seems to be somewhere else (and will likely happen with other SSDs). That drive is both slower and has lower endurance compared to your current one. It's also dramless.
  11. There are CAD files available for the expansion cards (here), so a module with a DB9 connector and an ftdi or any other serial-usb converter of your liking should be a breeze. RJ45 is a little harder due to space constraints but shouldn't be impossible, and I bet someone will come with it anyway given how universal it is, maybe frame.work will even release one officially, who knows. They're a startup, the best idea is to focus on a single product (basically your MVP), see how well the whole business model goes, they expand (which is already on the roadmap as per the video).
  12. At the time of Zen's conception, AMD had an modern ARM project (K12, as mentioned in the OP) along with a hybrid ARM/x86 project (skybridge), both led by Jim Keller, but then Zen had a tremendous success (enogh to warrant total focus on it) and Jim Keller left the company, but I guess they could simply continue from where they left off. AFAIK, AMD has no other x86 µArch design available, even their low end Athlons and entry Ryzen CPUs were Zen-based. I'm not sure if skybridge is meant as a single platform for multiple ISAs (as in, same motherboard/chipset for both ARM o
  13. I have both a laptop with arm and linux (a chromebook, actually), an arm cloud instance (graviton), a hackish phone with linux that I run some silly stuff on, and many other ARM SBCs, the only problem I face is when I need some proprietary stuff that has been only compiled for x86, other than that it works perfectly fine for me, but then again, none of those are used as a desktop, nor I'd consider myself a "regular" desktop user by this forum's standards.
  14. CPU is pretty irrelevant for machine learning, what you want is a reasonable nvidia GPU, unless you're working with basic tree models or things that don't really have the "learning" part (such as knn). Idk about your regular ram usage, but I usually recommend going for 2x32gb, specially if you plan on handling somewhat large datasets. RAM speed is pretty much irrelevant for development, quantity is more valuable.
  15. You could go for backblaze's B2, or Wasabi, it isn't much more expensive over their personal plan.