may save up for a new box

Discussion in 'Tech Heads' started by AgelessDrifter, Apr 13, 2019.

  1. AgelessDrifter

    AgelessDrifter TZT Neckbeard Lord

    Post Count:
    43,890
    Maybe not right away, but some time after I graduate I'll lose access to the GPU servers and clusters at my school and then I'll be looking for something with at least greater processing power than my 6 year old Surface Pro II (I actually just looked that up and can't really believe it isn't older than that)

    I know the shitcoin market ushered in a boom in GPU development--are there are cheap laptops with decent GPUs now?
     
  2. Agrul

    Agrul TZT Neckbeard Lord

    Post Count:
    45,673
    i bought this & am quite pleased w/ it

    https://smile.amazon.com/gp/product/B0713X45S1/ref=oh_aui_search_asin_title?ie=UTF8&psc=1

    it's not cheap rlly but you could probably get something w/ similar specs for like $200 less if you were to compromise on screen space. i generally look for i5 or later cpu; recognizably non-ancient nvidia gpu; 256GB+ SSD space; 8GB+ RAM; big 1080p+ screen; many consistently good reviews. maybe also look into the power supply if feeling like a responsible consumer

    if you're looking for power to train big ML models or something though you should just temporarily rent an AWS/azure/whatever the other one is cluster or VM(s) imo. your local machine or laptop should only need to be beefy enough to train small models while you test dev code
     
    Last edited: Apr 13, 2019
  3. Agrul

    Agrul TZT Neckbeard Lord

    Post Count:
    45,673
  4. AgelessDrifter

    AgelessDrifter TZT Neckbeard Lord

    Post Count:
    43,890
    Thanks, yeah I've been thinking about dropping a few bucks to fuck around on AWS. I dunno how much longer my interest in fucking around with tensorflow is going to hold out anyway so it wouldn't make sense to drop a forture on a beefy GPU for personal use. My current shit is just super duper slow even with toy examples and my harddrive is mysteriously always within epsilon of being completely full even though I've long ago deleted all music and videos and most pictures and datasets from school off of it

    That first link is just about at the upper bound of what I could see spending for just the right thing and does seem to check most of the right boxes
     
  5. AgelessDrifter

    AgelessDrifter TZT Neckbeard Lord

    Post Count:
    43,890
    I didn't see anything about it in the product description--I assume that first laptop is not touchscreen?
     
  6. Agrul

    Agrul TZT Neckbeard Lord

    Post Count:
    45,673
    definitely not touchscreen. aside from that, i think the biggest weaknesses of the 2 laptops i posted are A] not a ton of drive space (although the 1.25 TB on the 1st one isn't too bad) & B] the i5 CPUs are good enough for basic daily tasks but obviously will suffer some if you're doing any serious CPU intensive gaming or CG rendering or non-GPU ML model training. and they're from a lower tier of the i5 list

    this is your surface pro II? https://en.wikipedia.org/wiki/Surface_Pro_2 looks like it has an i5 series chip. i think the 1st laptop i linked has a CPU that's about 25% more performant for most tasks: https://cpu.userbenchmark.com/Compare/Intel-Core-i5-7200U-vs-Intel-Core-i5-4300U/m153577vsm4996

    surface pro II has integrated graphics chip though so that should be a lot worse. about 317% worse, apparently: https://gpu.userbenchmark.com/Compare/Nvidia-GTX-950M-vs-Intel-HD-4400-Desktop-115-GHz/m27713vsm8064 similar for the 2nd, cheaper laptop i posted: https://gpu.userbenchmark.com/Compare/Nvidia-MX150-0-vs-Intel-HD-4400-Desktop-115-GHz/m332822vsm8064

    re: AWS -- you can get a super beefy GPU node w/ tensorflow etc pre-installed for ~$5/hr that will kick the shit out of anything you're likely to buy. and paying AWS $5-30 to train a big-ish model keeps your personal use laptop or PC from becoming unusable in that period, + gives you valuable practice using a common cloud service. unless you're going to be training massive models every single day i think AWS/azure/etc's pricing models are hard to beat. you can also rent super small nodes for pennies per hour that will run tensorflow etc via cpu when you just want to test that you can ship your code to AWS through git or w/e cleanly

    to alleviate local storage concerns you could also store any "large" (like a few hundred GB, say) data sets in s3 or a similar service for a buck or two a month and just pull them out as needed. is literally ~2.5 cents per GB for frequently used storage, and 0.4 cents per GB for infrequently used stuff that you can wait a few hours to open (if you put it in glacier). shit's super cheap even without bothering to compress files
     
  7. Bailon

    Bailon TZT Neckbeard

    Post Count:
    11,950
  8. Vlaara

    Vlaara Maaruk the Mighty

    Post Count:
    22,147
    I take my previous comment about shitposting back, Bailon you have redeemed TZT.