scrobble.life
#pimp

Gen AI on AMD GPU

A little while back I made upgrades to my main rig I got a better GPU an 64 GB of ram... I got lucky and it was before shit started getting out of hand price wise, I also managed to build my AI server in time but that is not what this post is about.

2026-11-11-19h06m06s_seed172426767_3d cartoon , in a car , mother with her kids in th_1.jpg

So my main rig has an AMD card a 9070 XT with 16 GB of Vram, a great card , very capable but early on it was not very useful when it came to AI related work loads, well things have gotten a bit better but it is still far from the ideal option for an AI centric build, even something like a mac with unified memory is not ideal, don't get me wrong you can do a bunch of stuff still but you may have to jump through hoops trying to make things work or accept that you can only run certain things, the most simple and easy setup is a machine running windows with a nice NVidia GPU and decent, you can run the most models and it is fairly easy with an app called pinokio

But like I said things have gotten a lot better then it was a while back for AMD cards as well as apple silicone, and so in this post I will make some stuff with this 16 GB card and share with you all as well as tell you more about it.

So to get started I download and install the pinokio app, then through pinokio I install wan2gp, then I just start it up and I am more or less good to go, I select image, video, music, 3d, or world model and I can get started ,now with 16GB and an AMD card I can't do it all but still can do a fair bit and the models that fit work fairly quickly.

Here I used the qwen image 2.1 model (turbo 8 step setting) this model is great at a wide range of image generation as well as image editing capabilities and with these open models you can use something called a LORA to give it some extra capabilities to do specific types of images and styles but maybe more on that later.

2026-11-11-19h02m25s_seed221519644_3d cartoon, red bee, with a cigar in his mouth  si_1.jpg

The images you see in this post took 7 seconds each, and the one right below was an image edit operation moding the one right above and these take 15 seconds a pop

2026-11-11-19h09m46s_seed446742512_make the cigar a lit weed  blunt, and make his eye_1.jpg

Can also make music and videos but we just stick to the pics in this one.

2816696b41bf42f7acc3e9a8781460fd.gif My inleo subs can have something like 1080 version with sounds included made! and much more for only 5 HBD!

I am a financial advisor ! If you can't take personal responsibility for you investing, I hate to brake it to ya, but you ain't nothing but a sucka, you supposed to lose your shit.

@bitcoinman's bits of wisdom

  • Invest only what you can afford to lose and be mindful of the potential risks.
  • Unless you can accept delivery stay away from futures and stick to physical.
  • Leverage is a great tool for losing money fast.
  • Numismatics for the lows stick to smart stacking baby.
  • Be cautious of new schemes in crypto, like ICOs, DeFi, and NFTs, do your research before investing.


You can find out more about the #silvergoldstackers community in this post here


Weekly Post Schedule

Day Post
Monday Art&3D
Tuesday
Wednesday
Thursday Lil Metal series
Friday
Saturday Lazy Saturday post
Sunday Table silver series

That's it for today's post thanks for stopping by your support is truly appreciated.

Comments · 5

  • @thebighigg(78)· 4d

    Good job on beating the price surge. It's out of control right now, any upgrades cost an arm and a leg!

    !BBH !PIZZA !PIMP

  • @pizzabot(60)· 4d

    PIZZA!

    $PIZZA slices delivered: thebighigg tipped bitcoinman @bulliontools(3/20) tipped @bitcoinman (x2)

    Learn more at https://hive.pizza.

  • @bulliontools(76)· 4d

    That's not bad at all. Sounds like for about $2000 the average person could pull off a decent local AI agent capable of doing most things Gemini and ChatGPT can do.

    !BBH !PIZZA !ALIVE !PIMP

  • @abir70(48)· 4d

    7 seconds per image on a 16GB AMD card is a lot better than I expected, honestly. I always assumed you needed Nvidia for anything local, so hearing Wan2GP and Qwen Image 2.1 run fine through Pinokio changes the maths on a build. Did the 8-step turbo setting hold up on bigger images too, or does it start to struggle past a certain size?

  • @offgridlife(72)· 4d

    Awesome

    via HiveSuite