Sudo su
@sudoingX

GPU/local LLM. more RAM and OSS... everywhere

 Bangkok, Thailand
Based in Thailand
Joined August 2022
1,176 Photos and videos
Article

how i used x premium+ to go from 625 to 30k followers in just 4 months

i told people to go into debt for a 38 dollar subscription and i meant it. four months ago i had 625 followers. today it's 30,629. zero dollars from x to 5,923 in payouts and another 543 from

38
14
354
134,870
Article

your AI writes emails. i taught mine to find planets in real telescope data.

you've got the most capable machine ever built within arm's reach right now, and if you're anything like me, this week you used it to reword an email, summarize something you'll never reread, and

4
9
75
16,111
Article

so i tested wingman from emergent labs on my actual workflow and here is my first impression

i tweeted about this yesterday as the most underrated tool of 2026. now it is live and i can finally talk about what i have been using. wingman is a personal assistant agent inside the emergent

1
1
29
8,211

120B Parameters on $70K Hardware Lost to 27B on a $900 GPU. Here's the Data.

i tested two $70K datacenter flagships head to head then compared the same test on a $900 consumer GPU. the results broke how i pick models for autonomous coding. everything is reproducible. the

33
47
590
53,273
Article

12GB of VRAM runs more intelligence than you think in 2026.

i know because i tested it. one RTX 3060 with 12 gigs of VRAM. one 9 billion parameter model. zero handwritten code. the model wrote a full space shooter across 13 files, 3,263 lines, from a single

63
189
1,601
275,979
Article

Qwen3.5-35B-A3B: All You Need is a Single RTX 3090. 112 tok/s at Full 262K Context.

then the community showed up. 5 days. 15 GPUs. 176 tok/s peak. Qwen dropped a 35 billion parameter model that only activates 3 billion per token. Gated Delta Networks hybrid. 256 experts, 8 routed per

43
114
854
290,833
Article

i replaced anthropic's API with 2x 3090s. claude code on a local 80B Qwen model. every config.

i wanted to know if the local inference stack was ready for real agent work. not chat. not completion. a full autonomous coding agent that reads files, writes code, runs commands, debugs its own

25
41
541
56,033
Article

i benchmarked K2.5 (1T params) on 1x to 8x 3090s. every result, every crash, every dollar.

a single RTX 3090 can run a 1 trillion parameter model. a GPU you can find used for $800. the same card people game on just ran the largest open-weight model ever released. i wanted to know how far

25
29
372
107,255
Load more