unbug

Local LLM V100 96G & enjoy Omarchy; Tday ; CODELF (Github star 14k); MIHTool (Mentioned in Google I/O'13)

[unbug @unbug]

We just hit 300 members today—woke up to 100+ new people joining overnight. Come join these top DIY hackers using extreme optimization to make old hardware run Qwen3.8-Flash and GLM5.3-Flash. 🚀https://t.co/5W0ilKD1QY

[unbug @unbug]

Just a bit of insight: people are eager to build their own rigs for local LLMs, the same way we were obsessed with building gaming PCs 30 years ago. https://t.co/5W0ilKD1QY

[unbug @unbug]

Strata+DeepSeek🔥

[Michael Helcig @mhelcig]

Run a 763B-parameter model locally. DeepSeek-V4.1-Flash-1017 on a 12GB GPU.

Frontier quality with GSQ-RCO quants, directly supported in the Strata inference engine.

Further info below. Much more to come. Follow for updates.

[unbug @unbug]

My new power dashboard? I love it. 📊

[unbug @unbug]

@Bitcopath I believe you

[unbug @unbug]

China actually has plenty of homegrown inference GPUs, but prices haven’t dropped as expected—and they’re still not great for local LLMs. Literally one big reason: no CUDA. Without the ecosystem—PyTorch, kernels, quants, llama.cpp/Strata support—local LLM tinkering gets way harder.

[unbug @unbug]

@pheno7723 4x 40~55w

[unbug @unbug]

You guys keep asking about the power cost of my 4×V100. I asked Strata + Qwen3.8-Flash to build me a dashboard—here’s the daily, weekly, and monthly bill. Don’t be surprised 😎

[unbug @unbug]

@666Sebo it in RAM only

[unbug @unbug]

Quick tip for the new Strata build: don’t forget --conversation-cache-mib 0 to disable conversation caching. You’ll lose a little prefill speed—just a heads-up.

[unbug @unbug]

@SheppardBuilds that's the problem

[unbug @unbug]

@cutoffurmind it is, that's why NVIDIA is higher price than AMD, right?

[unbug @unbug]

@lucasteske 2 pair of NVLINK

[unbug @unbug]

@mesouug agree

[unbug @unbug]

We’re hitting 400 members this week! Join us and let’s rock local LLMs together. 🚀

[unbug @unbug]

We just hit 300 members today—woke up to 100+ new people joining overnight. Come join these top DIY hackers using extreme optimization to make old hardware run Qwen3.8-Flash and GLM5.3-Flash. 🚀https://t.co/5W0ilKD1QY

[unbug @unbug]

@curliph indeed, only for big company

[unbug @unbug]

@mhelcig @joaosump I'll keep my eyes open the whole night😆

[unbug @unbug]

@joaosump @YourBizraeli what project?

[unbug @unbug]

@gopinath9629 No rush, get yourself a good sleep

[unbug @unbug]

Oops—Strata + DFlash still doesn’t support multi-GPU yet. Guess I’ll wait for the next release. 😅

[Niko Veit @coldniko]

An update before big release 👇

DFlash2 base support is now here!
+ A lot of other improvements
+ Stability improvements

[unbug @unbug]

@slap__tjips I actually turned off the system every night 🤣

[unbug @unbug]

@populartourist I don’t have Batman’s super power 😂

[unbug @unbug]

Like I said, new models come and new models come go—we all end up running only Qwen locally. DeepSeek V4.1 Flash is ridiculously slow on my V100 rig (256GB RAM + 96GB VRAM).