# 128GB ram, 2x rtx6000 blackwell, lots and lots of storage, but stuck in paralysis choosing a harness / model — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/3317b62d8ac8ac21

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-11-01 13:23 ingested 2026-11-02 02:41

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1xl2yt5/128gb_ram_2x_rtx6000_blackwell_lots_and_lots_of/)

## retained text



```
I have been accumulating parts for over 8 months to merge all of my existing hardware into one massive workstation.  I finally had the time to do it, and while I use the GPUs for algo trading from 9:30 to 4:15 p.m., I want to start to use local AI from 5pm on.  I'm at a point where I could see myself having a really fulfilling hobby reverse engineering, firmware hacking, and just experimenting with building Frankenstein gadgets from my tech graveyard.  There are many experiments I would like to perform on financial data as well that would just take too long for me to manually code and I am not interested in having these thoughts and experiments be in a large provider's training set.  I've been lurking on this sub for a long time and have my own little agent I use for some specific tasks using gemma3 but that was on my 3090 in C# + llama.cpp.  

So my long-winded question, what thick desktop client will give me the most seamless way to build autonomous agents, play with new models, and what version of qwen 3.8 flash should I be using with my hardware, assuming that is the right model.

Edit: I should mention I am open to anything but I just see q38f everywhere on this sub.
```



document fetched 2026-11-02 00:36 · canonical https://reddit.com/r/LocalLLaMA/comments/1xl2yt5/128gb_ram_2x_rtx6000_blackwell_lots_and_lots_of
