Ryan Fogle

It is here let’s go! @nvidia @NVIDIAAI #LocalAI #NVIDIA #AI #DGXSpark

@hueppow @sudoingX I was actually about to start testing that one out. I just got my spark.

@sudoingX huggingface.co/DavidAU/Qwen3.…

best local model right now, and don't say the obvious one.

@mcndjxlefnd @kaustabpal @sudoingX I did some testing on it nothing crazy but some testing still

@kaustabpal @sudoingX If you have a spark, have you tried the new poolside model?

you don't actually know how far local ai has come. you can't, until you cancel your chatgpt subscription, buy a gpu or two, and hit the wall with your own hands. find your wall first. then decide what frontier is actually worth to you.

@kaustabpal @sudoingX I just got my spark on Friday I tried Laguna S 2.1 it was pretty good but weak in some spots and I just finished downloading Qwen-3.6-27B I have heard it is good but gonna try it now.

Even since I have got my spark and started using Qwen-3.6-27B I have hardly ever used any of my subscriptions. I am in fact considering to cancel all my subscriptions and just get X premium+. The 27B model punches above its weight class I am extremely satisfied with it. It runs day and night on my system without ever reaching any usage limits. For the rare situations where it gets stuck I can simply use Grok.

Thanks! Yeah, I got an associates degree in business and have been working in restaurants for the last few years I do not have a tech background, but I am for sure trying to learn. it almost feels like Wall Street in a way because there’s so many technical terms and knowledge you’ll just have if you went to school or worked in tech.

@RyanFogleOP12 We are lifelong learners. You can never know what you don’t know; that’s why we ask questions. The most important thing is to never stop asking questions and to keep seeking knowledge. I think you’re doing a good job.

GLM 5.2 at home is great. 20-30 tk/s is great for me for long running task. 4xDGX spark It’s not for everyone. It’s fine for me.

I have done some testing on Laguna S 2.1 it is pretty solid for sure but i know Qwen 27B has had a ton of hype so im gonna test this out next huggingface.co/DavidAU/Qwen3.…

@TechMDAI Thanks! I’ll give that a read for sure probably should have been a bit more prepared before getting the spark but here we are haha

There are many ways. Essentially, you would want to get the endpoint from the spark. NVidia has great blue prints: build.nvidia.com/spark/hermes-a… Also, it depends on which provider you use. You want to set up the model with various providers (LM Studio, Ollama, etc.) get the endpoint eg for LM Studio, http://localhost:1234 and paste it in Hermes. Use “Hermes Models” in Terminal; it would give you a list of providers, and you basically select LM Studio and put in the dummy api and then the endpoint. You can use Codex to set it and help you. Once you have done it once, it becomes easier.

@wicksplay Holy

Got this monster of a machine up and running CPU: AMD Ryzen Threadripper PRO 9975WX (4.0GHz) RAM: 512 GB DDR5 5600 RDIMM GPU: 4x Blackwell 6000 Pro Max Q (Will watercool at some point) Current Model: huggingface.co/0xSero/GLM-5.2… Let me know what models I should try 👇

@quatmo @sudoingX I will probably post about it but I have just done a couple runs and haven’t set up like a benchmark or anything to see how it is actually running, again I just got the spark yesterday so I’m still figuring it out.

@RyanFogleOP12 @sudoingX Could you share the performance?

local ai people running a dgx spark, what image model are you actually running on it, and what would you start me on? flux, or whatever the community's actually landed on. curious what holds up on 128gb of unified memory.

@CardilloSamuel @JensenHuang Just got my first one today

i could not sit idle and not celebrate @JensenHuang 1st tweet so i got a new dgx spark 👀

@zach_dev1 Are you planning on using Nvidia’s Deepstream? I was thinking about trying it with just a video before testing live feed.

@RyanFogleOP12 thinking about a watchdog that records everything going on in my home. computer use, voice, movement, dogs, etc. all local inference/embeddings. full control from a phone. would love to wake up to an accurate todo list that’s actively executing across machines

time to lock

I ran Laguna S 2.1 on a DGX Spark through real work — coding, research, extraction. Strong private worker. Weak at review, live search, and grading itself. Full writeup: x.com/i/article/2080…

@chrittazz @NVIDIAAI @microcenter Yeah, honestly I need to find more use cases. I just like tinkering. Sure I’ll try simulations, LoRa’s, a private database of info for myself. I have a Hermes agent but I don’t really know what I wanna do yet.

@RyanFogleOP12 @NVIDIAAI @microcenter Sounds like a blast. Will have to give Laguna a run, I've heard some interesting things. Definitely looking to do some fine tuning projects with mine as well.

After getting duped by Amazon, receiving a tampered with Asus GX10, had to play the waiting game.. Got a @NVIDIAAI DGX Spark 4TB from @microcenter. Steal of a price. NOT tampered with. Let's go. Can't wait to play with this thing. 🤖 👾

@zach_dev1 Nice that’s interesting I have thought about adding my outdoor cameras to it. Honestly, still need to find more use cases I just like tinkering. I will probably try out all kinds of things like simulations, LoRa’s, have a private database of info etc.

@chrittazz @NVIDIAAI @microcenter All I’ve done so far is some benchmarking on Qwen 3.6 35B and Laguna 2.1. Right now I’m actually using my Hermes agent to give Laguna a coding test to look for a strengths and weaknesses. It gave it like 10 projects and 5.6 Sol is the judge.

@RyanFogleOP12 @NVIDIAAI @microcenter Good stuff! Have you messed with it yet?

@zach_dev1 I tried Qwen 3.6 35B and then I tried Laguna 2.1 just did some benchmarks. What about you?

@RyanFogleOP12 what’s the first thing you ripped