∀𝗥𝕀千

[∀𝗥𝕀千 @ayayalar]

@Blackwellboy I hope he recovers quickly and healthy 🙏

[∀𝗥𝕀千 @ayayalar]

I’ve started my journey with rtx5090s (2x). Recently, I’ve been running dgx spark (2x). I know this is going to be a bold statement but dgx spark is the future for Local AI. It’s clearly meant to be used 24/7. Minimal power usage, quite, and doesn’t produce much heat. Perfect setup.

[∀𝗥𝕀千 @ayayalar]

@cryptoafterdark @TechMDAI Yeah. I was just teasing.

[∀𝗥𝕀千 @ayayalar]

@MiaAI_lab We might need a complete guide on how to protect your data from Grok 😕

[∀𝗥𝕀千 @ayayalar]

@MiaAI_lab When you say “others”, you mean @elonmusk ? 😂

[∀𝗥𝕀千 @ayayalar]

@loktar00 @MiaAI_lab I’ve been using it for agentic coding on production enterprise apps. I’m blown away. It replaced Qwen 3.6 27b for me. I’m on the fence canceling my Claude subscription. If we do get something better than v4 flash, I’ll hit that cancel button for sure!

[∀𝗥𝕀千 @ayayalar]

@Tech2Wild Instead of 3090s, I have the 5090. Better tok/sec. But I still prefer dgx. Silent and better suited to run 24/7 for a home setup

[∀𝗥𝕀千 @ayayalar]

@MiaAI_lab If we know you enough from your previous recipes, you’d make it run. And then make all of us chase after 3rd spark. You’re trouble.

[∀𝗥𝕀千 @ayayalar]

@0xSero I stack them side by side. Better for heat management.

[∀𝗥𝕀千 @ayayalar]

@Vamzzz93 @TheAhmadOsman I use those same exact models on large scale enterprise repos. Including legacy code bases. Contrary to your experience, I have great success. Never ran into thinking loops either.

[∀𝗥𝕀千 @ayayalar]

@auroter I agree with @MiaAI_lab

[Mia @MiaAI_lab]

Anyone — no matter how big their account or how important they think they are, who claims the DGX Spark isn’t suitable for real work is either ignorant or a liar.

No in-between.

[∀𝗥𝕀千 @ayayalar]

@LottoLabs If I have a capable model, ~30 tps floats my boat

[∀𝗥𝕀千 @ayayalar]

@raven @0xSero Will try it for sure. Thanks for sharing it.

[∀𝗥𝕀千 @ayayalar]

@raven @0xSero Are you experiencing any stability issues?

[∀𝗥𝕀千 @ayayalar]

@0xSero I also used @MiaAI_lab’s recipe. Getting some gibberish at times. Trying to figure out what causing it. It seems to related to tool calling

[∀𝗥𝕀千 @ayayalar]

2x DGX Spark running using @MiaAI_lab https://t.co/NUAwmW04Um. I really didn’t expect such a great performance. Everything I get from my X feed, this was supposed to be SLOW. Well, it is the complete OPPOSITE!

I checked my opencode config to make sure I wasn’t pointing to rtx5090 endpoint by mistake!

Whoever suggests, dgx sparks don’t perform, they’re clueless. If you’re on the fence to get one/two of these magic boxes, don’t wait another second.

Setup Experience:

@MiaAI_lab made this idiot proof. Works flawlessly.

Only issue/changes I needed (not with repo instructions or scripts) but in my own setup

- I set WORK_HOST=user@spark so the setup script doesn’t use root@spark and use my logged in account instead.
- I added my user to docker user group so I don’t need to run the script with sudo.

Thanks for the amazing work @MiaAI_lab !

[∀𝗥𝕀千 @ayayalar]

@JamesMcPherson @MiaAI_lab Yup. Followed these exact steps. Also shutdown and unplugged. Started backup. Do you know if there is a way to repeat the perf test runs in dgx sync cluster setup?