Landon

2x DGX Sparks | Local AI & LLMs

Sharing What Works

[Landon @landontgreen]

@jun_song @TechMDAI Speculative decoding improvements are making the low bandwidth not an issue

[Landon @landontgreen]

@0x3y3 @0xSero I don’t think i’ve ever realized this lol. i wonder…

[Landon @landontgreen]

@gnubyte_ @NVIDIAAI I’ve used the single sparks recipe since it originally came out and it was solid then. I’m sure it’s way better now as well

[Landon @landontgreen]

2nd DGX Spark secured! @NVIDIAAI

DeepSeek V4 Flash soon!

[Landon @landontgreen]

Going out of town for the weekend, hopefully come back to the weights :)

[DeepSeek @deepseek_ai]

DeepSeek-V4-Flash-Vision-Exp is now live on the DeepSeek API Platform! 🚀

🔹 This experimental multimodal model matches DeepSeek-V4-Flash on text capabilities—including agents, reasoning, and world knowledge.
🔹 On multimodal agent benchmarks, V4-Flash-Vision-Exp makes a major leap over V4-Flash, bringing multimodal agent performance close to Opus-4.8.

Try it with model='deepseek-v4-flash-vision-exp'. DeepSeek Harness 0.1.1 was released today with out-of-the-box support for the new model.

1/n

[Landon @landontgreen]

@Big_EG7 Good luck 👏

[Landon @landontgreen]

@Blau34 @NVIDIAAI Love sparkDash :)

[Landon @landontgreen]

@forumforartists @NVIDIAAI The Micro Center near me actually wouldn't do any shipping for it and I had to drive down to pick it up.

https://t.co/qlBAuauRPM

[Mia @MiaAI_lab]

Don't buy a DGX Spark from Amazon.

After receiving multiple reports of users receiving used or defective ASUS Ascent GX10 units from Amazon, I strongly recommend avoiding Amazon at all costs.

Some even received an empty box with no unit inside, and Amazon refused to issue a refund.

If you're planning to buy a DGX Spark, please purchase it from a reputable retailer instead.

[Landon @landontgreen]

@stevibe Thanks for sharing this test, I was really hoping to hear good news for things like this with this new release!

[Landon @landontgreen]

@forumforartists @NVIDIAAI Don't buy from amazon there is has been so many bad things happening to people buying their Sparks from there

[Landon @landontgreen]

@nohupBoris @NVIDIAAI Haha 😂

[Landon @landontgreen]

@Maritto_bk They haven't announced yet, but I'd imagine its openweights soon

[Landon @landontgreen]

Wow! Vision for DeepSeek V4 Flash!

It’s very cool to see them listening to the people, quickly as well 👏

[DeepSeek @deepseek_ai]

DeepSeek-V4-Flash-Vision-Exp is now live on the DeepSeek API Platform! 🚀

🔹 This experimental multimodal model matches DeepSeek-V4-Flash on text capabilities—including agents, reasoning, and world knowledge.
🔹 On multimodal agent benchmarks, V4-Flash-Vision-Exp makes a major leap over V4-Flash, bringing multimodal agent performance close to Opus-4.8.

Try it with model='deepseek-v4-flash-vision-exp'. DeepSeek Harness 0.1.1 was released today with out-of-the-box support for the new model.

1/n

[Landon @landontgreen]

@MrSage Congrats!

[Landon @landontgreen]

@TechMDAI Thank you 🙏

[Landon @landontgreen]

@Clawd_Max Thanks for sharing, I was thinking of setting mine to 372k or 500k so this is good to know

[Landon @landontgreen]

If you’re running DeepSeek V4 Flash on 2x DGX Sparks are you using the full 1m context window or less? And why?

[Landon @landontgreen]

@jun_song Definitely makes this the best model to run for majority of people.

[Landon @landontgreen]

@christopheduc @NVIDIAAI Not too bad!

[Landon @landontgreen]

@mr_r0b0t @NousResearch excited to hear how it goes!

[Landon @landontgreen]

@victortradesfx Makes sense, I usually have quite a few running parallel as well. Thanks for sharing!

[Landon @landontgreen]

@haarith How many agents do you usually run in parallel? That’s my only concern with the full 1m.