wick

Saint George and the Dragon

@zach_dev1 Everything

@wicksplay but what are you building

4x DGX Spark. Another 4 in the box for right now Got some heatsinks off of Amazon that fit very nicely on the top + sides to help with heat

i still don't know what the ideal soldering temperature is. i just picked 330C the day i started my journey and never changed it

@TechMDAI Godzilla is a good name for it lol

@wicksplay Woozers. That’s Godzilla

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 👇

My first interview with @sama, Co-Founder of @OpenAI. 0:04 How to start a startup 3:30 Trusting exponentials 4:57 Operating in chaotic environments 6:12 Learning to enjoy painful experiences 8:03 Creating abundant intelligence 11:15 Keeping core suppliers on OpenAI’s timelines 12:10 Invention of the joint-stock company 15:30 The best CEOs aren’t sociopaths 16:46 We are in the singularity 18:09 AI authoritarianism vs liberty 19:24 Texting 300-400 people a day 20:32 Having a small number of deep beliefs about the future 21:41 Critical path 22:25 Thinking about what’s next 23:38 Getting on planes in marginal situations 28:16 Buying lots of compute 30:51 Ambition 34:00 Google shouldn’t have let OpenAI survive 37:54 Having his life shot through a cannon after the launch of ChatGPT 41:20 The growth of Codex 42:02 The Death Star tweet 44:46 Status games and desire to be useful 47:57 Not being ambitious enough on compute investments 50:26 Execution 51:45 Ask for what you want 54:11 First few weeks of OpenAI 55:15 Shutting down Sora to focus on Codex 57:44 Designing beautiful products 59:01 Getting addicted to TikTok 1:01:03 Inventing a new device 1:02:53 Try to get better at your strengths 1:05:57 Masa is an n of 1 1:07:11 Real trends vs fake trends

@jonathan_conn Posted on my localmaxxing

@wicksplay Hell of a setup, congrats, can't wait to see what you do with it Would love to see your GLM 5.2 tps The spark community gets about 20-35tps

@gospaceport Shouldnt matter much because of the rubber base. Rubber is terrible at conducting heat

@wicksplay They may work better on their side so heat transfer from bottom to top unit is reduced also but yeah this looks good. Would have been cool if the surface of the DGX case was a mssive heatsink fin pattern OOB

@kkiran I found these heatsinks + a big fan nearby does the trick

I will add these to the list. I’m trying a bunch of things including a window AC! Garage vs bedroom and bedroom won given that central AC was not efficient and had to use Midea window AC. Works surprisingly well. A fan with raspberry pi driven switch during higher ambient temps during day is what I am thinking next. Glad the Sparks are quiet!

Took Thariq's context engineering rules and applied them to my local fleet last night Cut my biggest skill from 1211 to 385 lines Same capability, just trimming the fat. Took the handcuffs off of my supercomputer Copy paste the prompt screenshot into your ocr capable terminal

Married to the game 🥽🏌️

Google AI is largely in the same position that the United States was on the onset of the Second World War. The Americans had significantly more resources than anyone else but were mismanaging them to the point of complete dissonance. Once attacked, everyone rallied around the flag, giving rise to the greatest war machine in human history. Google's situation for a while has been the same Roosevelt faced in 1939: Lack of motivation, lack of structure, and lack of direction. This also just so happens to be the core of the reason that you should not write them off.

Very happy to support this on behalf of Google. We have long benefited from open source, are big contributors to open source and in fact have consistently made open weights models with Gemma available from @GoogleDeepMind @demishassabis . Onwards!

When multi-agent systems start acting like senior developers:

GLM 5.2 NVFP4 REAP 504B @ 87.1 tok/s output and 1559.9 tok/s prefill

@TechMDAI Works really well too with a fan infront of them

@wicksplay That’s super cool

DGX Spark heat issue solved ✅

My router: One queue, seven lanes behind it. brain (GLM 5.2 / Blackwell 6000 machine) is serialized, one active request, four queued, 429 if > 4. keeps shared traffic from stalling The setup: Brain 1M ctx the deep work, serialized Coder 128K bounded implementation Coder-xl 256K heavy context worker Vision 128K+img multimodal Scholar 256K critic Triage 32K fast classifier Embed - non-chat Rerank 4K non-chat Media box is separate, not behind the router Whole thing self hosted Stop paying for apis. Own it yourself

One 504B architect and seven specialist AI lanes Pi decides whether each task needs deep reasoning, fast coding, vision, long context or parallel workers then routes it to the right machine Live routing diagram and latency numbers next ✅