Josh Meyer

This is the ultimate hangout for Homelab enthusiasts, AI tinkerers, and Vibe Coders. Think of this as your digital basement: a place to swap stories.

[Josh Meyer @HiredGunDevDen]

Somebody told me to get more power, so I did. Turns out a terabyte still doesn't fit in 128GB. But my $3,250 rig runs a 125B MoE at 77 tok/s flat across 2K-128K, 25% ahead of the vendor at depth. Arithmetic beats more power. What does your rig do at 128K? @gospaceport @coldniko

https://t.co/SNenlUcQYJ

[Josh Meyer @HiredGunDevDen]

https://t.co/ZuRQEApWx6

[Josh Meyer @HiredGunDevDen]

I should clarify. Email traffic and Kanban tickets.

Elon Operates as my mediator.
All context and memory at his level is organizational based sub domain agents, like mark (zuck), Steve (jobs), Rachel (from the office), etc, hold seperate context on the more specific domain.

Subs task the smaller 30b to assessment of my request, then report back, mark and friends make a recommendation with a decision queue to me, elon negotiates workload with options to hire it out to a contractor.

Example. task (Need a UI refactor).. Goes to Elon, he grills me for specifics in an email, I respond to email he recommends the team lead (mark, steve, etc), They plan it and hand the plan back to my elon.. Plans get refined until elon brings it to me.. then i stamp it with my recommendations and he builds tickets.

[Josh Meyer @HiredGunDevDen]

Refinement is happening!

[Josh Meyer @HiredGunDevDen]

@simcomsystems email traffic

[Josh Meyer @HiredGunDevDen]

Appreciate the added angle. Solar plus western renewable buildout is a real factor, and local electricity subsidies plus state capital look like bigger pieces of the cost picture for Chinese AI right now.

Belt and Road feels more tied to exporting influence and infrastructure than to DeepSeek’s domestic pricing, though do you see a tighter connection there?

[Josh Meyer @HiredGunDevDen]

Has anyone ever considered the fact that the way China is able to offer Deep Seek at such ridiculous prices is because they have a massive solar grid powering their AI? US night hours = China peak solar time. @spacexai @SpaceXAIMemphis @gospaceport @NetworkChuck @AdrianDittmann

[Josh Meyer @HiredGunDevDen]

Can't beat that price.... :-\

[0xSero @0xSero]

Don’t let them gaslight you, Chinese models are absolutely eating up frontier demand.

Deepseek is #1 and #2 lol

[Josh Meyer @HiredGunDevDen]

The local AI community thanks you for your service!

[Digital Spaceport @gospaceport]

I validated it, NVIDIA did leave a free 60% on the table for Ampere/Ada GPUs 😎

Quad 3090s + Qwen 3.8 27B BF16 in vLLM:
Unpatched: 1114–1259 tok/s prefill
Patched: 1769–2020 tok/s prefill

FREE ~60%+ on prompt processing from 2K all the way to 128K. About 15 mins shaved off every hour for long running agentic tasks in Qwen 3.8 27B and other full precision dense models.

Decode, the same. Power utilization, the same. 🔥 risk, the same.

Quantized models + Imgen/Vidgen, testing soon to get hard numbers

Windows unlocked as well. 50X0, probably 🔜

Kernel module, no driver patching or vbios crap.

Easiest free win of the year for Local Ai 🎉

Notes + bench script: https://t.co/5rBgTrtIf0

[Josh Meyer @HiredGunDevDen]

@akaNemsko As a person that has worked in the food industry most of my early years.... This is hot garbage... you don't like the tips situation, go find a different job.

[Josh Meyer @HiredGunDevDen]

@gospaceport @spacexai @NousResearch @coldniko @NetworkChuck @QwenDevs The key to useful AI is brining it into context for yourself. Structure like this allows you to use familiar tools, like email and job queues.

[Josh Meyer @HiredGunDevDen]

Anyone else got their AI/SI system structured like a company? @gospaceport @spacexai @NousResearch @coldniko @NetworkChuck @QwenDevs

[Josh Meyer @HiredGunDevDen]

@simcomsystems @gospaceport @spacexai @NousResearch @coldniko @NetworkChuck @QwenDevs I'm working on this layer next!

[Josh Meyer @HiredGunDevDen]

Very Detailed!

[Simply Complex Systems @simcomsystems]

@HiredGunDevDen @gospaceport @spacexai @NousResearch @coldniko @NetworkChuck @QwenDevs We use a traditional organizational structure.

[Josh Meyer @HiredGunDevDen]

I think my AI rig is ~3K after all the upgrades I can make for my little B550 (128G DDR4, Risen 9 5900XT, 3 5060TI cards, 500G NVMe)... I spend less than ~100 a year in frontier tokens and I am pushing ~ half a million lines of code shipped. https://t.co/PkQtQf4gJs

I ran this on 32Gb DDR4 and a Risen 5 5500 before the upgrades last month.

[Digital Spaceport @gospaceport]

These ppl really think its $10k for a rig that can do serious work? They must not have met Strata or Maya yet 🤔

[Josh Meyer @HiredGunDevDen]

@gospaceport I ran this on 32G DDR4, and a Risen 5 5500 before I upgraded (last month).

[Josh Meyer @HiredGunDevDen]

@gospaceport I think my AI rig is ~3K after all the upgrades I can make for my little B550 (128G DDR4, Risen 9 5900XT, 3 5060TI cards, 500G NVMe)... I spend less than ~100 a year in frontier tokens and I am pushing ~ half a million lines of code shipped.

https://t.co/5Efm4byBzK

[Josh Meyer @HiredGunDevDen]

@elonmusk Is it true the next Grok model will exceed 50 Trillion Parameters?

[Josh Meyer @HiredGunDevDen]

I must say Qwen3.8 125B as my attention head for my agent swarm is working out swell with Strata. It's freed up my other 2 cards for additional resources (like a vision model and an Image model). SO i basically have 32GB of VRAM idle at the moment while I run Qwen3.8 125B. and it uses only about ~30% of my 128GB DDR4..

[Josh Meyer @HiredGunDevDen]

Unless of course @coldniko has a list of features he wants built by the community.

[Josh Meyer @HiredGunDevDen]

Just a thought... Strata is forkable.. might be worth doing your own optimizations to contribute.

[Josh Meyer @HiredGunDevDen]

Just a thought... Strata is forkable.. might be worth doing your own optimizations to contribute.

[udofu @udofuchi]

qwen3.8 flash nextの最適化は辞めちゃったのかな？

[Josh Meyer @HiredGunDevDen]

I don't see a benefit in using anthropic.. it's a catch all for lazy devs and companies that don't understand the value of seasoned coding teams. They gladly take your money because it's a fortnite game to most who don't know ai. 99% of ai code is a half asked product no matter what agent you use.