I made a Roblox game called voltline on a 2017 iMac 5k with Claude and codex and made like $400 today somehow
2026-10-03 04:27:55
118
ADTCoach :
Local models are soooooooo slow
2026-10-02 22:56:34
68
Ben :
Just for fun or are you losing money
2026-10-03 05:32:37
3
Blueforcer :
im running qwen 3.8 27B with 300t/s on my 5090
2026-10-05 09:49:51
7
Muzamel :
how do you afford it
2026-10-06 12:34:58
3
user27273646872 :
Waiting end of oct for the 512gb
2026-10-02 20:44:30
16
Brandon :
What key board is that homie?
2026-10-05 17:59:01
0
Ty :
Tensorfold inference engine!! Check it out on GitHub. There’s a few recipes that will get you over 100tok/s. You can try GLM 5.3 Flash and Qwen3.8 Flash those are the two best to run on it.
2026-10-02 22:32:39
13
Romero :
You linking them with a thunderbolt cable?
2026-10-02 21:01:16
2
unbowed :
building a local model when claude code is $200/month just doesn’t sound logical. until subsidies are gone hosting your AI makes so much more sense. I rather run Haiku or Sonnet than a local model. It’s practically free.
2026-10-08 00:33:18
1
Jorgilin :
Looks like you’re having fun
2026-10-04 19:38:40
3
Sean :
Using unified memory is super slow vs VRAM HBM…wonder he got 21 t/s…use vLLM server as high throughput to run over 256 clients…