Running 30 concurrent Gemma 4 E4B calls locally: how to build it under $7k
Plane_Garbage · reddit · 2026-09-02
A Reddit user is looking for a local high-concurrency inference setup: roughly 30 simultaneous Gemma 4 E4B calls at 15k input / 3k output tokens each, ideally all finishing within 2 minutes, on a budget of $7,000 USD. Candidate options include multiple RTX 5060 Ti cards, a Mac Studio Max, DGX Spark, or several Intel B580s.
More from Infra
- SGLang Summit sponsors grow from 11 to 36 in one week; Lip-Bu Tan, Lilian Weng to speak — BanghuaZ · 2026-10-18
- Yacine jokes his datacenter KPI is gigawatt-hours: get paged if not burning enough power — yacineMTB · 2026-10-18
- 1.68x inference speedup drops, directly accelerating GRPO post-training — yacineMTB · 2026-10-18
- OpenRouter is only a fraction of the total addressable market, poster argues — sloppenheimer · 2026-10-18
- Developer builds change alerts for LLM prices and retirements across Bedrock, Azure and Vertex — jsandhu0780 · 2026-10-18
- 12.3M tokens in 27 minutes for $0.14: agent session with 99.6% cache hit — teortaxesTex · 2026-10-18