金のニワトリ (@gosrum)

RTX 5090 환경에서 Nemotron-Cascade-2-30B-A3B(Q4_K_M)의 llama-bench 결과를 공유했다. prefill은 약 13000 tps, decord는 약 285 tps로, 최신 GPU에서의 추론 성능을 보여주는 벤치마크 트윗이다.

https://x.com/gosrum/status/2035545070355792319

#nvidia #rtx5090 #llamabench #benchmark #llm

We’ve entered a paradox. Local hardware like the RTX 5090 and Apple M5 is making "Inference Sovereignty" a reality for every desk. Yet, the demand for industrial-scale "Token Factories" is exploding.

In our final installment of the NVIDIA GTC 2026 series, we break down:
The Recompute Tax, Jevons Paradox, Trickle-Down Inference

https://www.buysellram.com/blog/hybrid-inference-architecture-why-the-token-factory-scales-as-local-ai-explodes/

#AIInfrastructure #NVIDIA #GTC2026 #HybridAI #GPU #DataCenter #Inference #RTX5090 #AgenticAI #LocalAIInference #TokenFactory #OnPremiseAI #tech

GPU return scams cost sellers thousands as eBay struggles to adapt

#eBayScam #GPU #RTX5090 #OnlineShopping #AusNews #AIGenerated

https://thedailyperspective.org/article/2026-03-20-gpu-return-scams-cost-sellers-thousands-as-ebay-struggles-to-adapt-5816bc16

System76 unveils redesigned Thelio Mira Linux desktop with Ryzen 9000 CPUs, RTX 5090 graphics, and workstation-class expandability

https://fed.brid.gy/r/https://nerds.xyz/2026/03/thelio-mira-linux-desktop/