🧰 AgentX

The items that name it 6

• SemiAnalysis (@SemiAnalysis_) on X
  thread2026-10-21

  Workload on which vLLM was measured at 200 P90 TPS.

• [AINews] DeepSeek DeepInfer: Open Source SOTA Inference running 155T toks/day/cluster making $22m/day at ~80% margin for a 81 day payback
  article2026-10-21

  Benchmark setting used for vLLM and SGLang results.

• How GLM5.3 Sparse Attention Affects HBM Memory Usage
  article2026-09-28

  InferenceX agentic workload whose September 28 snapshot results illustrate GLM-5.3 serving.

• Engrams Embedding Entendre: Codesign for Efficient DRAM/SSD Offloading
  article2026-09-18

  Agentic inference scenario in InferenceX used to test Engram DRAM offloading.

• Vera Rubin NVL72 Agentic Inference: 67x better Performance per Dollar
  article2026-09-14

  Agentic inference benchmark scenario that replays real world agentic traffic across SemiAnalysis's fleet of thousands of chips.

• Long Live the Short King: Why 4-hi HBM Wins
  article2026-09-13

  SemiAnalysis benchmarking of agentic workloads. Many workloads had pareto optimal results with KVCache offloading above certain concurrencies.