@FishAudio With SGLang-Omni's SPG topology, moving Drama 3's vocoder to a second H200 delivers 72% higher throughput and 37% lower median TTFA at 256 concurrent streams.
Learn how: https://www.sglang.io/blog/spg-topology
SGLang-Omni's SPG topology lifts Drama 3 vocoder throughput 72% on H200
AISummary
SGLang says that with its SPG topology, moving Fish Audio's Drama 3 vocoder to a second H200 raises throughput 72% and cuts median time to first audio 37% at 256 concurrent streams. The post links a blog page explaining the approach.
AIWhy it matters
The post gives concrete throughput and latency gains from moving a voice vocoder to a second GPU, useful for judging serving topology tradeoffs.
Post on XView on X
Source: SGLang · x.comPublished · added here