Skip to content
View original post on X: vLLMOfficial
Published
44/100

Built-in DSpark: 1.92–2.16x output throughput in FP8 and 2.00–2.23x in NVFP4, versus the matching baseline.

AIvLLM says its built-in DSpark support raises output throughput 1.92–2.16x in FP8 and 2.00–2.23x in NVFP4 versus each precision's matching baseline. The tests used TP8, 200 real prompts, greedy decoding, and up to 512 output tokens.

AIWhy it matters.

The post gives measured throughput ranges for FP8 and NVFP4 against matching baselines, along with the test setup, so readers can check the claimed speedup.

vLLMVerified on XOfficial
@vllm_project
View on X

Part of a thread · earlier post

Built-in DSpark: 1.92–2.16x output throughput in FP8 and 2.00–2.23x in NVFP4, versus the matching baseline.

TP8 · 200 real prompts · greedy decoding · ≤512 output tokens.

Cross-layer KV sharing, context parallelism and NIXL P/D serving handle long requests.

Source details

vLLM · x.comPublished

Topics

Open-source ecosystemDeployment & engineeringProduct updates