DeepSeek-V4-Flash tops out at 770 tok/s on one B300 in a vLLM batch test
Moreh · reddit · 2026-07-22
A Reddit user is benchmarking DeepSeek-V4-Flash on a single B300 with vLLM 0.25.0 and says throughput is only about 770 output tokens/s at batch size 256.
- They expected several thousand tok/s on a B300 and suspect a configuration issue.
- deepgemmmegamoe fails on a single GPU because it requires expert parallelism, so they fall back to flashinfertrtllm.
- Dropping speculative decoding roughly doubles throughput.
- They suspect sparse MLA attention may be running eagerly without CUDA graphs.
- The post asks for real-world tok/s numbers and working single-GPU MoE backends.
More from Infra
- Resin lets an agent build standalone LLM inference engines; vLLM and SGLang optimize DeepSeek kernels differently — AccBalanced · 2026-10-27
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- Anthropic S-1: CSP marketplace revenue jumped from 11% to 48% in three years — AccBalanced · 2026-10-27
- Alchemy adds effect/Workflow support across Durable Objects, Workflows and Lambda Durable Functions — samgoodwin89 · 2026-10-27
- Alchemy's Lambda Web Functions run plain Node web servers on Lambda, no OpenNext needed — samgoodwin89 · 2026-10-27
- Alchemy Cuts Cloudflare Bindings CPU Time by Up to 166× Via Profiling — samgoodwin89 · 2026-10-27