DS4F benchmarks hit 261 t/s single-session, 1479 t/s batch-16 on GB300 DGX
antirez · x · 2026-08-19
Developer MrCatid shared inference benchmarks for DS4F (DeepSeek 4 Flash) on a GB300 DGX Station using DFlash technology. The setup achieved 261 tokens/second in a single session and 1479 tokens/second with a batch size of 16. Redis creator antirez retweeted, expressing surprise at the performance and interest in acquiring the hardware.
Related event: DeepSeek-V4-Flash Benchmarks Show Up to 4,550 tok/s on a Single GB300(2 posts)→
More from Infra
- Hugging Face docs update: Shipping Triton autotune configurations — RisingSayak · 2026-08-19
- Matryoshka LM Suites: Train model suites in a single run — yoavartzi · 2026-08-19
- Liquid releases QAD quantized models, recovering 97% accuracy at 4-bit — helloiamleonie · 2026-08-19
- LoRA Training Test: Gradient Accumulation Is Not a Time-Zero Game — traceml-ai · 2026-08-19
- CXL standard nears deployment, benefiting Marvell and Google — BenBajarin · 2026-08-19
- LLM Midtraining Guide: Data Mixing and Overfitting Prevention — cwolferesearch · 2026-08-19