Pedro Cuenca(@pcuenq):With @jundotkim, we'd like to measure how much a DGX Spark would accelerate prefill of DeepSeek V4.1 Flash running on M3 Ultra. Something like this below, with the Spark being in charge of prefill. Has anybody done that? cc @Coolmark482, perhaps you know?

Pedro Cuenca

@pcuenq

ML Engineer at 🤗 Hugging Face | Co-founder at LateNiteSoft (Camera+). I love AI and photography.

Joined May 2019

1.1K Following 7.9K Followers

Pedro Cuenca@pcuenq

2026.11.04 10:09

With @jundotkim, we'd like to measure how much a DGX Spark would accelerate prefill of DeepSeek V4.1 Flash running on M3 Ultra. Something like this below, with the Spark being in charge of prefill. Has anybody done that? cc @Coolmark482, perhaps you know?

Pedro Cuenca@pcuenq

2026.10.07 13:29

It's crazy that I can run MiMo 2.6 Flash across my RTX 6000 GPU and my M5 laptop at 40 tokens/sec over 10 GbE 🤯 These are the native mxfp4 weights of a state-of-the-art model, on heterogeneous hardware. Supported out of the box in llama.cpp.