AIXDAabout 2 hours ago
Two DGX Sparks Run 284B Model at Cloud Speed
1 min read
Two Nvidia DGX Spark machines pool 256 GB of memory to run DeepSeek V4 Flash at 36 to 41 tokens per second, tripling single-machine throughput. The setup runs the model at native FP8, reaches 76 tokens per second on code, and never exceeds 272W.
Level
Hype check
Tap to vote and see what everyone thinks.
