Pinned
32 parallel runs on the AMD Strix Halo hitting 290 t/sec (highly prompt dependent) with Gemma4 26b QAT MTP.
I made 32 separate little web games and html pages.
Check them out here:
llm.ciru.ai/research/32slo…
That's about one finished game/app every 10 seconds. Some of them are
16 parallel runs of Gemma 4 26B A4B on a single NVIDIA DGX Spark!
Pushing 18 tok/s per instance and a 300 tok/s aggregate. It can even hit 32 parallel runs.
This level of concurrency highlights how efficient the architecture is.
