DeepSeek V4.1 Flash hits 222 tok/s on four DGX Sparks with capped output

V4.1 Flash 1017 reaches 222 tok/s on four DGX Sparks, though the author notes this edge case relies on capping max tokens at 10–200.

Norsyx

@norsyxArchived 9 days agoOriginally posted 2026.10.23

DeepSeek V4.1 Flash hits 222 tok/s on four DGX Sparks with capped output

Norsyx @norsyxOriginally posted 2026.10.23

Open original on X

222 tok/s 😹 DeepSeek v4.1 Flash 1017 on 4 DGX Sparks > Not a Clickbait but good for edge cases (kinda jev) > Set max tokens to 10-200 to restrict the decode > In the screenshot below that's a patched ZCode (to print the metrics)

x.com/norsyx/status/2113394895058772298Archived 2026.10.23 12:05

Open original on X