Running Qwen3.8-Flash-Next NVFP4 at 184 tok/s on a single RTX 6000 Pro

Been running versions of this on my pc, if you have 64 gigs ram, you can run it on a 12+ gig cards easily.

Been running Qwen3.8-Flash-Next NVFP4 for weeks now. What's new about this?

hn title is completely different from the article title, which takes clicking the hn title to discover.

> Our first model-kernel combos

They did a specific-hardware kernel optimization for the Pro 6000 (available on AWS as g7e).

On "just" a RTX 6000 Pro? You mean one of those $16,000 gpus?

it is very hard to run 150k token context with these kind of setups. even if you do, all of us are used to running multiple agents at once.

i hope ram prices will come down a little.