Finally! TT-Boltz now runs on a @tenstorrent QuietBox, fully parallelized across all four Blackhole cards.
— Moritz Thüning (@moritzthuening) 23 février 2026
This yields a 4x speedup, making QuietBox the best product for anyone who wants to run Boltz-2 locally at scale.
Very soon we'll parallelize it on all 32 processors of a… pic.twitter.com/jYKhkmzgKU
Finally! TT-Boltz now runs on a @tenstorrent QuietBox, fully parallelized across all four Blackhole cards. This yields a 4x speedup, making QuietBox the best product for anyone who wants to run Boltz-2 locally at scale. Very soon we'll parallelize it on all 32 processors of a Galaxy server. It’s pretty clear by now that GPUs aren’t the best bet for LLM inference. The same shift will happen to other fields like biotech.
→ View original post on X — @tenstorrent, 2026-02-23 20:05 UTC