Current required Topaz version is too old for new GPUs

Install, Configure and Update
1

Hi,

The current Topaz integration documentation [0] states that the only version supported is 0.2.5 (and with python 3.6 specifically). Unfortunately, this combination does not allow using more recent pytorch dependency and prevents Topaz to run on new GPU hardware (Nvidia Blackwell architecture for instance)

Thanks to the 4.7.1+cu12 build it’s possible to run CryoSPARC jobs on Nvidia Blackwell GPUs, but due to the current Topaz version restriction, one has to limit CryoSPARC workers to older GPUs or accept to fallback to CPU which is much slower.

Is there any work in progress to relax the Topaz version restriction and allow using more recent Topaz versions to leverage more recent Nvidia GPUs?

Maybe Topaz jobs are identifiable through the {{ job_type }} so one can redirect only those to older GPUs?

Thank you for your help!

Youssef

[0] Topaz (Bepler, et al) | CryoSPARC Guide

2

Cryosparc v5 is listed as compatible with Topaz 0.3.0+ : CryoSPARC: CryoSPARC v5.0 Updates

3

Thank you @boggild for pointing this out, I overlooked 5.x release notes since it’s still beta. Also multiple parts of the documentation has already been updated to reflect behaviour change between 4 and 5 but not Topaz one so I wrongly assumed that the 0.2.5 version lock was still effective.