Aykut Onol

FYI, I've just started to work on this. Just wondering whether there is any local progress on the benchmark?

Nice, thanks! I'm working on a relaxed-IK benchmark to perform an A/B comparison for this feature.

Thanks for sharing the result. It seems pretty similar to [the original result for that benchmark](https://github.com/RobotLocomotion/drake/pull/17373#pullrequestreview-1009578989), i.e., ~7% improvement. For reference, we were expecting a higher impact on the speed...

I'd like to run a 7.4 vs. 7.2 comparison (unless there is a more promising baseline version). However, it seems we have access to only 7.4, 7.5, and 7.6. Any...

I see, thank you both for the quick response!

I've finally managed to get SNOPT 7.2 working with the most recent Drake. You can see the changes in [this draft PR](https://reviewable.io/reviews/RobotLocomotion/drake/19103). I've run a few experiments to compare the...

I think it is okay to work in python when comparing only the SNOPT outputs, and not taking the timing results so seriously. Yeah, the little dog example is great....

@jwnimmer-tri, what would be your thoughts on changing the default version to 7.2 and making 7.4 optional?

Got it, that's a fair concern.

Yeah, I think we should support them all in the adapter. I guess you're right about thread safety as well b/c `SnoptSolverTest.MultiThreadTest` fails for 7.2.