With the latest batch of work units we've been able to give the 5800x3d a workout.

Was hoping the huge L3 Cache would make a big difference but performance is similar to other Ryzens.

With same ram and same # of workunits running concurrently the 5800x3d is outperforming a 3900x by 15%. I think a lot of that can be attributed to the 5th generation improvements as much as to the cache. The chip is very efficient but I suspect the 5700x or 5900x are better bang for buck (quite the premium for the x3d).

Anyways thought everyone ought to know :)

Thanks! I've been eyeing one of those too, but just couldn't justify dropping the coin on them on the hope that they'd improve performance. I've got a pair of 3900Xs purring away, and they're 64MB L3/ea (though across two of the chiplets). It would be interesting to compare it to a small-cache version of the same architecture. But my impression of the CPDN tasks is that they're just really hard on the RAM controllers, as there's a lot of data flowing around.

With how cheap 5900x's have become I had a funny feeling it wouldn't be good value but I had to try.. For Science! :)

I can't speak for the Hadley Centre models but bear in mind that OpenIFS is tuned for performance assuming small cache (actually my i7-11700). We have to aim for good performance across the board. I could make the model faster if I targeted specific chips but that's not practical for CPDN's very wide range of volunteer machines.

Having run a decent number of them I can confirm OpenIFS models do not benefit from the 3d cache in any meaningful way :)