Alex Izvorski

@peastman From reading the parmed issue, these CMAPs are per residue not per atom type? So if we created copies of each atom type for each residue we could have...

@jchodera That makes sense! However it looks like the SMIRNOFF spec doesn't have CMAPs yet https://github.com/openforcefield/standards/issues/14 There's basically three directions I can see for getting this working: 1. hack in...

@asgharrazaviroivant Maybe better question - why do you need to wrap? Distances have to be calculated in a pbc aware way anyway, wrapped or not. Is it just a visualization...

@asgharrazaviroivant "In my system the initial pdb file that simulation starts from is all properly wrapped, however, during simulations trajectories proteins jump around due to improper wrapping." - Yes, but...

@SHervoe-Hansen That works, thanks! However I find the idea of a topology having box vectors very weird... how can any object have box vectors as the only 3d info it...

> According to my measurements there are 589M "potential" interactions (i.e. all interactions we check in inner loops for r < cutoff) and only 160M or them are "positive" interactions...

> That assumes all reads can be fully coalesced. Reading the data for the first 32 atoms in each tile should be coalesced, since they're contiguous. The second set of...

> > I think the 64bit fixed point representation of forces is used because there is atomic 64bit addition? This accounts for about 3/4 of the total theoretical bandwidth >...

> The force calculation requires more data than just the positions and charges, right? The `LOAD_ATOM2_PARAMETERS` line expands into loading data from `global_nonbonded3_charge` and `global_nonbonded3_sigmaEpsilon` I think. @dmclark17 Yes, true,...

> They just get updated with `atomicAdd()` @peastman Isn't atomicAdd implemented as a read/add/write? I don't think the DRAM itself has the ability to add. > if the same block...