@foonathan I'm mostly impressed your compile jobs have enough parallelism for 128 cores. I guess the .cpp/.o translation unit model has its perks even if it's extremely wasteful/redundant in other respects.
@dotstdy@wolfpld@pervognsen@foonathan I might be mistaken, but I'd say the mold linker is able to do that.
At least max out 64 cores when linking something like Chrome or Firefox.
@molecularmusing@wolfpld@pervognsen@foonathan mold limits concurrency to 32 threads by default because it doesn't really scale beyond that. Still much better than all the others tho :')
I limit PDB loading to 16 threads because I had a hard time scaling beyond that. Some things are just not very thread-friendly, and/or you'll run into OS locks sooner or later.
@molecularmusing@dotstdy@wolfpld@foonathan Obligatory scalability graph of an LBM fluid solver's scalability graph. Note that x-axis is counted in units of 1024 processors.
@pervognsen@molecularmusing@dotstdy@wolfpld@foonathan Physicists have it easy because nature makes non-locality hard/weak. Out here in the man-made world, one ship stopping at an embargoed port can instantaneously cause trouble for the rest of the fleet around the globe (not that I was traumatised by liner shipping problems).
I bet linking could scale linearly if there was an inverse square law for correctness ;)
@pkhuong@molecularmusing@dotstdy@wolfpld@foonathan You say but of course all the conventional Navier-Stokes-based solvers have to deal with global conservation laws which limits their scalability. But yeah the fact that the underlying physical problem can be cast in terms of local particle interactions is what makes alternative methods like LBM possible.
@pervognsen 16GB but I do a lot of work over ssh on my dev server where I have 128GB of RAM, plenty of CPU, a couple TB of NVME storage and like 100TB of spinning rust
@zeux I'm day-dreaming about the workstation I want to build later this year when I come back from my trip to Denmark. I'm thinking I need at least 128 GB (well, "need" is a strong word). I have 32 GB in my current computer (a reasonably higher-end laptop) and it's definitely way too little.
Add comment