casey wrote:No idea - I never checked the final image. I will go re-run it and verify that the image is correct.
OK I re-ran the benchmark and had it output both an MXI and a PNG, since it seemed like that was what the benchmark wanted to do by default. Looks just like the reference image that you posted. The new score is: SL of 15.00. Benchmark of 562.938. Time: 19m04s
That's actually faster than the old score (559.394), so I assume that yes, the original posting was accurate and was using the images.
casey wrote:But the thing is, I don't see why you would have to. It seems like a Maxwell bug that it can't find the textures. Am I wrong? I mean if Maxwell can't find a texture in its prescribed path, should it search the -bitmaps specified path? It's clearly not doing that. Can someone else repro this to make sure it's not just me screwing up?
OK I figured out what was going on here. The problem seems to be that Maxwell doesn't interpret ~ in the paths correctly, perhaps including the working directory variable? If I expand out the -bitmaps, -mxs, -mxi, etc. paths manually on the command line to be /home/farm/ based instead of ~/ based, it locates the textures. If I don't, and they stay as ~/ based paths, then it can't.
- Casey
BTW. This is actually a six-machine farm configuration, and I ran the test on a single machine, so I'm assuming that the "aggregate" benchmark is therefore something like 3300?? That seems excellent since the entire farm cost was approx. $2000. Maybe we need a new metric: benchwells-per-dollar, or bpd's? So this is a 1.65bpd farm
