Paper work to do over break

[vsoch]

• Go through current paper and un-generalize (I left pieces general that didn't need pointing fingers)
• Write about issues with CUDA_VISIBLE_DEVICES
• Colors for different environments need to be consistent across plots for environments a731e64
• Finish adding each application, including description in methods (paper / docs citation), and results (and plots)
• Redo all plots with white and black axes (I decided the gray won't print well and is distracting)
• Think about what we want for single-node benchmarks (which we haven't solved)
• Figure out how to make institution latex wider so prettier looking
• Talk about GPU error correction setting as a possible explanation for stream on GPU (discussion)
• Add mixbench to paper, since we need results for error correction (note that I added a mention of it just in the context of stream, I don't think we can use the full results)

I will add more items as I hit them.

Discussion items (for after break):

• We have a lot of content - discussion of costs / contention of AI might be out of scope here.
• I don't think we should add multi-gpu-models - it didn't run anywhere except for lassen and compute engine, and there were a lot of error messages
• Should we add cores to the table of environments? (see node characteristics)
• Which apps/plots to include, and for which do we want speedup (I'm currently doing miniFE for speedups)
• I'm currently writing main parameters into application under methods. Is that sufficient?
• Do we still want to mention all apps (e.g., kripke / quicksilver) even if we don't include results / data?
• Look at stream docs and decide if we are using the right FOM / calculation, beyond reporting raw values.