I am now up to 50 phoronix workloads as summarized on the workloads page. For each one I have a graph and some pages of information. My general idea is to take the benchmark-based Phoronix articles and see if I can start to keep up with cpu-focused benchmarks used. So far there have been three articles:

• January 4th – https://www.phoronix.com/review/intel-meteorlake-gcc-clang compiler performance focus where I have most of the benchmarks and still need six more. What I haven’t done is reproduce the recompilations used in the article. I expect to get the last six benchmarks.
• January 5th – https://www.phoronix.com/review/intel-5th-gen-xeon-avx512 Emerald Rapids AVX-512 performance. I have all the benchmarks here, though haven’t recompiled to see where AVX-512 helps and how it alters the profiles.
• January 7th – https://www.phoronix.com/review/z1-extreme-meteorlake is more focused on graphics with one non-graphics benchmark missing (simdjson); having some difficulties getting this one to run. Not sure how much I will run the graphics benchmarks.

At this point I expect if I try knocking off one benchmark a day, I should be able to keep up with newer articles for most benchmarks. This also gets me a good database to compare various metrics both in the metrics table and separately I have an Excel sheet that looks like below. This helps me compare these workloads to see what high/low values for topdown/ipc might metrics might be. Listed in green are those in top 10% and red are bottom 10% of each metric. On the left column those in green were part of a phoronix article and others were not.

This has also let me tune my workflow and gradually consider adding graphs as it made sense. I can now also look at high/low outliers from table below and dive deeper , e.g. on areas where one metric dominates.

There are some areas it also lets me consider a little further…

• There is some additional automation I could do, for example
  • More automatically generate web pages; perhaps not in the wordpress style but I can automatically generate pages that insert graphs and labels.
  • Rather than populate the table manually, have my tests generate MySQL code that can insert entries into a database. This more easily would let me expand the table to include metrics I don’t currently such as the amount of floating point or branch/cache miss ratios.
• Phoronix has a notion of a “default-benchmark” I have been running all the benchmarks which sometimes gives me a view of differences, but using a default might make them more comparable.
• Phoronix saves test results in an XML file. At minimum I might also save that file, but also see about more automatically printing the phoronix result? Perhaps part of using phoronix generation scripts for printing results.

This also gives me an area to dig deeper on various workloads and then incrementally improve the tools as well.