Thank you for that link.
The Xeon E5 2687W v3 has some asymmetry in it. I wonder if that affects their test (where they are effectively flagging that some cores are off-limit for Lightroom):
http://anandtech.com/show/8584/intel-xeon-e5-2687w-v3-and-e5-2650-v3-review-haswell-ep-with-10-coresExporting raw images 80 ought to be a perfect candidate for multi-threading. Those 80 jobs should be AFAIK totally independent and each job is time-consuming. I am not sure that I am buying their conclusion that Lightroom scales well with #cpu cores but poorly with #cpus _because of_ interdependency.
Could it be that the PC they used for the test gets maxed out in memory bandwidth and adding one more cpu simply hits the same wall? How many passes and temporary buffers are needed in order to do that 1.8GB -> 700MB? (1800 [MB]+700 [MB])/100 [seconds] = 25MB/s of total read/write, not a very large number for cache systems or main memory (but within OOM for spinning harddrives?). Could it be that their method for reading/writing to disk somehow fails to take advantage of the bandwidth of their flash drive?
It seems to have 2.5MB of cache per core for a total of 10*2.5 = 25 MB per cpu for a total of 2*25 = 50MB for the system. Each raw file is 1800MB/80=22.5 MB, thus the working set of 20 threads (assuming that they are processed in one go) would be 450MB, far more than the total cache. Depending on how Adobe do things, this could mean significant extra traffic to main memory?
Another possibility would be that their threading is simply hardcoded limited to ~10 threads?
If one did a similar test with a dual 4-core system, one might be able to rule out some of the possibilities.
-h