P: Slow performance on Xeon CPUs

Same with 6/12 cores used in LR; 10 secs for rendering 40MB nefs + edits at start, and 25 secs after rendering 1000 images.

So its something to do with the edits that is slowing down rendering with time, not the no of cores.

Any ideas, Simon?

Bob Frost

CORRECTION! My old problem of 1:1 rendering slowing down with time only seems to be true if the nefs are unedited.

I just repeated one run with LR using just 8 cores, but with the following edits (Autotune/Punch10/Sharpen/CameraNeutral/Noise/LensCorrection/ WBAuto/GradFilter/8Spotsremove). I edited the first file and then pasted the edits to the rest of the 2700 nefs, and re-rendered standard previews.

When LR had settled down, I then rendered 1:1 previews of the 2700 files. At start the edited images were taking 12 secs to render, BUT after only 1000 images they were taking 20 secs to render!!

I stopped the rendering and restarted LR and the rendering was back to 12 secs. That has been my experience in the past; so the increase in rendering time has something to do with rendering the edits.

I'll repeat this with all 12 cores used in LR tomorrow.

Bob Frost

My main complaint in the past few years has been about the slowdown in rendering 1:1 previews of my 40MB nefs. However, some tests yesterday showed that this may have been cured somewhere along the update line to CC 2015.7

I selected 2700 40MB nefs (D810) without any edits (I reimported them into a separate folder and allowed LR to settle down for an hour or so), and then started rendering 1:1 previews (after allowing LR to render the standard previews as part of the import.)

At first they were taking about 4.0 secs per image, but soon settled down to 5.0 secs per image. They stayed at 5 secs until the completion in about 3.5 hrs. No slowdown with the i7-5930 6/12 core cpu (all 12 logical processors in use according to system info).

Then I ran LR (CC 2015.7) again (after deleting the 1:1 previews, restarting LR and allowing it to settle down again), but with the affinity set to FF (using logical processors 0-7). LR system info then said it was using 8 logical processors out of the 12 available. The time at start for each image was 4.6 secs, but soon settled down again to 5 secs, and stayed at 5 secs to the end.

Conclusions: No slowdown in rendering over 3.5 hrs, and no significant effect of using 4/8 cores or 6/12 cores. So my old problem seems to have been solved. I must do a big run sometime - overnight and see what happens then, but I'm optimistic.

But I haven't tested simply editing a load of files in succession with 8 or 12 cores. More coffee or something stronger needed for that!

Bob Frost

Yes the same with LR 2015.7. The problem with 2015.7 seams a little increased (=I can reproduce it faster!)

I can reproduce the observation that LR is performant right after a launch and then gets slower after intense brushing also with one single photo, but with walk over several photos I can reproduce it faster.

I tuned off all: sync, face recognition etc.

Now I will extent the summary:

1. Different photo type of different cameras: no difference
2. Now I'm not sure that a reduction of CPU kernels helps a lot. I tested with:

start /affinity F cmd.exe /c "c:\Program Files\Adobe\Adobe Lightroom\lightroom.exe"

and also with the tool "Process lasso".

Seems that not help much. Helps only a little to increase the time until the problem occurs. Or on 2015.7 helps fewer than with 2015.6.1 (I mad my first tests with 2015.6.1, now with 2015.7)

3. The problem is not only a brush or a clone problem. It is a general problem in the develop module when I must elaborate many photos in one session. But with brush and clone tool I can reproduce it very fast!