Skip to main content
.
.Michael.
Inspiring
May 24, 2025
Answered

Intel CPU: At photo export the pCores are ignored, only the eight eCores have 100 % load!

  • May 24, 2025
  • 2 replies
  • 1263 views

Hello,

with LrC 14.3.1 (and also some previous version) the picture export via 'File -> Export ...' is compared to the past limited regarding its CPU usage, i.e. now the pCores of my i7 13700K are almost ignored (idle at 800 MHz). This slows down the file export a lot. Do I have missed some new option? How do I get back the usage of the pCores? Maybe also important: My NVIDIA 4070S GPU is also only at approx. 15 % load.

Addendum:

Now, about 3 hours after start, the hardware usage looks completely different:

2 pCores are with about 80 % load in use (running with approx. 5 GHz each), the eCores and the GPU is idle. 

 

Greetings!

 

.Michael.

 

P.S. I use a PC with Win11 Home 24H2 and 32 GB of DDR5 RAM.

Correct answer .Michael.

"Why is a 100 % e-cores load plus approx. 10 % p-cores load faster then a 100 % p-cores load?"

 

The difference between 28 secs and 34 secs may not be significant here. In my experience doing measurements of LR over the years, to get consistent times, you need to use a larger workflow, e.g. 200 photos rather than 26. There's too much variability with smaller numbers of photos. Also important are repeating each test three times, alternating between the two things being measured, disabling background Window activities, disabling other plugins that might be running in background, starting each test with a freshly started Windows to have a cold file cache, a freshly started LR to have the same amount of initial memory (LR's memory allocation and garbage collection can introduce large amounts of variability), etc. will get more consistent results.


@johnrellis Thank you! Lets summarize:

  1. The p-cores alone won't help to improve the picture export speed. (Hint: My measured times are at least twice tested per test target, they are consistent for this purpose.)
  2. The usage of the e-cores together with some minor p-core load seems not to limit the picture export speed.
  3. Your HW is more up to date than mine and it shows similar behavior. I.e. there is no obvious picture export issue.
  4. The aspect that the picture export behavior "completely" changed after some hours: Was not covered in this thread, could be a glitch / random effect.

This means overall that my detected change at CPU usage is no problem. There was maybe never a change because I may have remembered to the time before I installed the 4070S GPU.

2 replies

johnrellis
johnrellis
Legend
May 24, 2025

First, apps don't control the assignment of threads to P-cores and E-cores -- that's done under the covers by Windows and the Intel Thread Director built-in to the hardware.

 

Second, Export's image rendering is split between the CPU and GPU, and CPU/GPU utilization less than 100% doesn't necessarily indicate that the computer could go faster if either was 100%. There may be another limiting factor, e.g. memory bandwidth between CPUs and GPUs or disk write speed (which should only be a factor with exports if the maximum disk speed is particularly slow, e.g. a slow network drive).

 

Similarly, that P-cores aren't being used doesn't mean that the export could go faster if they were. The GPU may be the limiting factor, and the E-cores are keeping up with it just fine.

 

I suggest the following two next steps:

 

1. Triple check you have the latest Nvidia Studio driver (576.52):

https://www.nvidia.com/en-us/drivers/details/245675/

 

2. Do this experiment: Select 50 files. Use your watch to time how long an export takes, with and without the GPU enabled for export. (Set Preferences > Performance > Use Graphics Processor to Custom, then check and uncheck Use GPU For Export). What are the two times? With a modern GPU (e.g. the 4070 Super) you should see significant speedups using the GPU. But let's verify that's true in your case.

    .
    .Michael.Author
    Inspiring
    May 25, 2025

    Thanks for your proposals, johnrellis.

    1) I have this GPU driver installed:

    Graphics Processor Info:
    DirectX: NVIDIA GeForce RTX 4070 SUPER (32.0.15.7652)
    Init State: GPU for Export supported by default
    User Preference: Auto
    Enable HDR in Library: OFF

    i.e. as recommended.

    2) If I export 26 of my current 40 MPixel DNGs to JPGs then I get these timings:

    2.1) with GPU: 28 s

    2.2) w/o GPU: 40 s

    at 2.1) only eCores + "less" 4070S GPU is in use, see the below attached screenshot of the Task Manager 

    at 2.2) eCores and pCores are 100 % in use

    3) You write: 

    quote

    Similarly, that P-cores aren't being used doesn't mean that the export could go faster if they were. The GPU may be the limiting factor, and the E-cores are keeping up with it just fine.

    By @johnrellis

    This is in general right but my two provided screenshots "with GPU in use" show this: 

    a) All of my eCores are at 100 %, i.e. a switch to pCores would gain performance.

    b) The GPU is very limited in use.

    -------------------------------------------------------

    side information: If LrC is not in focus, in my case the Task Manager is selected, then at 2) this can be seen:

    at 2.2) Only the eCores are 100 % in use and the export takes 73 s.

    at 2.1) the time doubles to 56 s but the GPU and eCores usage looks not much different to me ...

    Somehow it makes sense, that MS-Windows does not give the full CPU prio to an app which is not in focus, but I'm anyhow surprised about it (because the Task Manager does not need much performance).

    .
    .Michael.AuthorCorrect answer
    Inspiring
    May 28, 2025

    "Why is a 100 % e-cores load plus approx. 10 % p-cores load faster then a 100 % p-cores load?"

     

    The difference between 28 secs and 34 secs may not be significant here. In my experience doing measurements of LR over the years, to get consistent times, you need to use a larger workflow, e.g. 200 photos rather than 26. There's too much variability with smaller numbers of photos. Also important are repeating each test three times, alternating between the two things being measured, disabling background Window activities, disabling other plugins that might be running in background, starting each test with a freshly started Windows to have a cold file cache, a freshly started LR to have the same amount of initial memory (LR's memory allocation and garbage collection can introduce large amounts of variability), etc. will get more consistent results.


    @johnrellis Thank you! Lets summarize:

    1. The p-cores alone won't help to improve the picture export speed. (Hint: My measured times are at least twice tested per test target, they are consistent for this purpose.)
    2. The usage of the e-cores together with some minor p-core load seems not to limit the picture export speed.
    3. Your HW is more up to date than mine and it shows similar behavior. I.e. there is no obvious picture export issue.
    4. The aspect that the picture export behavior "completely" changed after some hours: Was not covered in this thread, could be a glitch / random effect.

    This means overall that my detected change at CPU usage is no problem. There was maybe never a change because I may have remembered to the time before I installed the 4070S GPU.

    GoldingD
    GoldingD
    Legend
    May 24, 2025

    In LrC, click on Help, then Systm info, Copy that Info , and paste it into a reply. Note that I am mostly interested in the section about your GPU. Following is an example:

     

     

    Graphics Processor Info:
    DirectX: NVIDIA GeForce RTX 3070 (32.0.15.6636)
    Init State: GPU for Export supported by default
    User Preference: Auto
    Enable HDR in Library: OFF

     

    Terms & ConditionsAccessibility statement
    Using the community Terms of use Privacy Cookie preferences Do not sell or share my personal information ad choicesAdChoices