Skip to main content
Participant
Jul 17, 2026 12:53 AM UTC
Question

Lightroom Classic 15.4.1: export caps P-cores at 2.4GHz on M5 Pro (macOS 26)

# Environment
- Mac: MacBook Pro 14" 2026, M5 Pro (18-core) (identifier: Mac17,9)
- RAM: 64 GB
- macOS: 26.5.1 (build 25F80)
- Lightroom Classic: 15.4.1 (latest)
- Power: AC connected, High Power Mode enabled (powermode 2)
- Catalog: freshly created; source files and export destination both on internal SSD
- Previous machine for comparison: MacBook Pro (M1 Pro)

 

# Symptom
During export, total CPU utilization plateaus around 50% and exports are much slower
than expected. Using powermetrics, I confirmed that while exporting, the P-cores
(P0 cluster) are ~100% active but locked at roughly 2.4 GHz (2304-2652 MHz),
never boosting above 3 GHz. The P1 cluster stays mostly idle.

Measured comparison against M1 Pro (same photos, same export settings):
- Export: 92 s on M1 Pro -> 70 s on M5 Pro (only ~24% faster)
- AI Denoise: 9.5 s on M1 Pro -> 4.5 s on M5 Pro (~2.1x faster)
GPU/Neural Engine-bound tasks like Denoise scale as expected with the new chip,
while the CPU-bound export barely improves. Also, on the M1 Pro all cores were
fully utilized during export, whereas on the M5 Pro the P-cores are frequency-capped
as described above. This looks like a regression in export behavior across
chip generations.

 

# What I have ruled out
1. Thermal throttling: thermal pressure stays "Nominal" during export
   (powermetrics --samplers thermal)
2. Low Power Mode: off (High Power Mode is enabled)
3. Background apps: reproduced with all third-party background apps quit
4. App Nap: disabled via NSAppSleepDisabled
5. taskpolicy -B applied to the Lightroom process: no change
6. Splitting the export into multiple simultaneous batches: CPU usage did not increase
7. Hardware sanity check: on the same machine, a synthetic CPU load test
   unrelated to Lightroom (running 12 instances of "yes > /dev/null" in Terminal
   to load all P-cores at 100%) boosts the P-cores to 4.2-4.4 GHz with thermal
   pressure still Nominal. This confirms the chip, cooling, and power settings
   are all healthy - the frequency cap only occurs during Lightroom exports.

 

# Conclusion (hypothesis)
The export worker threads appear to be created with a background/low QoS class,
so macOS runs them at reduced clock speeds even though the cores are available.
This matches reports in this thread:
https://community.adobe.com/t5/lightroom-classic-discussions/why-does-lightroom-not-use-the-performance-cores-m4max-during-export/td-p/15349358

 

# Request
Please review the QoS/priority of the export worker threads so that Apple Silicon
P-cores can run at their normal frequencies during export. I can provide full
powermetrics logs on request.

 

# Reference data
- During Lightroom export: P0 cluster averages ~2005-2449 MHz at 90-100% utilization
- Synthetic CPU load test: P0 cluster averages 3568 MHz (~53% residency at
  4308-4380 MHz), P1 cluster averages 4002 MHz

4 replies

TKAuthor
Participant
Nov 14, 2026 4:27 PM UTC

Update: Lightroom Classic 16.0 — partial improvement, but the frequency cap remains

Retested on 16.0 with the same machine, same photo set and same export settings.
The full powermetrics log from this run is attached.

Environment: MacBook Pro 14" 2026, M5 Pro (18-core, 12 P-cores in two clusters),
64 GB RAM, macOS 26.5.1. AC connected, lid open, thermal pressure Nominal
throughout the measurement.

## What improved

Core scheduling. On 15.4.1 / 15.5 / 15.5.1, the two P-core clusters alternated:
one ran at ~100% while the other sat at 18-43%, so only about six P-cores' worth
of work was ever in flight. On 16.0 both clusters run at 100% simultaneously
across every sample — all 12 P-cores are now being used.

Export time dropped accordingly:
- 15.4.1: ~70 s
- 15.5 / 15.5.1: ~75-80 s
- 16.0: 57 s

(The 57 s run was measured while a background mediaanalysisd task was consuming
roughly one core, so a clean run may be slightly faster.)

## What did not change

The clock cap. P-cores are still pinned to the same narrow band during export
(see the attached log for all five samples):

  P0-Cluster HW active frequency: 2513-2610 MHz
  P1-Cluster HW active frequency: 2517-2610 MHz
  Residency above 4308 MHz: 0.1-0.5%
  CPU Power: 12.0-14.6 W
  Current pressure level: Nominal (all samples)

Over 99% of the residency sits at just two steps, 2304 MHz and 2652 MHz, while
both clusters report 100% active residency. The cores are fully busy and still
held at roughly 57% of their maximum clock.

For reference, a synthetic all-core load (12x "yes > /dev/null") on the same
machine reaches 4.2-4.4 GHz with thermal pressure still Nominal — the log for
that is in my earlier post in this thread. So the silicon, cooling and power
settings are all fine — the cap appears only during Lightroom exports, which is
consistent with the export worker threads being created at a background/low QoS
class.

## Request

The 16.0 change to core scheduling is a real and welcome improvement. The
remaining piece is the thread QoS: lifting the clock cap should yield another
significant speedup on top of what 16.0 already delivers.

Attached: lrc_16_export.txt (powermetrics during export on 16.0, 5 samples)

Translate
TKAuthor
Participant
Aug 13, 2026 2:49 PM UTC

Update: still not fixed in Lightroom Classic 15.5.

I re-tested on 15.5 (MacBook Pro 14" 2026, M5 Pro 18-core, macOS 26.5.1) after a
clean reboot with all background apps quit. The behavior is unchanged: during export
the P-cores stay at roughly 2.3-2.6 GHz with essentially zero residency above 4 GHz,
while thermal pressure remains "Nominal" the whole time. Export times are the same as
on 15.4.1 (around 70-80 s for the same photo set and settings).

One minor difference: the load now shifts between the two P-core clusters (P1 runs at
100% while P0 sits at 34-48%, where previously it was the other way round), but the
total is still only about six P-cores' worth of work at a capped clock. CPU package
power stays around 11-18 W.

Translate
TKAuthor
Participant
Jul 17, 2026 6:20 AM UTC

Here's an excerpt from the log.

--- During Lightroom export ---
P0-Cluster HW active frequency: 2431 MHz
P0-Cluster HW active residency:  40.88% (1344 MHz: .18% 1644 MHz:   0% 1992 MHz: 5.9% 2304 MHz:  14% 2652 MHz:  20% 2964 MHz:   0% 3240 MHz:   0% 3504 MHz:   0% 3696 MHz:   0% 3876 MHz:   0% 4044 MHz:   0% 4176 MHz:   0% 4284 MHz:   0% 4308 MHz: .05% 4380 MHz:   0%)
P1-Cluster HW active frequency: 2505 MHz
P1-Cluster HW active residency: 100.00% (1344 MHz:   0% 1644 MHz:   0% 1992 MHz:  11% 2304 MHz:  16% 2652 MHz:  73% 2964 MHz: .14% 3240 MHz:   0% 3504 MHz:   0% 3696 MHz:   0% 3876 MHz:   0% 4044 MHz:   0% 4176 MHz:   0% 4284 MHz:   0% 4308 MHz: .24% 4380 MHz:   0%)
CPU Power: 13166 mW
Current pressure level: Nominal

--- Synthetic load (12x yes) ---
P0-Cluster HW active frequency: 4372 MHz
P0-Cluster HW active residency:  97.84% (1344 MHz:   0% 1644 MHz:   0% 1992 MHz:   0% 2304 MHz: .33% 2652 MHz:   0% 2964 MHz:   0% 3240 MHz:   0% 3504 MHz:   0% 3696 MHz:   0% 3876 MHz:   0% 4044 MHz:   0% 4176 MHz:   0% 4284 MHz:   0% 4308 MHz: 1.4% 4380 MHz:  96%)
P1-Cluster HW active frequency: 4346 MHz
P1-Cluster HW active residency: 100.00% (1344 MHz:   0% 1644 MHz:   0% 1992 MHz:   0% 2304 MHz: .17% 2652 MHz:   0% 2964 MHz:   0% 3240 MHz:   0% 3504 MHz:   0% 3696 MHz:   0% 3876 MHz:   0% 4044 MHz:   0% 4176 MHz:   0% 4284 MHz: 7.1% 4308 MHz:  33% 4380 MHz:  60%)
CPU Power: 12850 mW
Current pressure level: Nominal

Translate