Looking for comparative benchmarks of DeepPRIME processing on M4 Macs, as DP XD2s is the main bottleneck in my high-volume event photo biz. My M1 Max Studio averages 18s for DP XD2s on a 61MP RAW. The M4’s Neural Engine is twice as fast. Does this translate into 2x faster DP processing vs M1/2/3?
Probably not. Each image takes a few cycles on other processors for management. Also, only part of the processing is using the ANE. Moreover, timing also depends on how many images are treated concurrently.
Nevertheless, exports should finish in less time.
I was hoping for some actual benchmarks. I ran my own comparing my previous M1 Mac mini against my current M1 Max Mac Studio, and they were interesting. It was clear that the Studio’s extra GPU cores contributed substantially. This is why I’m trying to get a clear assessment of the contribution made by the M4’s beefed-up Neural Engine.
It’s possible that M4 doesn’t make much of a difference:
I do have an original M1 MacBook Pro, plus my current M3 Pro MacBook Pro, but I don’t have PL8 installed on the former. One thing to consider is this: unless you have a Mac Studio with Mx Ultra, then all of a given generation are equal for ANE because only the Ultras have anything other than 16 ANE cores.
Now, anecdotally, I reckon my XD2s exports run somewhat faster on my M3 than my plain DeepPRIME ones did on M1. (I never really used XD.) The claimed improvement M1-M3 for the ANE cores was 60% if I remember correctly.
If you want to supply me with the test files (including DOPs), I would be happy to run a test export on my M3 Pro using ANE. If using GPU I have an M3 Pro with 18 cores — substantially less than the Max chips.
Hello, there seems to be an issue with Apple silicon support. Surprisingly, the M1 Max performs faster than the M4 Max. Additionally, when I switch the rendering engine in the settings to M4 CPU or GPU, I don’t notice any significant performance improvement. Furthermore, the battery consumption remains the same, indicating that it doesn’t fully utilize the available performance.
Sounds like this needs to be raised with DxO via a Support Ticket. It seems you have the necessary hardware to provide them with clear test data too 