jhapeman
Offline
Upload & Sell: On
|
 New Mac and PC Lightroom Benchmarking | |
I have the luxury of having multiple Macs and PCs that are used in my business so last year I posted some comparisons on some common Lightroom and Topaz tasks. With the release of the AI-denoise in LR Classic v 12.3 I have now moved on to that product, so I am adding a chart for that here although I did also update the Topaz chart. I also am not putting in some charts I did before where almost all of the computers performed very similarly. Finally, we added a nice laptop for the CAD designers to use here, so I added that to the matrix for the PC users. It's not 100% top of the line, but pretty high end, with a discrete Nvidia RTX 3070 GPU. Observations as follows:
1) The Apple Silicon does an impressive job with many tasks, but Adobe's Denoise really sticks out. This result is particularly impressive as right now Adobe is not using the Apple neural engines as there's a bug in MacOS Ventura that causes issues, so they have disabled them until Apple fixes it. That will only increase the performance on Apple Silicon, and Adobe has stated it will be significantly faster. Right now the M2 Ultra has the fastest scores there, even beating out an Nvidia RTX 4080 GPU. It's likely an Nvidia RTX 4090 would do a little better, but that's the top of the line on Nvidia, so Apple can go toe-to-toe, which IMO is pretty surprising. It's good to note that Adobe's denoise is optimized for the tensor cores in the Nvidia GPUs so Apple is handicapped in comparison at the moment. One final note on the Adobe denoise is that it showed one of the biggest jumps in performance on the M2 vs. M1 generation comparison on Apple Silicon. In addition, Apple has advertised that the neural engines are 40% faster on M2 vs. M1 so I expect an even bigger boost when Adobe is able to leverage those after Apple's bug fix.
2) Staying with denoise, I did the same batch in Topaz as before. Because they are moving to Topaz Photo AI, they haven't updated the standalone in a while. It seems it has a scaling issue as the performance seems to cap out at about 6-7 seconds per file. I can't be sure of this, but I think it's slow on the DNG creation, and all of the machines seem to take a similar time for that process. I say this because the progress bar on processing moves at slightly different speeds across these, and then they all hang up for a few seconds at the end. The odd score for the Mac Pro is because Topaz has a bug that causes performance to slow down when there's multiple GPUs. The fact that the M2 Max MacBook Pro came out a little faster is odd; I'm not entirely sure why that is the case, but it's repeatable. It's not a huge difference though.
3) Preview render times are pretty self-explanatory. The big old 2019 Mac Pro does well here because LR mostly uses the CPU for preview rendering, and it's got more cores and those cores run at a faster base clock speed. The odd scores here at the two AMD Threadrippers which *should* do better, but don't for some reason. Watching the CPU utilization its clear they are stalling out and not fully utilizing the cores. It must have something to do with the AMD architecture on these processor models, but I don't know enough to speculate more. What is obvious is that core count is mostly king here--the more you have and the faster they are, the faster it goes.
4) Finally, jpeg export. This is another area where core count matters, but here it's both CPU and GPU as Adobe uses both. The M2 Max laptop didn't perform as well as one might expect here, and I think that's a clear case of thermal throttling--Apple isn't letting the SoC run as wild as it can in say a Mac Studio. I think the numbers the Ultra Studios post are very impressive considering the competition they are stacked up against here. These are powerful expensive PC workstations.
The M2 wasn't intended to be an M1 replacement, just the next iteration in the cycle of releases, but Apple has integrated some fixes into it that help with GPU and media engine scaling. I've seen some early benchmarks on video processing that show huge performance gains. I'm photography oriented so that matters less to me. If you have an M1 you have nothing to worry about with the M2 release. I do think we will see a big jump with the M3 as that is moving to a new processing node and the rumor is the it will bring hardware raytracing to the GPU.
I am eager to see Apple get the software bug in Ventura fixed so we can fully utilize the power of the neural engines on Apple Silicon. As more and more software becomes AI-optimized I think Apple is going to have an edge here with that technology already tried and tested and with access to the same pool of memory at the same speed as the CPU and GPU.
One other observation: The 12.4 release of LR really did improve the overall fluidity of browsing the filmstrip, but even so this is an area where the Ultras (M1 and M2) just seem to shine, probably because of the GPU core count--it's just super fluid browsing, even if I have not rendered previews beforehand. Between the performance updates by Adobe and the performance of these machines I can say LR Classic has never felt so fluid and enjoyable to use.
|
