I’m very meticulous when investing in new hardware and do a lot of research prior to purchasing. Once I buy something I like to understand the improvement in practical terms. I usually run performance benchmarks, but this time I went further with more tests and some graphs. These aren’t the fairest comparisons since the two machines use different SOC classes and have vastly different amounts of memory, but they represent the upgrade I’m getting. Let’s see the unfair comparison between my MacBook Air and my new MacBook Pro.
|MacBook Air (2020)||MacBook Pro (2023)|
|Processor||M1 – 4 efficiency, 4 performance cores||M2 – 4 efficiency, 8 performance cores|
|Graphics||8 graphics cores||19 graphics cores|
|AI||16 core Neural Engine (11 trillion ops/sec)||16 core Neural Engine (15.8 trillion ops/sec)|
|Cooling||No fan||A very quiet fan that barely comes on|
|Memory||8 GB LPDDR4X (4266 MT/s)||32 GB LPDDR5 (6400 MT/s)|
Geekbench is a nice synthetic benchmark to test raw performance and I like to baseline the hardware. There isn’t much surprise here from a CPU performance perspective. The single core performance of the M2 and its siblings is about 10% better than the M1. Some of that translates directly into twice the multicore performance, but the big difference here is that my MacBook Pro has twice the performance cores as my MacBook Air. The efficiency cores are also better performing than those in the M1.
The 2.3 – 2.6x boost in GPU performance isn’t unexpected either. The M2 Pro has more than twice the number of GPU cores of the M1 and each core is around 15% faster. These are great results and aren’t surprising; the M1 is a consumer design and M2 Pro is a performance design.
Synthetic benchmarks are great, but what about real-world results related to what I spend much of my time doing: Lightroom. It’s difficult to test the performance of fluid activities like editing, scrolling, and loading RAW files, but overall my Pro has been really, really smooth. Editing is snappy, especially complex masks, which don’t drag things down after a few images like they did on my Air. The speed comes from the M2 SOC and the fluidity by the 4x increase in memory. Lightroom is what I’d call a goldfish when it comes to memory – happily growing to consume as much as it can. On my system that can be upwards of 22 GB.
All that said, there are some parts of my workflow that I can test – things like imports and exports. So let’s look at those.
For the import test I imported 323 36 MP images from my D800, generated standard previews at 2048 pixels medium quality, and generated Smart Previews. There actually isn’t much difference in this test. I never found my Air slow at importing but I expected a bigger boost from having 4 extra cores available. I think the bottleneck might have been the SD card as both machines would import a batch of images, generate previews, and then idle for a few seconds before more images were ready. I might have seen a bigger difference with higher resolution Standard Previews at higher quality as they would stress the CPU more, but I tested what I use in my daily workflow.
The second test exported those 323 RAW images as JPEG files at 90% quality. In this case the extra power of the M2 really showed as I had expected at about 3.5x faster. This test was done without Lightroom’s recently-added GPU acceleration. Enabling GPU acceleration shaved 27% of the time from the MacBook Air and about 30% from the Pro. I’m surprised this doesn’t provide more of a boost – it either shows how strong the extra CPU cores in the M2 Pro are or it shows that Adobe still has some optimization to do with its GPU acceleration. I’d expect 19 GPU cores to be more than 30% faster than 12 CPU cores. Either way, a boost is a boost.
DXO Pure Raw
I use DXO Pure RAW to remove noise from my images and it’s really good but also compute intensive. I own version 1 which is an Intel build running under Rosetta 2 and isn’t optimized for the Neural Engine. It relies solely on GPU cores for processing. I downloaded a demo of version 2, which is Apple Silicon native and does take advantage of the Neural Engine. The results were surprising and I actually discovered a “bug”.
For these tests I ran Deep PRIME noise removal on a batch of 17 Canon R5 RAW files (45 MP) and 1 Nikon D800 RAW file (36 MP). The results for Pure RAW 1 are as expected, with the Pro coming in 2.7x faster than the Air due to its more than double GPU cores. Interestingly I didn’t see any memory pressure issues in this test even though the Air only has 8 GB to work with and it’s shared across the CPU and GPU.
The second test was rather surprising with the Air finishing about 4% faster than the Pro. That’s very confusing since Pure RAW 2 uses Apple’s Neural engine to accelerate processing. Both the M1 and M2 have 16 core Neural Engines, but the M2’s can complete about 50% more operations per second. If that’s the case then the Pro should be faster, right? It turns out that Pure RAW 2 running on an M2 Pro does not use the Neural Engine.
I confirmed this through Apple’s powermetrics tool, noting that the Neural Engine didn’t draw any power on the M2 while it did on the M1. I reached out to DXO and they indicated that they were planning to add support for it in the future. I find that a bit odd, since Apple’s framworks would automatically support new hardware, but perhaps DXO is doing something custom. Regardless, a “fix” is coming.
Despite this, Pure RAW 2 is still about 15% faster than version 1 due to it not running under Rosetta emulation. I’ve considered upgrading to version 2 for its faster performance, but I’m going to hold off until the Neural Engine is enabled for the M2 Pro. The boost I get on version 1 is plenty to keep me happy right now.
The final test was to transcode 4K videos into the H.265 compressed format using Handbrake and QuickTime. There was a drastic difference in the Handbrake test, with the M2 completing the work a little over twice as fast. QuickTime was a different story, with the Pro only beating the Air by about 3%. I think that has to do with the way QuickTime handles transcoding – I barely noticed any CPU or GPU utilization on either machine.
I’m happy that I took a bit of time to formally test my new machine. I discovered an interesting behavior with Pure RAW and I have baselines that I can use to test upgraded versions of my most common tools in the future. Overall, my MacBook Pro is meeting my expectations. It will be really cool to compare these results to my next Mac in the future.