Performance tuning Adobe Lightroom classic

When I was deciding on what to upgrade my main machine to, performance of Lightroom was one of the considerations. But what do you need to consider? Adobe has a page on the subject but it lacks quantitative numbers. Well, with that I decided to run some experiments. I was coming from an older Lenovo T480s Core i5 5300 dual core, hyper threaded, 8G RAM with an SSD. It was clear I needed some more bang for the buck as I became more efficient with Lightroom.

For the purpose of these tests I am going to use a Lenovo L14 Core i7-1165G7 quad core, hyper threaded @2.8GHZ with 16G of RAM and an SSD, running on Win 11, as well as a VMware VM running on Xeon E5-2660 V2 @2.2HGZ with the VM on an SSD running Win 10.

It’s worth noting, I am a basic Lightroom user, and have not as yet delved into some of the more complicated things it can do, so this article and these tests are written from that perspective.

Memory

For starters Adobe recommend a min of 12G, so obviously this was an area to focus on. I ran a few different tests starting at 16G, and 32G and found Adobe happily makes use of the larger amounts of memory but I didn’t see it push a whole beyond 16G. On the laptop it would bump up near the 16G, but in the VM it really never pushed beyond the 16G even when 32G was available. I recently upgraded my laptop from 16G to 32G and I noticed a much smoother experience while editing. I can’t really quantify it, but it was noticeable.

GPU

Most video cards have GPUs, a type of processor that’s REALLY good at floating point and the like. This is largely used by games but Adobe has an article on Lightroom’s use of a GPU … They are somewhat vague as to what actually uses it saying “speed up tasks of displaying and adjusting images”. So to put this to the test I borrowed a buddies powerhouse desktop with an NVIDIA GeForce GTX 3070 which clocks in at 11783 on Passmark’s test well exceeding the recommended 2000 limit. I rarely saw the GPU do much of anything throughout my import, edit and export of photos. So little you shouldn’t spend a dime on it IMHO. On my Iris embedded GPU I have seen no more than a little use of the GPU, which BTW scores in at 1424 below the recommended min from Adobe. To be sure I checked the settings for performance, as you can see the NVIDIA supports “Full acceleration” while the IRIS supports only basic.

LAN

I put my working files on a LAN drive so they can be backed up while they are being worked on. This means Lightroom is pulling them from, and saving them on the LAN drive. To be captain obvious, the speed of the LAN will make a HUGE difference. Wireless connections for example will be a HUGE bottleneck but even wired connections are of course going to be slower than a local SSD. How much slower? Well exporting the files to a LAN drive took 186 seconds with a 16 core VM with the processors being slowed by the network connection. Compare that with doing the same export to a local drive it took 157 seconds or 16% faster. As you can see in the performance graph the LAN is getting pushed but not max’d at 500Mb/s. The files were coming off a spinning drive so this too limited the speed to/from the LAN. This is on a wired 1G connection. The export ends up being around 1.48G in size.

Export to LAN

Export to local SSD

Processor

So now onto what I expected to be the more interesting part of this little experiment, playing with processors. By running on a VM I can easily adjust the number of processors, do a repeated task and see what difference it makes. For the purpose of this experiment I am going to use a directory of 109 RAW Canon images (CR2s) as well as 36 Nikon JPGs for a total of 145 images. I will import the images, manipulate them by batch adding lens correction to the CR2s, and then export them full size, as well as with a watermark imposed on them as well as resizing. The images are coming off a LAN drive

The import on the laptop took 56 seconds pretty much pounding all cores, while a 16 Core VM clocked in at 47 seconds with the cores waiting on the network and not maxing out, as discussed above. So the extra cores ended up about 16% faster.

Let’s look at exporting to a LAN drive from the laptop which clocked in at 295 seconds Vs a 16 core VM at 186 or 37% faster.

So the results of both of these tests tell me that cores help, to a point, especially when saving to a lan. The Quad core laptop processor max’d out. And what I can also say is that the laptop processor bumped into thermal limiting, a topic of another post I recently did. Basically the CPU slows down under heavy load when it is over heating. It’s a lot harder to cool a CPU in a space constrained laptop than a desktop. On the other extreme, at 16 cores on the VM (it’s running on a 10 core hyper threaded so some of those are not physical cores) the CPU did not max out and exceeded requirements. Again, saving to the local local SSD and pulling from the SSD would likely increase the CPU utilization significantly IMHO.

So … if your buying a new machine, cores are king, to a point, don’t spend on a GPU for lightroom, and make sure you have LOTS of memory. If your workflow can tolerate finding a way to have your files on a local SSD this can definitely help.

Update: I was reminded of a synthetic benchmark written for Lightroom, that runs as a plug in called Pugetbench. You can look into benchmarks. There’s an old saying benchmarks don’t lie, but only liars use benchmarks … I digress.

So let’s look at Pugetbench and compare two Core i9 results, one using Intel’s discrete video and one using an NVIDA RTX 4070, a card that’s worth $800 BTW:

Compare with video
i9 discrete video	1690.5	91.7	246.4	Results
i9 NVIDIA RTX 4070	1689	88.6	249.2	Results

As you can see the NVIDIA made NO difference in the benchmark, which agrees with what I had previously seen.

Let’s compare some processor results:

i7-1165G7 4 core hyper thread 8 threads laptop processor	634.5	67.1	59.8	Results
i7-8700 6 core hyper thread 12 thread	795.2	67.3	91.2	Results
i7-10700 8 core hyperthread 16 thread	916.6	85.4	97.9	Results
i9-12900K 8P 8E 24 threads	1402.5	108.2	172.3	Results
i9 13900K 8P 16E 32 threads	1665	98.1	235.4	Results

What you can see is this benchmark is all about threads. Interestingly enough, the efficiency cores of the new processors hold up ok. I include the first processor, because it’s what’s in my laptop. So let’s do some comparisons:

First and second line in the table … going from 8 threads to 12 is 25% faster

First and third line in the table … going from 8 threads to 16 is 44% faster

Neat stuff … and while it’s obviously not a straight line, 50% more threads only yielded 25% faster, and 100% more threads only yielded 44% faster.

BTW this benchmark does NOT seem to cover off Macs, not even Intel based …

This article is a start, and by no means complete, or thorough, I need to do more digging/testing, but for now I’ll leave it at that.