Explore related tutorials & articles
29
Replies
Some have suggested that an M1 Ultra chip does not make a difference over M1 Max for photoshop? And that getting more than the basic 10 cores is overkill.
By @samlarson
The reason is that if you look closely at the differences between the M1, M1 Pro, Max, and Ultra, there is really no major difference in the processor frequency. They are all more or less the same single-core speed. The main gains you get are in the number of cores and GPU power, and some other things that don’t apply to Photoshop such as hardware video encoders.
But in Photoshop you process just one image at a time, so that naturally limits how much use you can get out of lots of cores. Lots of cores are great when you need to push through lots of operations that are easy to parallelize, like rendering thousands of video frames, exporting hundreds of photos from Lightroom Classic, or rendering hundreds of vector objects and textures in a 3D application. But as a 2D photo editor, Photoshop just can’t put that many cores to work when you edit one document with some layers.
Yes, the current version of Photoshop does take more advantage of CPU and GPU cores than in the past, which is why Photoshop has a new Multithreaded Compositing setting in Preferences. That should help if your documents have many layers. But if you buy an M1 Ultra, Photoshop is still going to leave more CPU/GPU cores unused compared to what a video editor or 3D application are able to put to work. For large documents, you want more Unified Memory (probably at least 32GB) to make sure both Photoshop and the GPU have the memory they need for optimal performance, and must make sure there is lots of free space on the SSD you assign as Photoshop scratch.
Photoshop has been adding GPU acceleration to more areas, but again, not as many areas are GPU accelerated compared to video editing and 3D applications. So again, the benefit of upgrading to the more powerful GPU in the M1 Ultra may have only limited benefit in Photoshop.
Gains in processor frequency come with new generations of Apple Silicon. The M2 has somewhat faster single-core performance than the M1. But the M2 is currently available only as a basic consumer processor; Apple has not yet released M2 Pro, Max, or Ultra variants yet.
But all Apple Silicon Macs will probably feel faster than an Intel-based iMac, the Mac Studio should definitely feel faster.
You don't have to use an NVMe; it's just that they are so much faster - so fast, in fact, that there's very little practical difference between RAM and scratch disk. No more scratch disk bottleneck.
In any case, that's several orders of magnitude faster than any external drive. You will notice that.
I'd recommend you go to work on your system drive, and do a radical cleanup. Move everything you can, delete as much as you can. There will be a lot of junk in your user account, that's where most of it accumulates. If you can clear 500 GB or so that's a good start, Set that as primary scratch and whatever else you have as secondary.
A "standard" configuration of OS and applications should't take up much more than 90-120 GB. The rest is potentially free space.
For Windows there is an excellent little utility called WinDirStat that shows you graphically what fills up your drives and where it is, in detail. There is a Mac equivalent, I just can't remember what it's called.
Thunderbolt 3 and 4 have a maximum theoretical throughput of 40 gigabits per second, which translates to 5000 megabytes per second. If you cut that down to a real world data rate (after overhead, other devices on the same bus, etc.) it might be around 2500–3500MB/sec. That’s the throughput of a lot of affordable NVMe SSDs, so if you buy one of those, you might get the speed you paid for.
The latest generations of NVMe, including the kind inside the Mac Studio, are about twice as fast as that, out to a theoretical 7000MB/sec. Thunderbolt currently can’t go that fast. So if you get an NVMe SSD for external use, you don’t need to buy the fastest type unless it’s no problem to afford it.
Would (NVMe) make sense if it's not an internal drive?
By @samlarson
Despite those limitations, it’s still worth using NVMe for an external drive. The reason is that if you don’t get NVMe, the alternative is SATA. The problem with SATA is that it tops out at only about 500–600MB/sec…less than 10% of the top speed of a Mac Studio internal SSD. For about the same price you can get what are now “low-end” NVMe SSDs with throughput of 800–950MB/sec if connected by USB 3, or over 2000MB/sec if connected by Thunderbolt 3/4.
In other words--no significant time difference. Am I right to conclude that as far as opening and saving files it doesn't matter so much what kind of drive my files are on? So spend $$$$ on a fast scratch disk, but save $$$ on my work drive?
By @samlarson
That could well be the case. Saving and opening is very different from scratch file use. The scratch file is like a cache for what’s in working memory, so the closer the scratch media speed is to memory speed, the better.
But Photoshop is not very fast at saving a single document to a file. When watching storage throughput during a large Photoshop file save, I’ve never seen it go over 400MB/sec, which is slower than even SATA. (I haven’t measured it lately so feel free to try. Watch core usage and disk throughput during the save, in something like Activity Monitor, iStat Menus, or Stats.)
One explanation given in the past is that saving is still single-threaded, which means you can buy as many cores as you want but saving will use only one. But I’m not sure if that’s still true.
Like D Fosse said a lot of file saving is the CPU getting the bits arranged in the proper format for the file type (and compression if used). With that overhead, document saving can’t be a straight write to storage like an abstract speed test would be, and may never hit the maximum speed of the storage medium being used. That said, some other applications seem to be faster at saving the same file types, so something is not quite right.
"One explanation given in the past is that saving is still single-threaded, which means you can buy as many cores as you want but saving will use only one. But I’m not sure if that’s still true."
According to PugetBench PugetBench List (pugetsystems.com) saving is not the same for all systems. Two of 19 things they are testing are opening and saving file. I can not get conclusion how much saving speed depend on disk/CPU because they are not providing info about disk drive. Only CPU, GPU, RAM, OS and Motherboard are clearly stated. Also RAM manufacturer is missing including latency. I am posting this because there is plugin for Photoshop that can be used for test. On their site one can compare results and get some conclusions since they are testing machines using 100MP file.
One explanation given in the past is that saving is still single-threaded, which means you can buy as many cores as you want but saving will use only one. But I’m not sure if that’s still true.
By @Conrad_C
Not single threaded, but certainly not using all cores equally.
15 GB uncompressed PSB
NVMe generation 4 at nominal speed 8000 MB/s.
Save completed in 23 seconds
But in any case, Conrad is right - saving a file is not at all the same as scratch disk operation, so in a sense this is apples and oranges.
"
15 GB uncompressed PSB
NVMe generation 4 at nominal speed 8000 MB/s.
Save completed in 23 seconds"
It will be interesting to see your PugetBench result. Here is 17 12700K save and open score.
I do not know. Plugin starts with their own image then resize it to 100MP I believe and later everything is done using that 100MP image. The rest is unknown.
Here is one good 11700KF bench with high score.
AdChoices