jhapeman
Offline
• • • •
Upload & Sell: On
|
OK, so I have the good fortune of having two new Mac Studios. I was on vacation until Sunday so I just got them set up and ran a set of basic tests. I also have two 2019 Mac Pros--a 16-core with dual Radeon Pro Vega II MPX modules and a 28-core with a W6800X MPX module. For travel and home use I have a 16" M1 Max MBP with 64GB of RAM. Finally, I have a 64-core AMD Threadripper 3990x workstation with dual Nvidia RTX6000 cards. As part of my business, we do a lot of 3D rendering, hence some of the big iron, but also a lot of studio photography shooting hundreds of images to focus-stack and/or just edit to upload to our ecommerce website. Faster processing is always a plus in these cases, and they are business expenses, so they are written off.
I needed a minimum of one new computer for a new hire coming up, so moments after the launch I ordered a loaded Studio Ultra with 8TB of storage and 128GB of RAM. After sleeping on it, I went to order lower-spec Ultra and delivery dates had already drifted out three months. Arrghhh! Luckily I was near an Apple Store on the official launch day so that morning I logged in and reserved a base model Ultra for pickup, and managed to grab one before they quickly sold out as well. I had to schlepp it home from vacation on the plane, but with the small box with a built-in fabric strap handle, it was a piece of cake.
First, some basic observations. It's much more attractive than it seemed online. A beautiful dense little cube of computing power. It's really quiet. Not silent--but you need a really quiet room to hear it. It's dense but not super heavy and has a nice heft. Lots of ports, and they're fast--I imported 1465 Sony A1 raw files with minimal previews in just 2:40. That's about 30% faster than the Mac Pro, even though they are all Thunderbolt devices, and the Sony card reader is only USB 3.2, not USB 4.0.
Speaking of speed, now on to some tests I ran with those 1465 Sony A1 raw files. I use lossless compressed raw, so that has some overhead although I'm not sure what--it's irrelevant since I used the same files on all of the machines. On to some graphs. They will all be below, so you might have to scroll back and forth. These are pretty fascinating and should be helpful for those considering the various models of the Ultra.
Rendering Previews: Both the 128GB/64GPU and 64GB/48GPU perform identically. Not surprising as this is mainly a processor-bound task and both have identical numbers of cores that run at the same frequency. What is surprising is that they cleaned up the competition--they both outperformed even the 28-core Mac Pro and the 64-core AMD TR workstation. The slowest was the 16" MBP, and it did get hot and the fans ran (in its defense, the 28-core and 16-core Mac Pro also spun up their fans, and the AMD sounded like a jet engine). Both Studios did spin up their fans slightly but it was very faint and subtle. This is one of the only tests where the M1 Max in the MBP did not scale linearly in performance--my guess is that thermal throttling was at play there.
Exporting full-sized JPEGS: Once again, the two Ultras performed identically, and while the AMD was a slight winner here, it wasn't by much. Also keep in mind that the AMD processor alone cost the same as the entire base model Ultra. Both Studios beat the top of the line 2019 28-core Mac Pro. That's remarkable.
Exporting TIFF files to Topaz DeNoise: I use Topaz DeNoise and Sharpen fairly often, although DeNoise the most often. I'm running the latest beta of DeNoise, which is close to release and fully AS native. The export plugin seems to be constrained to only using six cores, which is how Lightroom Classic used to be for many functions until some recent releases to optimize performance. I don't know if that's still the case for some functions or if Topaz is using some legacy export functionality, but it does limit the scalability of this function as the core count scales and levels the playing field quite a bit. I tested doing a larger number of files with the M1 Max, Ultra, TR and 28-core Mac Pro and it scaled linearly with file count, while never utilizing more than six cores on any of the machines.
Processing files in Topaz DeNoise: This is another surprise test, and shows what well-optimized code can do and/or the advantages of the unified memory architecture and the way the GPU cores are presented to the application when written for Metal. I was surprised to see the Studios beat even the workhorse Nvidia RTX6000. Once again, this is a video card that costs more than the base Studio Ultra. Apple has a reputation for being expensive, but these machines provide amazing power for the price point. It also shows how synthetic benchmarks don't mean a lot compared to real-world results.
Bottom line: Even for large workloads the base Ultra is an amazing workhorse and more than most will need. I suspect the only advantage for a photographer of 128GB of RAM would be if you are doing massive panoramas or HDR merging in LR and/or PS. I'd like to say image stacking as well, but unfortunately HeliconSoft has only limited AS support right now, and uses the deprecated OpenCL functionality, while Zerene is built on a Java Runtime Engine and in my testing was always slow as a dog as a result. I'm hoping that one of them will finally get around to updating their software but I'm not holding my breath.
I'm happy to answer any questions anyone might have and if there's a reasonable test you'd like me to try, I can see what I can do. My personal recommendation is if you can spring for the Ultra, do so for the base model but unless you've got extra money needing to spent or massive needs for memory I'd suggest the 128GB of RAM and anything more than 48 GPU cores are unnecessary.
|
