Topaz Capabilities on Mac Computers

General Discussion Cameras and Equipment
1

Hopefully this post is appreciated and in the preferred location within the community. I’m putting this post here as it’s the only place I’m seeing benchmark comparisons. I have tried nearly everything I can to get an Apple computer (MacBook, Mac mini, Mac Studio) to work well with the Topaz suites. Below are some lessons learned that hopefully can help those trying to match a computer to the software while staying in the MacOS environment. First the computer specs, then the insights:

Computers Tested:

#1: Macbook M1, 8 CPU, 14 GPU, 16 GB RAM, 512 GB SSD
#2: Mac Mini M4, 10 CPU, 10 GPU, 16 GB RAM, 512 GB SSD
#3, Mac Mini M4 Pro, 12 CPU, 16 GPU, 24 GB RAM, 1 TB SSD
#4: Mac Studio M1 Ultra, 20 CPU, 64 GPU, 128 GB RAM, 2 TB SSD
#5: Macbook M4 Max 14 Inch, 14 CPU, 32 GPU, 36 GB RAM, 1 TB SSD
#6: Macbook M4 Max 14 Inch, 16 CPU, 40 GPU, 64 GB RAM, 1 TB SSD

I also tested my brother’s laptop with a NVIDIA 3070 GPU (other specs not known). It crushed the first 3 tested and beat out the last 3 (4 through 6).

Overall:

I’ve listed the computers in order of performance with #6 being the highest performance specifically to my experience in the 3 Topaz products. Move down the line to increase performance.

Running Video AI, Generative models in Gigapixel, or Superfocus Beta in Photo AI will max out the GPUs on every computer listed.

Every computer consistently ran into issues with Topaz software. Most of the issues documented on this site were ones I experienced (force quitting, “film negative” rendering, non rendering, B&W rendering, etc.). I’ve concluded that Topaz simply isn’t a “stable” product on Macs. I regularly uninstalled and reinstalled apps, removed cached folders and files, and installed older versions of the software as temporary solutions to the consistent problems. The problems keep coming back after consistent use.

The capability of the software is phenomenal, and to me, worth the money spent even with the consistent issues had. Move to a PC if you’re not okay with the bugginess.

Computer Insights:

M1 vs M4 Chip: the M1 struggles even on the M1 Studio. However, the M1 studio did far better than M1 MacBook. If choosing between M1 Studio and M4 Pro, they were near equals. I’d recommend going with the M4 Pro over the M1 studio based on my experience. If I had to guess, I’d guess the M4 chip holding other things equal is 2x the performance of an M1 chip in Topaz.

M4 Macbook vs M4 Mac Mini: I did not notice a performance difference.

M4 vs M4 Pro vs M4 Max Chip: any differences identified had to do with the GPU based on my user experience.

CPU: Honestly, I never noticed CPU being an issue even at 8 CPU.

GPU: Other things held equal, 25% more cores will give you 25% more efficient render times. When purchasing a Mac, simply evaluate if the extra cores as a percent increase is worth the additional money and time savings. 20 to 40 cores would double your speed for example but don’t forget that the chip matters.

RAM: When working in a single Topaz program, there is very little benefit to purchase RAM above 48 GB but there were notable gains up to 48. When working in 2+ programs (I often render in Gigapixel and Video AI at the same time), 64 GB RAM is very helpful but additional RAM did nothing for performance.

SSD: you do not want your SSD near full for smooth operation, especially if you skimp on RAM. Be sure to allow for 25% free space at 512 GB or 1 TB, 15% at 2 TB to ensure this doesn’t impact performance.

Power Management: When running the two M4 Max MacBooks side-by-side, I accidentally learned that the 32 GPU 36 GB Ram option in high power mode was nearly identical to the 40 GPU 64 GP RAM option when in normal power mode. Putting the 40 GPU computer in high power mode effectively boosted performance by 20%.

Topaz Insights:

Topaz Gigapixel: When working with the Basic Models in Gigapixel, all computers tested were up to the task but most consistently had issues in the Generative Models. The M1 MacBook struggled too much for my satisfaction on even basic models.
I tested my brother’s laptop with a NVIDIA 3070 and it was faster than every Mac option I tested and by a fair margin. If you want to use Topaz in near-real-time, no Mac will give you this joy today; go the PC route.

Topaz Photo AI: With the exception of SuperFocus Beta, all computers got the job done with increasing performance as you moved from #1 through #6 computer. For Super Focus Beta, all computers are SLOW; get a NVIDIA GPU if you want image renders faster than every 10-20 minutes (based on roughly 1k x 2k photo with 2x upscaling).

Topaz Video AI: It’s all about the GPUs baby! Frame rate production appears to be a linear relationship to the number of GPUs, holding other things constant.

Topaz is a great product with poor buggy performance on Macs today when doing anything beyond the basic models. It’s worth the money to me though and I appreciate the solution Topaz has brought to the market. I’ve tested other software solutions for both photo and video rendering and find Topaz to be the best when it works.

3 Likes
2

Hi Dave - this is a very useful bit of research -thank you for posting it! I am hesitating on the purchase of an M4 Pro Mac Mini after reading this since it likely will not resolve my issues with Gigapixel AI. On a 2017 iMac with 40gb ram, the program struggles to render a single raw image from LR. I mostly work in LR, PS and would like to dabble in video. I agree with your assessment that Topaz does an amazing job and I can confirm this having rescued a very low res image of my grandfather that alone made it worth the purchase of the software. I am now second guessing the plan to spend about $2K on the Mini Pro just to address this issue since LR and PS work just fine on my current 2017 rig. I’m thinking I may wait for the M4 Studio and just tolerate the slow process in Gigapixel . Thoughts?
Mark

3

I’ve adjusted my workflow so that I preview things during the day and render over night. While slower than desired, this meets my needs and keeps me in MacOS. If you wait for the M4 or M5 studio, you could likely double your specs and cut your render times in half. This will still be slow but less so.

4

Any Apple Silicon computer will be considerably faster than those old intel based machines and their awkward built in GPU.

And consume a lot less power, be cooler and more quiet at the same time.

So you definitely might want to consider the switch.

1 Like
5

Ok - thanks!

6

Hi Mark, the best solution for your problem is the first run the Camera Raw straight to the Topaz Photo Ai and let Topaz process and save them as DNG’s. Then you take those DNG’s to the lightroom. I have been using Topaz Photo Ai, Video Ai and Gigapixel from last three years and I have always got best results this way. DNG’s are equally good as Camera Raw for editing and long term archival purposes. Nobody can ever go wrong when using DNG’s. Best idea is to run the Camera Raw files overnight and your mornings will be smooth.

1 Like
7

This was a lot of work. Thank you for sharing it with the community! :grinning_face:

And so many machines at your disposal. Very insightful. From what I read anecdotally, it seems the findings have held true over many versions of Topaz software. PC’s rule. And Nvida has an edge. Don’t skimp on VRAM. RAM matters, but only up to a point. 64GB to 128GB gets you almost nothing…from what I read.

8

I never even thought of that. I have been living in a TIFF work for 25 years and just now finding the new formats abound, and some with notable benefits!

9

I helped a friend spec a Apple Studio M4 Max with 96 GB. I was not too happy to hear he didn’t get much of performace boost. He is using Topaz Gigapixel for upscaling and Affinity for hand coloring images of the Panama-California Exposition of 1915 for a non-profit organization.

He might spend 100 hours on a single image.

10

As someone posted above, I also managed to upscale a large number of old JPG scans overnight using generative models. On my basic M4 Mac mini this took several months on and off, at approx 20 minutes per image. Going forward, I guess Mac users who want full access will have to go with the new subscription model and do their rendering in the cloud.
Meanwhile, I can now upscale locally using On1 Resize at approx 90 seconds per the same low res image.

11

Of course, if someone is putting that much handcrafted love and detail into those old images the speed of the AI isn’t coming that much into account.

Still regarding the sheer processing time of TPAIs models the M based machine will run circles round the Intel one - even my M1 Pro MacBook Pro is WAY faster than the latest state of the (Apple) art Core i7 I also have here. And using much less power.

A M4 Max should be somewhere comparable to a RTX 4070 performance-wise.

Of course, only looking at TPAI the basic 32GB variant would’ve been fully sufficient.
You don’t get more speed with bigger RAM here - and RAM unfortunately is extremely expensive with Apple.

1 Like
12

Very helpful, jo.vo. Thank you.

1 Like
13

I think it comes close to a RTX 4070 in terms of CUDA cores. Not AI. AI cores on mac are fixed and defined by Neural engine. Only the Ultra has double the Neural cores. I think the M5 looks more like Nvidia GPU where AI/RT cores are behind the cuda cores like a pipeline and scale up when you have more GPU cores. Speed is not defined by VRAM but also AI power. And there also AMX coprocessors in the M series but i have no idea if they are used by Topaz Video. They can do matrix calculations too.

14

Topaz software doesn’t really use the Neural Engine of AppleSilicon very much, but mostly the GPU cores.

(And in certain cases when it does it’s even better forcing it not to do so on machines with high GPU count)

15

Thanks! I just read 7.1.5 does use the neural engine. If my info is correct the Neural engine and AMX coprocessors can do the fastest matrix calculations in the M chips. And it would be weird for an AI program not using them but i have no idea how the program really works.

16

A boost over what? I went from a 2019 iMac w/64GB RAM and 8xi9 CPUs and a boring Radeon GPU to a Mac Studio M4 Max with 64GB RAM and huge number of CPU/GPU cores. My Topaz Photo AI and Gigapixel AI (and Photoshop and Lightroom for that matter) SCREAM on my machine. I can export a Raw from Lightroom Classic to Photo AI, let autopilot do its thing, export the results back to LrC, all in less than 10 seconds per image. I don’t do video, but for photos the M4 Max is a brilliant machine.