Topaz Giga AI, Sharpen AI, Denoise AI & Video AI Apple M3 Gen Test.
HTML-код
- Опубликовано: 10 июн 2024
- Topaz Giga AI, Sharpen AI, Denoise AI & Video AI Apple M3, M3 Pro, M3 Max test on 14 & 16" MacBook Pro. Using both RAW and JPEG.
Base M3 Max vs Top M3 Pro - Justifications for a $1000 Upgrade!
• Base M3 Max vs Top M3 ...
Top M3 Max vs Base M3 Max 16" MacBook Pro, is the max worth $500 more?
• Top M3 Max vs Base M3 ...
16 MacBook Pro Top M3 Pro vs Top M3 Max is $1500 worth it?
• Top M3 Pro vs Top M3 M...
Base M3 Pro 11/14 vs Top M3 Pro 12/18 - $400 Upgrade - Results will surprise you.
• Base M3 Pro 11/14 vs T...
M3 vs M2 vs M1 | Pro Photo Workflow Test - Good improvement & "Pro" SoC wannabe
• M3 vs M2 vs M1 | Pro P...
M3 Pro vs M2 Pro vs M1 Pro | Pro Photo Workflow Test - Disappointing performance?
• M3 Pro vs M2 Pro vs M1...
M3 Max vs M2 Max vs M1 Max | Pro Photo Workflow Test - This Rival the Ultra!
• M3 Max vs M2 Max vs M1...
Does memory bandwidth matters? 32GB on M1 PRO vs M1 MAX
• 32GB on M1 PRO vs M1 M...
Should you buy the fastest SSD for Pro Photo + Video Workflow? Does internal vs external matters?
• Should you buy the fas...
Ultimate Calibration Guide for Apple 14" & 16" MacBook Pro XDR Display - This is the way!
• Ultimate Calibration G...
iStat Menu bjango.com/mac/istatmenus/
Looking back - Pre M3 release videos
My thoughts about the new M3 Macs, Overall configuration advise
• My thoughts about the ...
M3 Macs - A few additional thoughts before the official ship date!
• M3 Macs - A few additi...
Battery Test on M1 Pro vs Max
• 14” & 16” MacBook Pro ...
00:00 Preface
00:46 Test Systems
01:54 Reading Spec
02:05 Test Images
02:45 Settings
04:24 Sharpen AI Aurora
07:31 Sharpen AI Pano
10:00 Denoise AI Aurora
11:27 Denoise AI Pano
13:39 Giga AI Aurora
15:18 Giga AI Pano
17:09 Video Enhance AI 4
19:17 Conclusion
Art's Tip Jar
www.paypal.com/donate/?busine...
Full Disclosure:
ArtIsRight is a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a way for websites to earn advertising revenues by advertising and linking to amazon.com
Some of the links below are affiliate links. This means that, at zero cost to you, I will earn an affiliate commission if you click through the link and finalize a purchase.
#Topaz #m3Pro #m3max 
Наука
Every Mac-using photographer and videographer needs to have subbed to Art's channel.
Thank you!
Finally the video I have been waiting for. Thank you, Art🎉
You are so welcome!
These tests you are doing are really insightful! Thanks Art!
My pleasure!
You confirmed that I made the right choice buying a maxed out M1M instead of an M3. Great video.
Nice 👍
Love how analytical your work is! Thank you
I appreciate that!
Thank you for taking the time to do this!
I think Topaz apps in general are not the best when it comes to hardware optimization, especially on Mac. They really need to improve it, I was blown away when my 3070 was much faster in Video AI than the M2 Max with 64GB Ram.
any comparison if you don't mind sharing? I am using the M2Max with 64GB RAM and considering whether or not and which model to upgrade to.
@@joyKafka frankly I returned the laptop months ago, but again, my 3070 PC was faster. If a video (dunno the settings) took 15 min on my PC, it would take maybe 25 on the Macbook. In general if you're going to use the laptop for AI stuff, I think you're better off with Nvidia stuff (not that you can't with the macbook, but most apps I found are better optimized for Nvidia Cuda and Tensor cores)
@@agiverreviga4592 Thank you. The advantage using M3Max for AI in my case could be the 96GB or 128GB unified RAM for GPU, plus its neural engine. For media and video editing, I enjoy the current dual and multiple monitor environment, mobility plus the well built ecology. I am thinking to upgrade to 14/30/96 mainly because it seems the 64GB RAM is already a constraint when the RAM is divided among wired, compressed, cache, app and GPU demands. Take Davinci resolve for example, the max RAM DR would use is 48 GB out of the 64GB RAM (GPU usage would be only a portion of it), and usually it would be around 20ish GB since a max of 32 GB is reserved for Fusion (VFX effects) cache. Sadly there is no 14/30/128 option. The other reason is the neural engine and 45% more transistors. This neural engine seems to make M3Max a better machine than MS M2Ultra.
This would be an interesting test to do on the Razer Blade 18 that I have.... hmmm... thinking....
Big thanks for testing!
You bet!
That’s was fantastic. A big help in deciding what system I should go with. I use Gigapixel daily in my workflow.
awesome
I very much appreciate all your videos and was pleasantly surprised to see this one in the mix. I’ll get in line here to request you do a similar video with Affinity Photo as it is gaining quite a following (and yes I use it along with Topaz for specific work on an M1 Studio Max 64/1. I venture to a couple of forums elsewhere and have linked your vids at time as they are practical and useful for both pro and hobbyists.
Thank you.
Good evening from the Netherlands. Thanks again for the clear and good explanation. I followed your advice last week and ordered the M3pro base model. With an upgrade of ram and SSD, as you explained in a couple of other video's. Next week it will be delivered and I'm curious to see the difference with the current i9 MacBook. Thanks!
For me it was a huge leap, hope that it is for you as well
Seeing any differences?
This guy does the most!
FYI Art - you have a rendering issue with your placeholder drop - idk if you care/want to reupload -- still great video!
I know, I saw that after the video launched. However, I am choosing not to replace it to mess up with the stats.
Again great and informative!
Would love to see you do a test with Neat Video denoising, (the performance check/test of the software) as the simple m2 specs bump didn't result into a big difference.
The software is now updated for M3/M3P/M3M, and it might end up being a much bigger jump now.
...but I haven't seen anyone do any tests as of yet.
No idea if this is something you could test, but would appreciate if you could..
Either way, thanks for making all these videos for us contend creators.
You're Welcome!
Amazing test!! Topaz is a heavy boi!
Thank You!
Thanks Art. I pulled the trigger on a M3 Pro with 36gb RAM. Seems to be the best balance in price, power, and battery life.
Watched every Video from Art on M3 Chips.
Read every single Comment left.
Seems like the choices easily are 2-1 from people here buying the Macbook Pro M3 Max Base Model
@@JeffAlanWolf-ij3lsyep can’t go wrong with the M3 Max, it’s the best bang for the buck in terms of power. I didn’t need that extra power and the premium wasn’t worth it for my usecase
Great choice with M3 Pro and 36GB. And thank you @JeffAlanWolf-ij3ls
You deserve 500k subs
Thank You!
Thank you so much, that saved me from upgrading.
You are welcome!
Thanks!
Welcome!
Thank you, Art, as always, for your most insightful analysis and recommendations.
I don’t live in the US so Mac prices are higher and there is no refurbished program here. So for older Macs, they will be 2nd hand ones with few years of use.
Question: If I want to buy a new MBP now (coming from 2017 Intel i5 MBP 13” with 8GB Ram, 512GB SSD) at around $2,000 price point, which M SoC should I get? Spec? Especially if my focus is Topaz Video AI and video editing?
probably M2 Gen, or if you can get the Pro / Max in M1 that would be great as well.
Excellent job on the AI tests. Thank you again for all the hard work and detailed explanation.
In terms of RAM requirement for AI and machine learning, take llama 2 70-billion-parameter models for example, it would need probably 40ish GB RAM to load, and my M2 Max with 64GB cannot load it at all. For deep learning with 4K images, it would require about 160GB RAM for the GPU for a batch size of 16, or 40GB RAM for a batch size of 4. Of course when we are running the AI Denoise or AI sharpen, we are not training the machine to build the model. Instead we are running the already trained model with minimum parameters. That's perhaps why RAM is no issue here. Maybe 8-12 GB Ram is enough. And in this case, I wonder the performance may actually depend more on the whole architecture, the better design for neural engine, or perhaps the number of transistors, since there's no apparent bottleneck here.
In terms of video requirement, for me a video project size can easily go up to 1TB (in ProRes or RAW), and much more when all caches and proxies included. The point to bring the storage size in is I wonder the RAM is already occupied by a lot of other things when processing large and many files, such as app, wired, compressed, cached memory and the memory allocated for GPU. This doesn't count the multi-cam scenarios in yet. But again if the project size is small, the timeline is short, the video resolution is 4K or below, and only one effect is used, perhaps RAM is not critical any more when it reaches a certain minimum.
If we translate the four-second performance of AI denoise/sharpen on the 36MP still image to 1 second per frame for a 4K video (8MP/frame). That is actually, very, slow.
Understandably what you've done is the best strategy to make all other things equal when performing the test and what you've shown here tell us a lot of things. As much as the tests show that RAM is not so much relevant, I fear in real life application, it is still important. Thank you very much again.
I hear you but that is talking about a different model and tasks all together.
@@ArtIsRightthe new dynamic caching is said to be able to allocate the memory more efficiently in a way that the 64GB ram could be as good as 96GB ram with the traditional caching method that will usually generate a lot of wasted cache. This might be the reason why you found the speed is not in proportion to the size of RAM.
most people misunderstood this, dynamic caching is for GPU memory allocation only, not CPU or NPU. and Topaz software never does well. Code base is antiquated
@@ArtIsRight Good to know. Thank you for the input. This is why Apple's unified RAM becomes even more valuable with this dynamic caching technology. Take RTX 6000 Ada with 48 GB Ram for example, it costs about USD 10,000, almost twice the price of MS M2Ultra with 192 GB RAM. With the new dynamic caching, it could be equivalent to 256GB RAM or more, the RAM the size of an SSD.
256GB RAM in a dedicated graphic card is not something anyone could afford but Apple makes it possible. Admittedly it would still be half the speed of the RTX 6000 at its best but this size of RAM could do many things the 48GB RTX 6000 can't do at all, say parameters beyond 70b. This really makes a lot of things (like dreams) possible for many people. Understandably it's still not for everyone, but I am really into it.
For video AI, the number of processes is just the number of videos it will work on at a time, so it doesn’t really have a bearing on your test unless your test involves processing multiple exports.
Good to know, thank you.
Have you tested fan noise and throttling on the 14" 14/30 and 14" 16/40?
I don't have those in the studio. If you want performance get 16, portability get 14
Is MBP M3 Pro base model with 512GB storage is enough for TopazLa? Or I have to upgrade to 1TB storage? Actually what make me wanna buy MBP M3 Pro is because I want to run well the TopazLab softwares, Blender, Adobe softwares, and DaVinci Resolve software.
faster with the M3 Pro, 512GB is fine depending on how you store your files.
Art is Right is among the best and most thorough source for this genre of tech. Consistent, and tone make a huge difference. Thank you Art for your reviews and time spent testing.
You're Welcome!
Could you please do a 18 vs 36GB RAM base M3 PRO?
😀
Art another Informative Video
I know I have asked before but would you kindly give your opinion on:
Your Opinion matters.
External SSD Drive between
Samsung T 7 vs T 9
Both 2TB
Great sales on both
$49 difference
I ordered both and I am returning one
I stayed away from Sandisk because of their design flaw issue (losing data)
Thanks Art!
Jeff, you might have missed what you like to voice my opinion on. Or just dm me
I tried to DM You
Nothing
😕
Could you give us a direct link please?
I did Facebook but nothing happened so it’s probably me art so please understand. Thank you.
I have to say after watching this video that I’m feeling much much better about getting the M1 max with 64 GB of memory over just an M1 with 16 gigs of memory. in most cases at least here and as a photographer, it’s performance is right in there with the M2 and even M3 mid range models, and I saved a bunch of money. Mine does have a one terabyte SSD versus a 2 TB but I would assume that has absolutely nothing to do with performance unless the drive was extremely full?
About SSD not so much, if at all and overall, well said
Are there any differences in speed/lag between the M3 Pros and M3 Max chips in Lightroom Classic editing with high megapixel raw files? Zooming in to 200%, using the heal tool frequently, brushing/masking, etc. My current 13" 2020 M1 16GB has significant lag with my Canon R5 45mp compressed raw files in LR Classic when zooming to 100% or 200%, when using the heal tool over and over. I'll click and have to wait several seconds for it respond and catch up to my input. I'm wondering if the M3 Max is worth the extra money over the M3 Pro in removing lag during these tasks.
Great question
Between the SoC not so much but RAM is a huge factor in LRC. especially with the various localized adjustment tools. The issue with the regular M1 is that they don't age well at all 4 years later with Sonoma. It was really hot SoC when it came out but you can see in most of these test, it is not holding it own any more. If anyone wants longevity get something with the Pro, Max, Ultra suffix. And at this point in time regard of gen. You can see that M1 Max or even pro is doing quite well.
Video Ai tends to make the m1 ultra get very hot wonder how much wear and tear it’s causing to the cpu… will be interesting to see how next years M4 Max performs
nil
Thanks a lot Art. I understand 64 GB is a good choice. The Mac Studio is not available with m3 Max at the moment and might be more expensive with all possible cores. Therefore Studio M1 or M2 is my question: M2max 64/1 or M1ultra 64/1
Same price at the moment
Ultra all the way. Max comes somewhat close but for the most part falls short
Cheers
Also still go for M1ultra? No issues with codecs on the old one?
i have m3 base model, the video ai program run good but when I start export Mac start very high sound, its normal, I don't care about that?
yes normal
Please tell me the length of video source used in topaz video AI?? 🙏
That was an information gap, really.
How close it is to real time...
I mentioned in the Video AI intro 4min 30 sec video
@@ArtIsRightSorry I wasn't listening to that part but focused on the presentation... 😅
all good
1:14 right most M3 max top 32GB. I thought apple only sell M3 max top MBP in 48/64/128 GB? How do you get 32GB?
Probably an error on my charts, I have to check
I just received an email response from Topaz when asking about upgrades and I was told Topaz is NOT going to update Sharpener or Denoise anymore! They are only keeping Photo Ai and Gigapixel going forward. A bit disappointed as I found these two programs often did a bit better job than Photo Ai for their respective tasks.
Wow, thank you for the heads up. This is good information. It looks like I have to figure out what is next in my topaz test; either that or just do giga and lump it with the rest of my standard review/test :)
Well, I went to their webpage last night and I noticed that it still lists Sharpener and Denoise on their product page, but there’s a label on each one that says “now in photo AI”
👍🏼
The more I watch those comparisons, the more I'm inclined to wait for the M4, or simply leave it as is and see what Snapdragon has to offer.
I want to see how that one does
My dude said consider upgrading from m1 or m2….. I work in tech. If you have an m1 or m2 even base with 16/32 gb of ram. Do not upgrade. Waste of money. Just wait 2/3 more years. If you have intel, upgrade now to any m3 with 16/32 gb of ram.
BS. M1/M2 Vanilla are really lagging behind. M1max and upwards, yes don't upgrade yet unless you get any of the m3max
ditto to @PKperformanceEU and I do tech and have a computer engineering background.
@@ArtIsRight i didn't question that. I myself am CS student and from what I see, even the old m1max is massively faster than a vanilla m2, using various of my hand written benchmarks. Both ST and MT as well as SIMDified benchmarks
@PKperformanceEU should have been clearer, that was a 2 part answer, one is to ditto you and the second is a reply to the OG comment. I'll be more clear next time. never thought you doubt me :)
First
nice
You show the speed of Topaz Denoise AI which can use the Apple Neural Engine. Just imagine how much faster LrC Denoise would be if it also could use the Neural Engine instead of just the GPU.
True
Your top model doesn’t exist. Unbinned the M3 Max is 48GB Unified Ram, not 32
Then that is the top model, error in the chart, one man crew, etc.
@@ArtIsRight they unfortunately force you to upgrade to 48GB if you choose the 16/40, leading to the increase of $500
Mentioned this in numerous videos already, this video is an addendum
@@ArtIsRight someone’s sensitive. An inaccurate addendum.
If anyone in the comments can please help me? I am thinking of buying a MacBook M3 pro 11 core, 12 core pro or a 14 core M3 Max, 18 GB or 36 would be enough, but I would like to know the difference when running MATLAB, Apple's office page is very ambiguous in the data, gives some graphs without specifying which test was done, and in the end he says little about what he did, such as the number 12 or the numeral 14-- for example 12) Testing conducted by Apple in September and October 2023 using preproduction 14-inch MacBook Pro systems with Apple M3 Pro, 12-core CPU, 18-core GPU, 36GB of RAM, and 4TB SSD, production 14-inch MacBook Pro systems with Apple M2 Pro, 12-core CPU, 19-core GPU, 32GB of RAM, and 8TB SSD, production 14-inch MacBook Pro systems with Apple M1 Pro, 10-core CPU, 16-core GPU, 32GB of RAM, and 8TB SSD, and production 2.3GHz quad-core Intel Core i7-based 13-inch MacBook Pro systems with Intel Iris Plus Graphics, 32GB of RAM, and 4TB SSD. Tested with MATLAB and Simulink R2023b v23.2.0 and Parallel Computing Toolbox using a vehicle dynamics model. Performance tests are conducted using specific computer systems and reflect the approximate performance of MacBook Pro.
Yeah difficult one. Look at MatLap and Simulink and see what system resources these program are using. From there choose the machine based on that