I don't game but last year my GPU was getting frustratingly slow (I have the patience of a 2-year old) - I replaced it with a Geforce RTX 4070. LR Denoise with an A1 FF image takes 10 seconds.
DWOfPaul wrote:
Have you happened to find a denoise tool that works as good as DeNoise Ai for TIF files?
This is exactly the problem I have. The DeNoise AI was good for TIFFs, but not fully developed, then it died out. The Photo AI was not quite as good or at least not well controlled. DXO is great with RAW, but nearly useless for TIFFs.
I don't know if there are any good AI programs for TIFFs, but I really could use one.
Out of curiosity, I did a quick performance test using LR denoise on two different systems using A7rIV 60mp raw files.
System 1:
2019 MacBook Pro
i9 (8 cores hyper threaded)
64 gigs of RAM
AMD mobile graphics card 8GB of VRAM
Results = Approximately 90 seconds per photo
System 2:
2025 Lenovo Legion Pro 7i gen 10
Intel Ultra 9 275hx (8 performance cores, 16 efficiency cores)
32 gigs of RAM
Nvidia 5080 mobile graphics card with 16GB of VRAM
Results = Approximately 15 seconds per photo
While my test was not perfectly scientific, I do feel it clearly shows that if LR AI Denoise is slow for you, upgrading your graphics card can gain you much better performance.
EB-1 wrote:
How fast does that big gaming laptop run Topaz or DXO?
EBH
So far, I have only put LR and PS on it. I intended to use the Lenovo mostly for gaming and programming. At the end of the day I prefer Apple, and I used to use Boot Camp or Parallels for those tasks. But 5 years into to M world, and there is still not a great way to run high performence Windows applications on an M chip, so I gave in and got my first PC in 15+ years. Once the M5 MacBook Pros come out, I will probably end up buying one for my photo editing, video editing, and daily use. I have to admit, though, I am very curious now about how Topaz will perform on the Lenovo, so I might have to set up a test over the weekend.
Maybe it is out of scope of the thread, but it's difficult to find comparisons across software. Everyone is comparing one software (like Abobe) across different computers, but that's kind of useless for me to compare RAW conversion with NR on Abode, Topaz, and DXO on one machine. I'm not naturally a computer person like you, so managing a dozen boxes with Linux and/or Windows is enough already without even considering Apples.
EB-1 wrote:
Maybe it is out of scope of the thread, but it's difficult to find comparisons across software. Everyone is comparing one software (like Abobe) across different computers, but that's kind of useless for me to compare RAW conversion with NR on Abode, Topaz, and DXO on one machine. I'm not naturally a computer person like you, so managing a dozen boxes with Linux and/or Windows is enough already without even considering Apples.
EBH
IDK if I can get to it today, but I can definitely time all 3 on my Mac. Overall, I think the models take a similar enough amount of time that most people go with the one they feel will produce the best results. While I don't believe any of the major players are sharing their exact approach, there are common approaches and tools for building AI image denoisers, which probably explains the similar quality and performance we are seeing from most of the major players.
At this point, operating systems are basically a blur to me between the Apple, Windows, and Linux (including Android) worlds. It makes me good at picking up new products and figuring out how they work, but bad when someone asks me for help with their computer and needs exact step by step instructions off the top of my head
A few weeks ago, I went down the Google rabbit hole of trying to figure out how fast the M4, vs M4 Pro, vs M4 Max are in LR denoise, to try and figure out if denoise performance scales linearly with GPU core count. For example, if this is the case, and say a photo takes 20 seconds on an M4 Pro with 20 cores, it should take around 10 seconds on an M4 Max with 40 GPU cores, and around 40 seconds on an M4 with 10 GPU cores. Unfortunately, I have not found enough data online to figure this out for sure. My hunch is the M4 Pro chip with 14 CPU cores and 20 GPU cores is probably the sweet spot for most photographers, but I would like to try and prove it.
That's the one I have. It's about 25s on the A7RV 60mp files. It's quicker on some older A74 files. Much faster than the old computer I had but I'm spoiled now and want faster. It would help if they allowed you to move onto the next file as it processes like someone mentioned. Hopefully they'll change that in a future update.
Ugh... I just looked at the subscription Topaz decided to charge... $199 annual for Personal and $599 for Pro. They renamed Photo AI to Photo, but I strongly believe that this is the same Photo AI in disguize.
Absolutely not interested in this subscription.
I wonder if they may come to regret the change in a few months. It is hard to imagine a large number of photographers jumping on this subscription.
It's $199 each for Gigawixel and Photo (AI), so they expect you will get the Studio at $444. Supposedly they have millions of pro users, but the $900 pro version is only necesary if you have more than a million revenue per year. I'm not sure what percentage that is, maybe 20%? Assuming that figure Topaz would be a billion dollar annual company if those 2 million users all become annal.
I'm perplexed that if one signs into their account, they still try to sell you upgrades to GP AI and TP AI even though there are no further updates. Seems like they are trying to shakedown every $. No thanks Topaz, I've moved on!
DWOfPaul wrote:
A few weeks ago, I went down the Google rabbit hole of trying to figure out how fast the M4, vs M4 Pro, vs M4 Max are in LR denoise, to try and figure out if denoise performance scales linearly with GPU core count. For example, if this is the case, and say a photo takes 20 seconds on an M4 Pro with 20 cores, it should take around 10 seconds on an M4 Max with 40 GPU cores, and around 40 seconds on an M4 with 10 GPU cores. Unfortunately, I have not found enough data online to figure this out for sure. My hunch is the M4 Pro chip with 14 CPU cores and 20 GPU cores is probably the sweet spot for most photographers, but I would like to try and prove it. ...Show more →
It correlates almost perfectly with # of cores, even across generations, though the correlation improves if you add a fudge factor for processing speed across generations. When I get some time later today I'll post graphs.
DWOfPaul wrote:
A few weeks ago, I went down the Google rabbit hole of trying to figure out how fast the M4, vs M4 Pro, vs M4 Max are in LR denoise, to try and figure out if denoise performance scales linearly with GPU core count. For example, if this is the case, and say a photo takes 20 seconds on an M4 Pro with 20 cores, it should take around 10 seconds on an M4 Max with 40 GPU cores, and around 40 seconds on an M4 with 10 GPU cores. Unfortunately, I have not found enough data online to figure this out for sure. My hunch is the M4 Pro chip with 14 CPU cores and 20 GPU cores is probably the sweet spot for most photographers, but I would like to try and prove it. ...Show more →
If you are using the Apple Silicon it is probably using the 16 core neural engine and not the GPU or CPU unless you have an option to select which chip is used.
My M3Max has a 40 core GPU and it is quite a bit quicker than using the 16 core neural engine when using DxO Photolab - but that may not be true for a 20 core GPU.
Here's a correlation for # of GPU cores, across M1 to M4 devices, vs Adobe Denoise using LRC.
I took the data from the last Art is Right review. It was a few months ago I think. The table below shows the data I used. Art uses Nikon and 36MP raw files for these tests. I'm only looking at GPU cores, not CPU cores, but I do include (in the table) which M device generation it comes from as I use it in the 2nd graph to sWAG a generation speed improvement.
The correlation above is good but we know that speed improves from gen to gen so I added a 12% fudge factor. ie a factor to "allow" a speed improvement of 12 % from M1 to M2 (etc). I did this as a multiplier to the # of cores (kind of adding virtual cores). As you can see the correlation tightens up somewhat as the M1 through M4 devices are separated. Virtual GPU cores vs Denoise from Art is Right data by Mike Broomfield, on Flickr
.
I don't even see a trial option for Topaz Photo.
I wanted to confirm that this is renamed Topaz Photo AI (that I already own), but I don't see an option to try without subscribing.
It all seems very confusing…. I think I’m going to move to PhotoLab9 and ditch C1 & PhotoAI. I still have Gigapixel which I use occasionally and I guess that will keep working until an OS upgrade breaks it.
