I have been giving this some thought. Do you need an internet connection to use AI in PL9? Most companies, the answer is yes.
So if you internet connections isn’t very fast……………….. does this cause problems when you go to render photos?
Does DxO use the data from our photos to improve their AI?
AI masking does not require an Internet connection. I just tested it to make sure.
I don’t know if DXO uses customer images, maybe send them an email for a definitive response.
So a slow internet connection doesn’t cause the problems……………… hey ho
Wait a minute. If you don’t need an internet connection then the AI engine is on my machine and the one thing we know about AI is that it is very hungry, very. The problem might be just that.
Exactly, that’s why you download extra models during the install phase, and why a GPU with 6 to 8GB of VRAM is needed on top of 16GB of RAM (according to PL9’s release notes): to be able to run the models locally.
Instead of speculating, just check → https://support.dxo.com/hc/en-us/articles/6758873797917-Will-my-images-be-sent-to-DxO-servers-when-using-DeepPRIME-technologies-or-AI-Masks
But it takes a huge amount of grunt to run AI. Meta and Microsoft are at the moment trying to build two nuclear power pants to provided electricity for their AI - that’s how much grunt you need.
Using AI and training AI are two different things. AI is trained on data to create a model which is then used to perform the AI tasks.
Training takes the grunt and power not the use of the models generated by the training (although they do require the latest generation of GPUs and NPUs.)
I do get that but………… according to my PC it takes a lot more grunt if you use the AI masks. Certainly, there is more to this than just CPU, GPU and VRAM.
I’m certainly no computer whiz, but as I read all of these posts from thread to thread, I’m starting to think the problem is more with the processor than the GPU, and more than likely some combinations. I have a Sparkle Guardian OC Arc B570 10 GB Video Card, which by all accounts is near the bottom of the barrel for performance, yet it runs just fine. I have 32G of DDR5 RAM, so no issues there. Why does it run ok on my machine?
I have an Intel Core 7 Ultra 265K processor. The processor is part of Intel’s Arrow Lake line released in late 2024. From PC magazine: “Arrow Lake also sees Intel introducing its first neural processing unit (NPU) in a desktop CPU. The Core Ultra 7 265K has a pair of Intel’s third-gen Neural Compute Engines, which the company says has AI performance of 13 TOPS”
The processer is designed to run AI. That’s the difference in ‘grunt’. Interestingly, when we built this computer (which is specifically for image/video processing) we were going to put an i9 in it, but it went off-sale, and this one came on. After investigating we went for it. If not for the sale, would I have had as good of a DxO experience with the other processor?
The minimum requirements frankly are hog-wash. Interestingly, these aren’t the original minimum requirements when I upgraded from PL7. Seems the goalposts have changed. Anyway…..
8 GB RAM (16 GB RAM for AI Mask) - There are difference types of RAM, DDR 3, DDR4, DDR5. I suspect that the first one won’t be compatible with the motherboards used with the recommended CPUs. So say DDR4 and DDR5. And both of them have different speeds. You could have DDR5 6000 or you could have DDR4 2133 - the difference is like a Rolls Royce and a Fait.
There are many aspects of a motherboard that effect the ability to move data around. Just from a novices point of view, if you use a WD Black SN8100 M.2 2TB, then that is incredibly fast, but if you use a SSD through your SATA port on your laptop, not so fast, not fast at all.
The motherboard is essential for connecting and allowing communication between the CPU, GPU, RAM, and other components - poorly or cheaply designed motherboards can limit performance knows as bottlenecking.
DxO don’t even mention motherboards. I get that it is very difficult to cover ever eventuality but to turn round to an exist customer and refuse to refund when the software doesn’t work on a system that meets minimum requirements - well….. that pretty crappy. Sort of thing you might expect off Apple but DxO should be better than that.
Let’s see and compare what a few of the alternatives describe under system requirements.
https://helpx.adobe.com/lightroom-cc/system-requirements.html
I think its a ‘Bigger, Better, Faster, More!’ (Album title from 4 Non Blondes, '92) area. Yes, even system RAM (like the DDR4 vs DDR5) may affects a lot (may for some workload not to much). As you mention, motherboards ‘just’ the ‘Glue’. And ‘you can build’ (order) whatever you want (of course on your budget) - that’s the nice thing on PC. I think its ‘clear’ the PL9 vs pre-defined AI masks (and all AI related, like manual AI masks, DP NR) the GPU VRAM amount is the most important, and secondly the ‘GPU speed’. PL9 need to be work whatever ‘older’ GPU you has, until it has enough GPU VRAM. Other things like CPU or system RAM is less important until if it some decent.
For only instructive reasons i describe a real-world case for some similar - but its cant be comparable for DxO PL case - how some industry handle ‘recommendation’. Again: i don’t think its can be valid / works for DxO or similar cases. Get a video post-production company, with 10 workstation (edit bay): 8 for Avid Media Composer, 2 for Autodesk Flame artists. Avid give ‘certificated’ PC list, Autodesk (DL) also provides. Usually like: HP workstation Z6, Z8 or Dell Precision xyz model(s), etc. And they also describe: where RAM is exactly in this slots only, GPU is only this slot(s), and so on. For post-production houses the PC price is less important, as they also has storages start from few hundered of TB and far more, archiving systems, professional services (support) and so on, what’s also quite costly. Avid annual price in some 500USD/EUR range, Flame may around 5000 per year and so on → For this software price ranges, and ensure they smooth operations → they just get the ‘certified’ PC (some company also sell their custom build certified system like: BOXX) Note: good Flame workstation may need 256GB of system RAM, and like 2 of very high-end nVidia like RTX 5000 with 32+GB VRAM and so on.
So, versus ‘motherboards recommendation’ some industry say: ‘certified system’. But its all high-end, and for their whole operations (and revenues) the PC cost not really matter. So, even DxO say: this is a certified workstation for you not in real advance, as its has least 16GB of GPU VRAM card, etc.