Confusion about the order of multiple instances of Diffuse or Sharpen

Do you use a pre-built, packaged version, or do you compile your own (you may be able to improve performance by compiling yourself, specifically for your machine)? If the latter, do you build with RelWithDebInfo or with Release? I build my own, and Release was way faster for me than RelWithDebInfo.

The time difference between scaled-down and normal export is shocking; I thought scaling down was the last operation, and it’s not so resource intensive. Have you tried running with -d perf to see which operation takes up much more time?

Thanks for sharing these tips, @kofa

No, I use the OBS release.

Regarding Release X RelWithDebInfo, do you mean this?

./build.sh --prefix /opt/darktable --build-type Release

Yes, I meant --build-type Release.

1 Like

Maybe this is not the right thread to ask this question, but I did not want to create a new one.

Which existing profile in Diffuse and Sharpen would be most appropriate (according to you) for a developed tiff/jpg file in a target printing resolution? This sharpening would be the last action before printing.

That depends on the size of the image, the size of the print and the eyes looking at the print.
There unfortunately is no easy answer.

I’d try local contrast, though (the fast version, in my case).

btw, i love the name of the module. The DoS really is a Denial of Service attack on my computer :slight_smile: (no opencl, unfortunately).

1 Like

Yep. DoS was the thing that finally persuaded me to upgrade my GPU!

My laptop is a little better that yours (not that much) but diffusion or sharpening is a hog, mainly because of the iterations.

While I do the editing, I change the iterations to 1 and it makes a huge difference.

Don’t forget, I occasionally do, to reset the iteration to either the default: 10 or to xxx before exporting.

1 Like

Thanks, good tip.

But the main issue is in exporting.

For now, it’s a no go to me, until my budget allows a brand new computer + a decent gpu

A decent GPU alone will take you a long way

2 Likes

Hard to do just the GPU if you’re starting from a notebook though.

eGPU is a thing.

eGPU isn’t a very viable option these days… you’d have to buy a GPU card and even years-old ones are absurdly expensive due to cryptocurrency mining and the chip shortage.

(Slightly newer ones are even worse…)

So eGPU isn’t really an option… unless someone here knows of a good eGPU case + some GPU card that works well enough and isn’t sky-high-expensive.

Of course, these prices also make a desktop computer with a dedicated GPU also cost prohibitive.

Does this mean the only option right now is buying a new laptop with a dedicated GPU? (Even some laptops with integrated GPUs are on backorder for half a year or more.)

Or is there a GPU that isn’t so great for the latest games, but good enough for darktable that works well in an eGPU that is not so pricey?

1 Like

GPU was the main reason I moved from laptop to desktop. That plus the ability to upgrade/swap out individual components instead of buying a whole computer each time. You don’t need a very recent/high-spec GPU for darktable. The one I’m currently using (6Gb Gtx 1060) is from 2018 (I’m very happy with it) and the one I used to use (2Gb Gtx 770) from 2013 (still better than CPU only). I seem to recall briefly considering an eGPU when I had a laptop but it looked prohibitively expensive.

What I have done occasionally is couple that with a low-spec laptop and remote into my desktop over a 4G connection (using NoMachine or VNC). Obviously that doesn’t work for colour managed image editing but it’s fine for pretty much everything else.

I always baulk at the cost of new PC stuff, so with a few exceptions only really buy pre-owned. Haven’t had any major issues yet (touch wood).

If you can’t find a decent eGPU (I assume these can be made to work well with OpenCL?) and you aren’t willing to go for a desktop, possibly that will be your “best” bet. Again, pre-owned would probably save you a good chunk of money, though likely you’d have to live without a long warranty.

2 Likes

I’ve been happy enough with my XPS 13 9360, which has Intel GPU only, but I have OpenCL on. However, the diffuse or sharpen module is quite slow. I’m happy that people work on performance for every darktable release. Hopefully DoS will get faster too. (Even still, it’s usually a last step to turn it on or back on. Thankfully darktable’s order isn’t the order you enable/disable things, for so many reasons.)

I have dual boot hackintosh/linux Ubuntu 20.04 built in 2018 with i7 8700k cpu, RX 570 GPU with 16GB RAM. On the Hackintosh (Big Sur) using a single instance of diffuse and sharpen is workable and sometimes two instances works ok depending on the other modules used, while in Linux the response is very good even with 2/3 instances of diffuse and sharpen. In 2018 I bought the RX 570 with 4GB RAM around US $130 in India.

Does Darktable on Big Sur use OpenCL in your case? Diffuse and Sharpen is rather computation-intensive, and I see a big difference when I disable openCL under Linux.

1 Like

OpenCL is working and enabled.

@Aliks This was the thread concerned. Read from the top, as it quickly diverged.