Intel CPU with INTEGRATED GPU and opencl for Darktable

Software darktable
1

I am looking for a minipc in the size of my Beelink (AMD) which supports opencl. I gave up to get opencl work with my Beelink.

I think the best thing to start is to find out wich integrated GPU supports opencl. It looks like AMD doesn’t support opencl at an acceptable level with integrated GPUs. Maybe Intel does, but it looks like there are problems too with Intel.

After it is clear which integrated GPU supports opencl I can try to find a minipc with it.

Assume, I will use Ubuntu 24.04 or later.

Can you recommend a CPU / GPU?

2

In my opinion rusticl should support integrated AMD GPUs, too. Sure it depends a bit on the CPU built in into your Beelink. I am running Darktable on a laptop with AMD Ryzen 9 5900HX with Ubuntu 24.04 using the rusticl drivers for opencl.

1 Like
3

I’m on Windows 11 currently, but I think openCL worked on Ubuntu 22.04 with my Intel Iris Xe 96eu (i7-11370H)… and it’s pretty fast for an iGPU, too. It actually makes a noticeable difference for me.

1 Like
4

I didn’t get it work. See Radeon 680M and opencl?

5

Could you please do a test for me:

Go to Index of /~sarunas/bench_raw/setubal and download setubal.orf and setubal.orf.xmp, Please note Firefox did not download the correct exetension name, if, rename it according to the size.

Check the right configuration, if Darktable can use all the available ressources. This setting can be a big difference in speed. I got the double speed with “large”.

Settings / processing / CPU - Memory / darktable resources: large

It took me weeks to find this out with my Intel B650. Compare No opencl with Intel B580 (Asrock Challenger) - #24 by linuxuser In this thread you can find also what I did with Ubuntu to get opencl working.

With the B580 I have now:

pixel pipeline processing took 1.415 secs

Of course this value never can be reached with an integrated gpu.

With my Beelink SER6 Max minipc, where I did not get opencl working, I get

pixel pipeline processing took 11.437 secs (Full log attached)

After everything is prepared please do:

darktable-cli setubal.orf setubal.orf.xmp test.jpg --core -d perf -d opencl

The interesting value is “pixel pipeline processing took” at the end of the log. Please attach the full log.

The question is, if it is noticeable faster than my SER6.

setubal-beelink.ser6.max.ryzen7,7735hs.32gb_resources.large-log.txt (4.5 KB)

6

With daktable resources set to large, I still get tiling on demosaic, denoise and lens, etc. so the result is kinda slow… (29.436 s)
darktable_iris_xe_01.txt (10.8 KB)

1 Like
7

I had a discussion already when I made a decision for a dedicated graphics adapter. There have been real odd results. My very old Nvidia was faster than a newer one. One used Windows, one Linux. This was the reason why I have chosen the Intel Arc B580.

I have found another CPU:

Intel Core Ultra 5 Prozessor 125H. This CPU has an Arc GPU integrated. Looks very interesting.

Ok, not a bargain,

and

Intel Core UItra 9 185H

8

The Ultra 5 125H doesn’t sound too bad, honestly… For the size of that thing it’s pretty good.
Sure, if you built your own desktop you’d be able to get some 8GB dedicated GPU (probably an older AMD), but if you need it to be this compact, it seems okay.

9

As long the price doesn’t change. Hours before I saw one with Arc8 cheaper, the 125H is Arc7.

You have to be very careful about the other details, when comparing prices.

My fave at the moment is:

64G RAM and 2TB NVME, probably the same slow NVME, which I already have in my Beelink.

10

Besides half of the slides talking about AI, yeah, that seems like a solid choice for the price and I think darktable should run pretty well on it.

1 Like
11

I found there are more Intel CPUs with Arc GPU, like the Intel Core U9-285H

If I am right, the power of the CPU doesn’t matter a lot when opencl is available…

There is GEEKOM GT1 Mega Mini AI PC with Intel Core U9-185H

And there is probably an outdated Geekom IT1 with U9-185H

So for Darktable the cheapest U9-185H is enough. It looks like the prices are not a lot different at the end.

12

There is a new expensive competitor

Beelink SER9 Pro AMD Ryzen™ AI 9 HX 370

Ok, this is not an Intel GPU but it could be that opencl works with AMD Radeon 890M

Maybe there are special offers in summer.

13

Mmm not bad, according to some article headline it should beat GTX 1650 gaming performance… hard to tell how that translate to DT, I’d have to look up some benchmarks

14

What I have learned, that the darktable configuration must be perfect. It took me a long time to find out how to get my Intel Arc B580 with “darktable resources: large” 2x faster.

The other question is, if the AMD GPU gets an advantage with “darktable resources: large”, someone said there is no difference.

You can read sometimes in the forum, that some features cannot be used with AMD.

So if nobody can test it with linux, it is guessing only.

It looks like the HX 370 / Radeon needs a newer kernel than the Arc 8. Kernel 6.14 is out with Ubuntu 24.04.2 and 24.04.3 shouöd be published soon, so I think this is not so import. I have read about graphical issues, black screens, a.s.o with the Radeon.

At the moment the AMD minipc is a lot more expensive, than the Intel.

15

In that case I’d go Intel then…

16

At the moment yes. AMD is 30% more expensive, but I am unsure if the tests with DT I have found, are optimal configured.

If the AMD could be 2x faster like my not optimal configured B580, then everything changes. The AMD test is 50% slower with Windows.

I read so often, that dt is slower with Windows, so it makes no sense for me to compare this.

17

In the case, that size matters, I found, that a Fractal Design Node 202 Mini-ITX case could be an option

The case is 377 x 82 x 330 mm, of course bigger than a Beelink MiniPC, but you can use some dedicated GPU, like the Intel B580.

At the end the Mini-ITX solution is a little bit (about 150) more expensive

1 Like
18

A dedicated GPU is way better than the iGPU.

1 Like
19

I am sure, but it is not so easy to build a ITX-mini pc.

And that is not a good idea. In case you get a PCIe 4.0 dGPU work with PCIe 3.0 with riser. What do you expect you loose with speed?

The Node 202 riser card does not support PCIe 4.0 and as it looks now, we do not have any plans of making a revised cable either.

There is the Fractal Ridge, but note:

Not every version supports PICe 4.0

Then you have to choose the GPU very carefully.

While with the Node 202 the ASRock > Intel Arc B580 Challenger 12GB OC would fit, it is too long for the Ridge.

The 12GB Sparkle Intel ARC B580 TITAN would fit, it is longer, but not so high.
https://www.sparkle.com.tw/en/B580-TITAN

I am still unsure, what would be the best GPU build, if I look at Ridge GPU Installation & information : Fractal Design Support

The goal is to use fans, but not the original ones.

20

I feel like with such a tiny setup even if you find a fitting and compatible dGPU, it’d be a risk of struggling to find and upgrade in the future that fits too.

1 Like