I think it’s time for an upgrade.
What would be the minimum spec for running multiple instances of various foss software (DT, RT, Gimp 2.9, Hugin, Luminance, etc…)?
I could either swing to Mac or Win, though if it were Win it would be quickly converted to dual boot Linux. The other bonus for Win is I already have peripherals, so just need a tower box - helping to further reduce costs.
Any help would be much appreciated!
[quote=“fotonut, post:1, topic:565”]
What would be the minimum spec for running multiple instances of various foss software (DT, RT, Gimp 2.9, Hugin, Luminance, etc…)?
[/quote]It’s limited by RAM and mostly depends on what you want to open in said software. I’d say go with at least 8 GB. If you deal with big panoramas you might need even more.
I hope that helps.
Thanks for your reply Jonas.
What about CPU and GPU options?
Celeron? I heard that was a cutdown version and probably not suitable for CPU intensive work?
[quote=“fotonut, post:3, topic:565”]
What about CPU and GPU options?
[/quote]The more you pay to faster they become. In general memory limits the amount and size of images you can open, the cpu dictates how fast processing is going to be. A celeron is among the cheapest things you can get and I wouldn’t recommend them if you can afford something more powerful.
Thanks for that.
Would you consider an Intel i5, or is that a bit slow now?
[quote=“fotonut, post:5, topic:565”]
Would you consider an Intel i5, or is that a bit slow now?
[/quote]They are generally fine. Maybe 10-20% slower than an i7.
So this could be worth looking at?
I’d need more disk space and more ram. But I can’t answer that question for you.
It’s upgradable to 16gb ram, and when the need arises, fit another HD.
For processors, an i7’s advantage over an i5 is twofold: they are clocked higher to start, and they have hyperthreading.
In Filmulator’s tone mapping step it alternately performs a blur and a big old bundle of multiplications and additions. On my i5 desktop, they take an equal amount of time, but on my i7 laptop, the blur takes half the time of the other step.
So for certain image processing work, an i7 could possibly be slightly more than twice as fast as an i5, but that’s not going to be true all the time.
I was talking about the differences given equal clock speeds something like this:
If the i7 runs at a higher clock speed the differences are of course a bit bigger:
@fotonut If you already know that you will need a bigger disk it’s usually cheaper and less effort to just start with a bigger one. Also the Box lacks a dedicate graphics card. With Software that makes good use of the graphics card that can make a big difference.
Darktable supports OpenCL (computing using the Graphics Card) for quite a few operations. I think the next version of Gimp/GEGL do too but I’m not certain.
My concern with the Anandtech benchmarks is that I don’t know how they translate into performance on the programs we’re using.
One of the “Agisoft Photoscan” benchmarks shows a 50% improvement, and others, not much. But what’s actually going on? Are they applicable?
How much faster does Amaze get with hyperthreading? RT’s Contrast by Detail Levels? NR? Sharpening?
These would be nice to know, and I was offering my own results from Filmulator.
[quote=“CarVac, post:12, topic:565”]
How much faster does Amaze get with hyperthreading? RT’s Contrast by Detail Levels? NR? Sharpening?
[/quote]I don’t know for certain but I do assume that they behave like most other software in that they only show small benefits but yes they could be exceptions too.
I did a totally non scientific test with darktables opencl support. Using a GTX 970 with OpenCL was roughly twice as fast (1.3s vs 2.8s) as using just the i7 6700k. I expected the difference to be even bigger (it might be with more complex edits) but that’s still a nice speed up. Changing the number of threads from 4 to 8 made no noticeable difference (both with opencl on and off).
How did you change that?
[quote=“houz, post:14, topic:565”]
How did you change that?
[/quote]Using GOMP_CPU_AFFINITY=0,1,2,3. If I set it to one or two cores the performance drops as expected.
The rough numbers are:
1 Core: 11.4s
4 Cores: 3.1s
8 Cores: 3s
OpenCL: 1.3s
To be clear: This is not a proper controlled benchmark, this is just me starting up darktable and zooming to 100% a few times. There are tons of uncontrolled variables.
Makes sense, after some point the bottleneck isn’t the computation any longer I guess.
Very interesting. What resolution is your screen?
Can any of the darktable devs explain what would probably be included in this timing? Is Amaze (seems not to have any OpenCL implementation) running? On just the visible part of the image, or the whole thing?
[quote=“CarVac, post:17, topic:565”]
Very interesting. What resolution is your screen?
[/quote]2560x1600.
It looks like amazee is running on the CPU even with OpenCL on but it’s rather quick:
[dev_pixelpipe] took 0.058 secs (0.364 CPU) processed `demosaic’ on CPU, blended on CPU [full]
Maybe it’s cached.
3 seconds, cached, for just a part of the image?
Do you have sharpening and NR active?
[quote=“CarVac, post:19, topic:565”]
Do you have sharpening and NR active?
[/quote]I closed the console that I used for this but yes, I threw in a lot of operations to make it run a bit longer. Among them 3 instances of denoise profiled those are fairly expensive. 