I’m in process of specing another PC to scratch build for myself - but I’m uncertain what I should put more money on, and what is less important, I recently installed a decent graphics card in my current PC, set PL to use OpenCL and saw a modest improvement in ‘export’ functions.
I’m on a 4 core i5 now, bags of memory, SSL, so what should I throw at this? More cores? Raw core performance? Storage system? I don’t see PL using a huge proportion of my current 32Mb, so I don’t think ‘more memory’ is the answer. But a different chipset (so a bit faster I/O might be)
FWIW, I’m a (retired) software developer. I get that multiprocessing is hard*, and that, as you add threads and processes, you get diminishing returns based on locking sections. So I’m not expecting an i9 (12/18 cores) to do much for me.
*not so much hard to write but hard to debug when something goes wrong, especially on a machine with timings other than your QA test mules.
I’d love to know what machine setups developers are targeting. That was the ‘secret sauce’ when building Windows drivers. Any chance of finding that out?
one of our members on the forum called “Benci” is much more of an expert on this than me but he replied once on a similar topic:
"If nobody has more comment …
Chose highest core type CPU ( AMD 12 core or Intel 8 core )
Use >16 GB RAM memory
Install OS on SSD
Select a separate GPU card ( no necessary the very expensive, just distribute rendering )
Select a motherboard with M.2 bootable memory socket and e.g. Samsung 960Pro SSD
( it has a transfer rate >2700 Mb/s )
However, the rendering ( Export ) is independent from editing, you can run when go to sleep.
The CPU speed, memory bus speed and GPU bus speed are not so important. You can reach some percent improvement only. A 45 Mb image file need during rendering ~2 Gb RAM."
there are several HW threads on the old forum http://forum.dxo.com/5
GPU doesn’t seem to help much.
I’m also interested to hear from DxO developers the secrets about building a reasonable fast machine…
I’ve read it. I was asking yet again because it doesn’t answer my question, which is “what hardware to optimize?” I’m looking at xeons, hepta-core i7s and Ryzens, so I need to choose between single-core power and number of cores.
I’d like displayed changes to edits to be more responsive, would like to focus hardware changes on improving that.
CPU processing power and the amount of installed memory are the primary factors affecting system performance with PL. Second to this is GPU which offloads a small percentage of rendering from the CPU.
CPU processing power and the amount of installed memory are the primary factors affecting system performance with PL
Yeah, OK, CPU and memory, like for everything else that isn’t disk bound - but, for the CPU, is there more benefit in per-core performance, number of cores or number of threads?
FWIW, I have 32G of memory in my desktop PC, and have never seen PL use all of it. My notebook (8G) is a different matter. I see as many as 100 page faults/sec when PL is scanning a folder. (D750 and 600 so 26 -30m raw files). Zero page faults or one when in ‘customize’ on one of those files. I’m guessing the sweet spot is around 12 or 16 on a W10 box.
On export, DxO Photo Lab uses all of the cores. For editing Photo Lab seems to only use about four of them.
Here’s some real world Mac examples: editing on my MacBook Pro 2011 17" 2.2 GHz 4 core with 16 GB RAM doesn’t feel much slower than on my Mac Pro 12 3.33 GHz core computer with 128 GB of memory and Radeon RX580 with 8 GB of VRAM.
On the other hand, when I go to export a set of 50 photos with Prime Denoise (a mix of 5DS R and 5DIII RAW files), the MBP has full fans on and takes over an hour. The Mac Pro does the job in about 12 minutes. All cores are maxed (don’t plan on doing much multi-tasking, although web browsing or writing will still work).
GPU: I have special software to monitor the GPU. It doesn’t seem to be taxed at any point when running DxO Photo Lab Elite 2.
I expect the engineering is similar for Windows. If it’s not, I’d love to learn about the differences.
Nice to know. I see 4 cores in use when I’m editing. Don’t have a machine with more. Sounds like the answer is ‘get four cores, then focus on individual core performance’. So I can stop wondering about one of the hexacore CPUs.
It would be great to have, from DxO developpers, more information than just minimum requirement.
What is better in order to run PhotoLab with Prime and all functions ?
For display and adjustments vs during export :
more CPU power with integrated grahics or less CPU power + additional GPU ?
GPU card for game or for acceleration ?
“N” cores or “N + 50%” threads ?
full cores frequency or turbo frequency ?
which benchmark can be used as a reference for PhotoLab ?