Well to be honest I'm not too sure if this is an improvement in 3.0. My computer is quite new (last year) and I chose it wisely so that it would perform very well for editing picture. This was the chance for me just to look at how it compares against others. But still, I'd be curious to see the performance difference when you'll run the same export with 3.0.
Cheers On Fri, 20 Dec 2019 at 07:37, Jochen Keil <[email protected]> wrote: > Hi Sébastien, > > thank you for taking the time to perform the benchmark! > > I also re-ran my benchmarks with the masks disabled (for the exposure & > tone curve modules) and got a much better performance. > > Your results suggest that there have been some significant improvements on > this in DT 3.x, so I'm looking very much forward to this! > > Thanks again and best wishes, > > Jochen > > > > > > On Thu, Dec 19, 2019 at 11:21 PM Sébastien Chaurin < > [email protected]> wrote: > >> Thanks for your reactive and detailed answer Holger ! >> >> I’m currently running DT 3.0 rc2, more precisely commit >> 680668c58a1415322b89af0a689cb792e747bd6f >> >> I’m a little bit behind, but since we’re running in the final stage of >> issuing 3.0, most of the effort at the moment is being put on translations >> documentations etc so I don’t think any performance improvement is to be >> expected when 3.0 will be out. >> >> Cheers >> >> On 19 Dec 2019, at 20:46, Holger Wünsche <[email protected]> >> wrote: >> >> HI, >> >> >> looking at the output from darktable it seems your the masked modules are >> a lot faster than on my system. Which version of darktable do you use? Your >> demosaic and defringe are comparable to mine, the others are just >> significantly faster so we didn't really paid a lot of attention to them. >> >> The numbers from time (real, user, sys) are just the the total runtime >> (real) and the amount spend in the application (User) and the time spend in >> the operating system (sys). The last two are the sum across all cores, so >> having 8 cors calculate 1 second will result in 8s in user-time. >> >> The time from 9.5 to 17.8 is almost the same for me and I think it is the >> time needed to compress the png (but I don't know ;) ). >> >> >> Regards, >> >> Holger >> On 12/19/19 8:19 PM, Sébastien Chaurin wrote: >> >> For the sake of benchmarking my system, I also used your files and >> command from your email : >> >> $ time darktable-cli 2019-11-23T23_23_35+0100_7871.arw >> 2019-11-23T23_23_35+0100_7871.arw.xmp 2019-11-23T23_23_35+0100_7871.png >> --core --library :memory: -d opencl -d perf 2>&1 > dt_log.txt >> >> Although it took less than 18s, I’m not to understand the other numbers : >> >> real 0m17.835s >> user 0m48.886s >> sys 0m3.417s >> >> And from the log it created I have : >> >> 9.516693 [dev_process_export] pixel pipeline processing took 8.120 secs >> (37.632 CPU) >> 17.805166 [opencl_summary_statistics] device 'AMD Radeon Pro 580 Compute >> Engine' (0): 295 out of 295 events were successful and 0 events lost >> >> Does this mean it took 17.8 - 9.51 just to write the png file on the disk >> ? >> >> I find in my case that by far the 2 modules that used the most power >> were demoniac and defringe. Not sure why this does not seem to be the case >> for others… Apart from those 2, yes tone curve 3 and exposure 1/2 >> were slightly higher that the others… >> >> CPU and GPU were reasonably loaded while exporting but not 100%. >> Full log is attached. >> >> I’d be happy for someone to enlighten me on these results :) >> >> >> ____________________________________________________________________________ >> darktable user mailing list to unsubscribe send a mail to >> [email protected] >> >> >> >> ____________________________________________________________________________ >> darktable user mailing list to unsubscribe send a mail to >> [email protected] >> > ____________________________________________________________________________ darktable user mailing list to unsubscribe send a mail to [email protected]
