I just built a computer specifically for darktable use. It is: Intel Core i5-4690 3.5GHz Socket 1150 ASUS H97-Pro Gamer LGA1150 ATX Motherboard Gigabyte Radeon R9 280 OC 3GB Corsair Vengeance LP 16GB (2 x 8GB) 240-Pin DDR3 SDRAM 1600 (PC3 12800) Desktop Memory Samsung 840 EVO Series 250GB SSD Seagate 3TB 7200RPM SATA3 Internal Hard Drive ST3000DM001 Antec True Power Classic Series 650W 80Plus Gold Power Supply Antec P100 Professional Silver Mid Tower Case A wireless PCI-X card and a DVD drive
This cost me less than AUD $1500 from a local parts supplier. I designed this machine to be in a sweet spot between price, power consumption and performance. I specifically bought an AMD video card as OpenCL performance is reputed to be better on AMD cards than nVidia ones for a given price/power point. The Core i5 is the second-top of the line, just below the overclockable "K" version. The RAM is medium speed (i.e. one step up from the basic one in terms of latency). Subjectively, it's the fastest performance I've seen for dt of any computer I've used, which isn't surprising seeing as it was specifically designed for the purpose. I'd say the quality of the AMD proprietary drivers is below that of the nVidia ones (I'm using the ones that come with Ubuntu 14.10, and I've heard the new ones that ATI released recently are significantly better, so I'm hoping for an improvement when I upgrade to 15.04). I see some visual glitches with the chromium browser and get occasional funkiness in dt where the occasional image thumbnail will come up black (opening the image in the darkroom then switching back to lighttable fixes that) and I've seen one image export almost completely black). These are minor, and I'm hoping will improve with the drivers, but are something to consider as I've never seen any equivalent glitches with my previous machines running nVidia cards. The case is actually really nice, plenty of ventilation, but padded on the inside to reduce noise, and with nice conduits to allow routing of cables on either side of the motherboard mount plate so you can keep the upper surface relatively free of cables. The one thing I'd do differently next time is not buying that 3TB HDD, check it out on here: https://www.backblaze.com/blog/best-hard-drive/ Otherwise it's a good machine, with the SSD and USB3 particularly making a huge difference to subjective performance (I have my images on external USB drives). If I can make the time I'll do some dt benchmarks and report back. Cheers, Rob On 14/02/15 04:24, Matthieu Moy wrote: > ----- Original Message ----- >> Matthieu Moy <matthieu....@grenoble-inp.fr> writes: >> >>> I'm considering replacing my old computer with one that would let >>> darktable run better. I'm seeking advices on which hardware pieces are >>> the most important. >> >> Thanks to all those who responded. Based on these advices, it seems this >> configuration would be a nice compromise: > > I ended up buying this configuration: > >> http://www.ldlc.com/fiche/PB00126228.html (french only I guess, sorry) >> Core i5-4460 => not the best, but a good quad-core >> Radeon R9 270 2 Gb => good floating point performance according to >> gpuboss (2,355 GFLOPS) >> 8 Gb RAM DDR3 => not huge, but 2 slots available so a jump to 16 Gb or >> even 24 Gb is rather easy. >> Disk 1 To => enough to hold a few year's of RAWs for me ;-). >> No OS >> 690 € > > What a breath of fresh air to use darktable on this compared to my old > computer... A little feedback to help other people looking for advices: > > The OpenCL performance is really good compared to the pure software > implementation (I had to install the proprietary driver to get OpenCL, not > sure whether it's supposed to work with 100% free software). > > I benchmarked a bit (using darktable-cli): > > One image with a lot of processing, with OpenCL: > 7.12s user 1.20s system 118% cpu 7.033 total > > Same image, without OpenCL: > 107.04s user 1.37s system 368% cpu 29.392 total > > 30 images with reasonably amount of processing, with OpenCL: > 128.39s user 17.74s system 127% cpu 1:54.91 total > > Same 30 images, without OpenCL: > 1247.44s user 11.92s system 355% cpu 5:54.10 total > > Therefore: > > * OpenCL is 3 to 4 times faster than non-OpenCL on this machine. > > * When using OpenCL, the CPU is not used much (around 120% CPU out of 4 CPU = > around 1/3 of the CPU available). When watching the CPU usage with > gnome-system-monitor, I see that the CPU usage never goes above 60% for a > given CPU. Good job dt developers, you're using OpenCL wherever possible it > seems :-). > > So, I confirm that having a good GPU is more important than having a good CPU > for darktable, and spending more money on a better processor (like i5 -> i7) > would probably not have improved much. > > I watched the RAM after a reasonably long darktable session, with Digikam and > Firefox open too, and the 8 Gb were not fully used (and swap wasn't touched). > So, I'll wait a bit before buying more RAM (but will sure need to do so > before the end of the machine's life). > ------------------------------------------------------------------------------ Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server from Actuate! Instantly Supercharge Your Business Reports and Dashboards with Interactivity, Sharing, Native Excel Exports, App Integration & more Get technology previously reserved for billion-dollar corporations, FREE http://pubads.g.doubleclick.net/gampad/clk?id=190641631&iu=/4140/ostg.clktrk _______________________________________________ Darktable-users mailing list Darktable-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/darktable-users
