stock would be better, since I will be doing long hours of compute... thanks in advance
Duly noted, I'll do both runs for giggs. But I'll be sure they're clearly labelled for you.
I meant to do this last night on my 2 Sapphire Fury Tri-X OC cards, but Fallout 4, sorry. I'll try and remember to do it tonight. I have a R9 290 on a Fedora 22 machine as well but I'm not really that good on linux so I don't know if I can get it work or not.
windows 7
sapphire r9 290x 4GB oc to 1148core 1500 mem
high align setting
high build density
photo set monument
Device 1 performance: 1353.49 million samples/sec (Hawaii)
Total performance: 1353.49 million samples/sec
Cool, just make sure xfire is disabled and then restart before running the benchmark, it might make quite a difference
My Full Machine: http://pcpartpicker.com/p/CFrjrH
The noise was fine. The card ran at 75-78 degrees, and the fan speed was below 50% the whole time. I sit about 2 feet from my machine, and while the AC was running I couldn't hear my machine. When the AC cut off I could hear it if I was listening for it, but it's not really that loud.
I ran these tests in windows (sorry, all my linux stuff is done with VirtualBox on this machine) while working, so I had an IRC client, a couple putty windows, a couple cygwin windows, notepad++, and my VPN all running. I turned off everything else, including spotify and my browser, so those should be the only things effecting the test. Most of the resources would have been CPU resources, so it shouldn't skew the results too bad, and it's probably a more real world test (let me dream, alright).
Monument
Stock Clock-High: Device 1 performance: 1552.84 million samples/sec (GeForce GTX 980 Ti)
Stock Clock-Ultra: Device 1 performance: 1518.87 million samples/sec (GeForce GTX 980 Ti)
Building
Stock Clock-High: Device 1 performance: 1313.7 million samples/sec (GeForce GTX 980 Ti)
Stock Clock-Ultra: Device 1 performance: 1425.2 million samples/sec (GeForce GTX 980 Ti)
i7-5930k, 16GB DDR4, Samsung 850 Pro 256GB SSD
Windows 10
Sapphire Fury Tri-X OC 4GB HBM 1040mhz x2
Monument:
High:
finished depth reconstruction in 235.855 seconds
Device 1 performance: 1416.29 million samples/sec (Fiji)
Device 2 performance: 1456.78 million samples/sec (Fiji)
Total performance: 2873.07 million samples/sec
Ultra High:
finished depth reconstruction in 1413.51 seconds
Device 1 performance: 1318.94 million samples/sec (Fiji)
Device 2 performance: 1292.07 million samples/sec (Fiji)
Total performance: 2611 million samples/sec
Building:
High:
finished depth reconstruction in 65.3 seconds
Device 1 performance: 1076.88 million samples/sec (Fiji)
Device 2 performance: 1103.65 million samples/sec (Fiji)
Total performance: 2180.53 million samples/sec
Ultra High:
finished depth reconstruction in 363.375 seconds
Device 1 performance: 1140.94 million samples/sec (Fiji)
Device 2 performance: 1144.03 million samples/sec (Fiji)
Total performance: 2284.97 million samples/sec
A couple things about my Fury cards. The gpu frequencies are the stock frequencies set in the BIOS from Sapphire but I have flashed the BIOS to unlock additional shader units on both gpus. The stock shader count for a Fury is 3548 and the Fury X/Nano is 4096. These gpus have 3840 units. I also can't comment on noise or heat. All my gpus are water cooled so the temps are always low. They never broke 36C during any of the runs. These gpus also have dual BIOS but I can't reach the switch to set them back to the stock BIOS at the moment. My water block setup is blocking the BIOS switches so I would have to tear it all down in order to run the tests with stock BIOS.
thanks for posting sli fury results :) or xfire if you prefer the amd naming scheme.
pretty impressive, wich photo sample did you use?
pretty good, but weirdly better than the fury X.... lets wait for more results :) thanks!
Updated my post.
1 Like
Yeah, I thought that odd as well. I expected the the results to be close to a Fury X but not above. It might have something to do with differences in overall system specs even though this should be utilizing only the gpu's.
It seems that the 290x/390x would be the best bang for your buck.
for now it seems, although I would like to see more furyX results, because 200 million samples/sec edge of the 908ti doesnt look too much time saving, for 50 12MP photos, I ussually make project of 350-500 18MP photographs, so the minutes difference in the benchmarks mean hours in real case scenarios...
for the price of 2 908ti you could get 3 290x/ 390x and surpass them in performance easy.
yes! but I have a budget of 750€ 800 tops, so the options are 1 980ti, 1 fury x or 2 390s, I would prefer 2 390x's but the thing is I would have to buy another power supply, and that would put me out of budget for dual card scenarios ( I have a 850 w power supply and a 8320 @ 4.6 ghz, and compute on the graphics card it's a powr consumption bitch!) I'm not worried about the power bill but If I can avoid buying another power supply, much better... 3 way card are discarted for an h440, yes they fit, but it would be a thermal nightmare in such a case
Although you're not interested in the 980, I'll post some results when I get home (about 3 hours) just so you can compare them.
Considering the performance differences, I would honestly suggest that you get a 390 and a new psu now and then add a second 390 later. That would result in the best performance per dollar by far.
1 Like
yeah that's on the table too, still the decision it's being hard right now XD, I have a month to decide, I'll wait for more benchmarks, so we can come with a clearer.
that's cool I might not be interested, but someone may be in a position of interest in the 980 and photoscan (budget, need of cuda for other software.. personal preference...)