A request for 980ti and furyx owners

It's been a while since last time I posted here, but always kept following the content and the forum.

I have a request for anybody using this gpus, I need a benchmark that no one does the gpu million samples/sec for the dense point cloud generation in agisoft photoscan, this benchmark is way more important to me than gaming benchmarks will determine my holiday buying decision

The steps are the following.

  1. download agisoft photscan demo (if you have it and use it regularly way better)

  2. download the sample data http://www.agisoft.com/downloads/sample-data/ , I'm specially interested in the monument and building sample data

  3. under the tools menu select preferences then go to the opencl tab; select the gpu and set the active cpu core/threads to 0, so we will have the gpu working only.

  4. on the bottom you'll see a tab named console, is besides the one that says photos

  5. start the procesing of the photos, under the workflow menu add photos (make sure to load all of them) the run align photos (set it to low) and the run build dense cloud in high and ultra high ( the later only if you have 16+ gb of ram)

  6. let the reconstructing depth phase run until it finish, the you can cancel the process, and on the console copy and paste here the results. It should look something like this:

Device 1 performance: 588.824 million samples/sec (GeForce GTX 570)
Device 2 performance: 563.746 million samples/sec (GeForce GTX 570)
Total performance: 1152.57 million samples/sec

also temps and (subjective) noise are taken into acount, and if you have any other card don't hesitate topost (a 390x might be considered, for another reasons)

Also please post OS photoscan runs on mac,win and linux, and linux results are the most important to me

Thanks in advance


UPDATE: provisional result tables

gtx 460 stock (monument sample, w 8.1) 277.979 million samples/sec
gtx 460 @ 910 mhz (monument sample, w 8.1) 302.175 million samples/sec
r9 290x @ 1148 mhz (building, w 7, high) 1236.58 million samples/sec
gtx 980 sc acx 2.0 stock (win 10) 1088.03 million samples/sec
r9 fury X (win 10) 1267.63 million samples/sec
r9 290x @ 1148 mhz (monument, w 7, high) 1353.49 million samples/sec
gtx 980 ti stock (monument, windows, high) 1552.84 million samples/sec
gtx 980 ti stock (monument, windows, ultra) 1518.87 million samples/sec
gtx 980 ti stock building, windows, high) 1313.7 million samples/sec
gtx 980 ti stock (building, windows, ultra) 1425.2 million samples/se
r 9 fury unlocked @ 1040mhz (monument, win 10, high) 1456.78 million samples/sec
r 9 fury unlocked @ 1040mhz (monument, win 10, ultra) 1318.94 million samples/sec
r 9 fury unlocked @ 1040mhz (building, win 10, high) 1103.65 million samples/sec
r 9 fury unlocked @ 1040mhz (building, win 10, ultra) 1144.03 million samples/sec
gtx 970 @ 1450mhz(monument, win 8.1, high) 897.446 million samples/sec
gtx 970 @ 1450mhz(building, win 8.1, high) 734.815 million samples/sec

1 Like

Looks like its open cl so I would say amd would ha e the best results. Most cloud point software is cuda acceleration so if your doing environment laser scanning most times nvidia will perform better and they always recommend xeon 8+ core. You may want to go pro class cards also.

I'm about to go to the gym but I'll test my fury x once I'm back at home

Kevin agisoft photoscan is not lase scanning (wich can be offload to gpu also) it's photogrammetry, My gtx 570 dispalys pretty well, so it will be only for compute, therefore going pro, would only make me go out of budget, wich is 750 € (so 980ti /fury X is the most I can get), also I'm only upgrading the gpu, in the 2016 holidays we'll talk about cpus. AMD should get the win, but I also do rendering so the 6 gb of the 980ti would be better against the 4 of the fury, but I'm willing to sacrifice IF the fury gets a great edge on that, since my photogrammetry projects are kind of huge.

Don't know how well/at all crossfire works with compute stuff but with your budget you could just get 2 R9 390. Have you already considered that?

What about saying buying 780(tis) used for SLI?

I know locally someone sold 780tis fro $300 a pop 3 months ago.

well I xfire/sli sucks at compute but you can disable it, again, if it gets the highest score I will consider it, although I'm pretty sure the power consume would rocket, and some days I do like 8 hours of compute, soo it better be really better than one fury X

the problem with that is "only" 3 gb... for the other things I do, like rendering it's a little bit on the edge...

makes sense, I would go for the 980ti then, or maybe try to find a first gen titan.

I say this since the fury lacks ROPs, and has less memory.

390(x) might be worth checking out too, I'll try to show my 970 SLI when I get home tmr.

I don't have either of the cards you want to benchmark but I do have agisoft so I wanted to see what my current system looked like.

Sample: monument
OS = windows 8.1
GPU = GeForce GTX 460
Stock Speed benchmark:
2015-11-11 11:48:45 Device 1 performance: 277.979 million samples/sec (GeForce GTX 460)
2015-11-11 11:48:45 Total performance: 277.979 million samples/sec

Overclocked to 910Mhz
2015-11-11 12:11:49 Device 1 performance: 302.175 million samples/sec (GeForce GTX 460)
2015-11-11 12:11:49 Total performance: 302.175 million samples/sec

windows 7
sapphire r9 290x 4GB oc to 1148core 1500 mem
high align setting
high build density
Device 1 performance: 1236.58 million samples/sec (Hawaii)
Total performance: 1236.58 million samples/sec
working on ultra density now will edit with results

Wow! It's the the monumento por the building? I think it's on par of a titán x an netter than my 2 570 together!

it is the building., the monument i forgot to capture the results
where could i find a list of results of other gpu's?

http://www.agisoft.com/forum/index.php?topic=3678.0. And http://wiki.agisoft.com/wiki/Hardware

In both cases are missing the cards within my budget... Oh since the compute tales hours, I wont OC the gpus

Win 10
EVGA GTX 980 SC ACX 2.0 - not overclocked
8GB 2133 RAM
i5 3570k

Device 1 performance: 1088.03 million samples/sec (GeForce GTX 980)
Total performance: 1088.03 million samples/sec

windows 7
sapphire r9 290x 4GB oc to 1148core 1500 mem
high align setting
ultra build density
photo set buildings
Device 1 performance: 1256.19 million samples/sec (Hawaii)
Total performance: 1256.19 million samples/sec

going to try and find a distro with kernel 4.3 native and post results

Windows 10
Fury X
Total performance: 1267.63 million samples/sec
That doesn't seem to be that much better than the 290x from the guy above though I don't know if I did anything wrong or not as I do not have anything experience with this program

it weird indeed... I was hoping for more, maybe win 10 driver not mature in that sense...

Well like I said I could have done something wrong. I'll test it on linux later but I doubt it will perform any better there as the linux drivers for GCN 1.2 cards are awful

I'm downloading it now, I've got a decently OCd TI, would you like stock and OC results?