I'm hoping someone can clear this up for me before I go insane... I used to run poser on an AMD FX8320 paired with a GTX1060 (6gig) graphic card. It got to feeling slow, and I saw a 1080ti (11gig) on ebay that wouldnt require a mortgage, so I grabbed it. Card works, but I found the performance to be rather lackluster in the FX machine. I was pointed to a site that said my existing hardware was bottlenecking the 1080 by as much as 50% sooooo... One Ryzen2700 later, and I'm still quite unimpressed. So, I put the FX and Ryzen on the desk at the same time, created the same scene on both- empty space with SAKitty standing there, both rendering at 800x600 and with the same Superfly numbers Result? FX machine storms home in first place, 130/130 tiles done. Ryzen? Still strolling through tile 39/130 What gives? Both machines are running Windows 7x64, yes I got that wedged into a Ryzen system, thank you Gigabyte for supporting Windows7 with a USB fix. Both machines have 32gig of ram, obviously the FX has considerably slower ram (1600 VS 3000) The FX also has 4x8 while the Ryzen has 2x16 sticks. FX system has been running the same copy of windows for about 5 years, Ryzen obviously is fresh. I've run the Valley benchmark through both machines on the ultra HD setting FX scores : FPS 58.3 Score 2439 Min FPS 18.8 Max FPS 100.1

Ryzen scores: FPS 101.0 Score 4227 Min FPS 22.2 Max FPS 147.2

Looking at the leaderboard for the benchmarker, that score for the Ryzen is only about half of the score of the first user with a 1080ti that I found. Did I pick up a dud card here or have I missed something? Both machines are running the latest nvidia drivers, updated thismorning. The only other difference, is the fx is on an HD monitor and the Ryzen is running a 4K monitor (which used to be on the FX)

If you're rendering using GPU in Superfly what matters most as far as I understand it (based on using Octane) is CUDA cores which are the processors that handle the render calculations. The 1080ti has over 3,500 while the 1060 has about 1,200. Both are based on the Pascal architecture (I mention this because Superfly has not been updated in a while and doesn't work apparently with the latest cards based on Turing). When you choose GPU in the render settings Poser will transfer all the geometry and textures to the card, which is where amount of VRAM comes in, then the card will handle everything from there - the CPU, RAM etc will be redundant at that point. The 1080ti has a lot more CUDA cores so should be significantly faster than the 1060. If you swap the cards over i.e. keep everything else the same in the two systems, do you get the same result?

That's the extent of my tech knowledge! Maybe try running GPU-Z and seeing what it reports?

I returned the EVGA 1080 TI SC2 I was using, believing it to be defective. Today my Gigabyte Aorus 1080 TI Extreme edition arrived, and its a bit better than the EVGA A =BIT= better... Doing the same basic render, it finished in about 402 seconds, where the EVGA took about 440 seconds or there abouts. My old machine with its 1060 did the same render in 115 seconds. The 1080 should be something like 3 times faster than the 1060, not four times slower, I'm at my wits end here, what gives!? I have just spent £1200 on this new machine, and its considerably WORSE than my five year old machine that cost half as much back then............. I'm honestly thinking of just giving up on rendering all together now, I couldnt afford this machine when I thought it would be a massive upgrade, now its as good as a damn paperweight unless I can get it to run like the numbers say it should =_=

I'm REALLY hoping its not down to the Windows version. I do =not= want to go to Win10

There is much detail in your post so I've probably misunderstood something but why 130 tiles for the gpu render? Default GPU size when using Superfly is 256 which suggests you're doing a very large output image or you're using an inefficient tile size. A 1080 should be able handle 512 tile size for most scenes. Details such as output render size, render settings, GPU utilisation during the render and the poser log would be useful.

When I say 130 tiles, I mean there are 130 chunks to the image, not bucket size. I'm keeping the test render to 800x600 with a bucket size of 64, because thats what my old system seems happiest running at and I want to keep the duel between old and new fair. You are right about the bucket size though, at a bucket of 64 it takes about 266 seconds now, a bucket of 512 took around 169, while bucket 1024 only took 158 seconds.

There are the render settings I'm using, basicly GPU medium quality, but now with a bucket size of 512. Poser is 11.2.296 Computer is a Ryzen 7 2700 with default clocking, 32gig ram (16x2) Gfx card is a Gigabyte Aorus GTX180ti Extreme edition, 11gig which is running a 4K monitor @ 60hz over Display Port. I know the card is working ok, since it can throw around an older game like Borderlands 2 at 1440p @ 260+ fps, and Farcry New Dawn on ultra settings 1440p gives me around 100fps average. What I'm thinking is that its down to the CUDA drivers nvidia is putting out now. They come with CUDA10 now, but I think poser was written for CUDA8? I've seen in some three year old forum threads that people have been going back to nvidia driver 373.06 or earlier, but when I try installing that version I keep running into 'cannot find compatible hardware' For the moment I have version 384.76 installed which is CUDA 9.0 and the performance has improved somewhat, roughly 200% what it is under CUDA10 but its still slower than my old machine with its GTX1060. Incidently, tried the same render under Windows 10, it is marginally faster, but not worth having windows 10 for =p

There were problems with the nvidia driver but that was resolved sometime ago. I tried a 800x600 render of SAKitty on my machine (GTX1070 founders edition, i6700K, no overclocking on GPU) and came up with much different render times
Bucket size of 64 was 109 seconds,
Bucket size of 1024 (or 512) 34 seconds.
I used your render settings but the scene file could be quite different so its not a reliable comparison.
Have put my GPU-Z stats below. The blue bars on the percap line of the sensor screen shot are a voltage reliability issue. The graph shows two renders 64 bucket size followed by 1024 bucket size

On reflection your 1060 @64 matches my 1070@64 quite closely suggesting our test scenes are very similar, if my 1070@1024 is 38s your 1080Ti@1024 should clearly be much better than that. The GPU-Z utility has a load test function that can be used just to check the card is not being thottled due to power or heat reasons so might be worth a try if not already done. Another thought is to use Blender benchmark renders to confirm card is ok. Really we need another 1080ti user to chip in but the only ones I've found are using Octane rather than Superfly.

I really should look into this octane thing, I've heard about it but I've never actually seen it. I ran GPU-Z during a render and my PerfCap is showing a solid VRel bar which puzzles me, my PSU is a modular, and I have each of the power connections to the card on its own cable so voltage should be steady?

GPU-z during the render of 'duel' about halfway through And this is GPU-z after the render had finished

I downloaded the blender benchmarker, did a quick bench which rendered bmw27 and classroom.

My results were :

bmw27 - 109 seconds [average 343]

classroom - 257 seconds [average 1053]

Averages taken from this page - https://opendata.blender.org/query/time-per-scene

I'll just drop this in too incase it helps at all. I notice PhysX isnt ticked but its definatly installed

Oh by the way Ironsoul, the 'duel' scene is the factory default scene with Andy/Femme deleted [which is set as my prefered scene so I always start with an empty space] then SAKitty is loaded. No posing, no light changes, no material changes, using Main Camera in its default position. Ground guide still on. After loading Duel and using a 512 bucket size on my 1060, it rendered duel in 43 seconds. Which makes my 1080ti look even worse =p

Graytail posted at 10:25AM Mon, 07 October 2019 - #4366256

Oh by the way Ironsoul, the 'duel' scene is the factory default scene with Andy/Femme deleted [which is set as my prefered scene so I always start with an empty space] then SAKitty is loaded. No posing, no light changes, no material changes, using Main Camera in its default position. Ground guide still on. After loading Duel and using a 512 bucket size on my 1060, it rendered duel in 43 seconds. Which makes my 1080ti look even worse =p

Yeah, that 1060 is nipping at the heels of my 1070 all too well, guess I should have waited for a non foundation version to comeout :)

My Blender benchmarks for the 1070 are

bmw27 - 92 seconds
classroom - 343 seconds

Not sure what to make of the differences but I would expect the 1080ti to be better than the 1070 for the bmw27 if working ok.

If you've not already done so might be worth searching for a standalone benchmark software that covers render performance to see if it also shows a problem.
Another thought is to look at the overclocking on the GPU card to see if it can be turned off just to see if that's causing the problem.