ghostship2 posted at 8:55AM Sun, 27 October 2019 - #4368229

here is a link to that standard test scene

http://www.sharecg.com/v/86389/browse/11/Poser/Superfly-render-test-scene

Thanks!

caisson posted at 11:37AM Thu, 24 October 2019 - #4368116

mr_phoenyxx, the problem is that the 1080Ti and Poser are speaking different languages and do not understand each other. The 1080Ti was released in 2017, Poser 11 in 2015. GPU rendering is still relatively new and nowhere near as mature or stable as CPU rendering, and the various cards have gone through major changes internally over time. It could be that the slowdown you've seen is to do with changes in the graphics drivers. It isn't Poser because Superfly has not been updated since the initial release. Octane Render is exactly the same - when I was using it, every time a new round of Nvidia cards got released it would have to be updated to work properly. At least you don't have a new RTX card as they just don't work at all.

So I think that both Poser and the 1080Ti are working, they just cannot communicate efficiently. If your rendering speed is good when using two cards together then that is your best solution for GPU rendering right now; alternatively switch to CPU. The long term solution is for Superfly to get an update (and I hope that it's high on the dev teams list).

I am aware of all of that. I've had similar issues in other software packages with updating NVidia drivers and seeing changed behavior. However, that is not the case this time. What you are saying does not make sense. I was already running Poser 11.1, which is older than Poser 11.2 as you indicated. I'm not sure when the 1080 Ti was released, but I will take your word for it as far as Poser 11 being older than the 1080 Ti.

My NVidia drivers were already running the latest iteration. My Windows was already as up-to-date as it could be. Those updates were actually done the week before. And my 1080 Ti was working properly in Poser 11.1 with all those updates installed. I am running the exact same drivers, the exact same hardware, the exact same OS as what I was using with Poser 11.1. The only change was moving to 11.2, which then broke my 1080 Ti in Poser. So yes, it's a Poser issue, as that is the only thing that changed. They may not have announced that they changed the version of Superfly, but they have clearly done something.

Which I mean that isn't surprising. It's a different version. Obviously they have changed things. I am also not 100% convinced that it's the 1080 Ti. What I am saying is that there is some combination of specific things happening on my computer that has broken GPU rendering in Superfly with my 1080 Ti by itself. At the moment, the only tests I can think to do seem to indicate that the problem is the 1080 Ti or its drivers. But the change wasn't done in the hardware or in NVidian's drivers. The change was in Poser. So in my opinion, Bondware is responsible for fixing it.

EClark1894 posted at 10:51AM Wed, 23 October 2019 - #4368099

How many cores are you using?

That might apply to some of the people that are saying they see slow down even on a CPU render, but that doesn't currently apply to my specific situation. I'm using GPU renders. But to answer your question, my Poser is set to use all 12 threads of my 6 core, hyper-threaded processor.

ghostship2 posted at 10:32AM Wed, 23 October 2019 - #4368034

I'm not missing the point. I'm running the same software that everyone else here is running and haven't noticed any difference in render times up or down. What I am reading is that the OP is using ridiculous render settings that WILL INCREASE RENDER TIMES to crazy levels. I suggest using settings that optimize light diffusion, lower noise and speed render times before we get all insane about 12 hour render times. Optimize your settings first, then if your render times are still crap you can gripe and moan about it.

There was a standard render test scene over at SM forum that someone posted a few years ago. It uses figures and props that come with the program and represents a typical Poser scene with figure, hair, clothing and a room with appropriate lighting.

I appreciate your suggestions. I do. But the render settings do not matter to this issue. I understand the point you are trying to make, but I can use the lowest setting available for the fastest render possible, not caring about the quality of the result, and my system will take ten times longer to render as soon as I tell it to use only my GTX 1080 Ti. It performs properly with a CPU render, with the 1050 Ti, and with the 1080 Ti and 1050 Ti working in tandem. It only does **not **work properly when I tell it to use the 1080 Ti by itself, and that is a problem. Especially when it's not a little slower. It's a lot slower.

That is the problem I am trying to solve. Adjusting render settings is meaningless to the issue. The effect occurs at all render settings.

I do like your suggested settings for high quality renders though. I tried them out, and I agree that the result is superior. Once this problem is solved, I will likely use your settings moving forward.

I also thought about using that test scene you mentioned. I remember it well from the Smith Micro forums. Honestly, I just didn't feel like going there to track it down. I felt the default La Femme scene was easier and faster to use.

Also, I'll share the results of the test I did with your settings. This is the La Femme default scene at 1440X1080 with 70 Pixel Samples and 6 Diffuse Bounces. The exact settings you suggested.

GTX 1080 Ti by itself - 26705.03 seconds. Almost 8 hours to render!

GTX 1080 Ti + 1050 Ti - 1387.46 seconds. 38 minutes.

Whatever is wrong with the 1080 Ti or my Poser, it needs to be fixed. This is a serious issue.

Also, just out of curiosity, I ran the same test with my usual settings of 100 pixel samples and only 4 diffuse bounces. Render time with the 1080 Ti + 1050 Ti was 2381.48 seconds. So my usual settings are about 20 minutes slower than your suggested settings, and provide a slightly dimmer result. As you pointed out, your settings are superior. I'll be using them moving forward.

But I still need this very odd issue with 1080 Ti's (or with my specific install) figured out.

ghostship2 posted at 3:58PM Tue, 22 October 2019 - #4368021

rendering 10,000 samples per tile ain't gonna be quick.

mr_phoenyxx posted at 3:43PM Tue, 22 October 2019 - #4367993

@ghostship2 I think you missed the point? I wasn't trying to get a high quality render result. I was trying to come up with a standard test that would complete quickly in order to compare render times. A standardized test is the only way to do valid comparisons in order to determine what's going on.

Correct. Before the 11.2 update, it would take roughly 4 to 6 hours depending on the scene. Now it takes days. I'm aware that I don't use the most optimal render settings, but they worked perfectly fine before the update. Now they don't. My render settings haven't changed. Something about Poser has.

@ghostship2 I think you missed the point? I wasn't trying to get a high quality render result. I was trying to come up with a standard test that would complete quickly in order to compare render times. A standardized test is the only way to do valid comparisons in order to determine what's going on.

willdial posted at 11:44AM Tue, 22 October 2019 - #4367917

@mr_phoenyxx

The 1080 Ti should be faster than the 1050 Ti. I saw some posts from a couple of years ago with people complaining about slowness in Cycles with the 1080 Ti. Unfortunately, they don't mention a fix.

I'm not seeing any slowness from 11.2 update. But, I'm running a regular 1080 on Windows 7 with older drivers. I only update when absolutely have to update.

As for the 100 Pixel samples, Superfly is noisy especially for indoor scenes. To get a clean image, you have to crank up the samples or use a denoiser.

Thanks will! It hasn't been an issue until 11.2. Previously my 1080 Ti was faster than anything else on my machine: CPU or 1050. I've tried doing lower pixel samples with a denoiser. It doesn't look good in my opinion, which is why (as you commented) I crank the pixel samples up in order to get any kind of decent render.

ironsoul posted at 11:41AM Tue, 22 October 2019 - #4367924

Most likely a futile, "throw enough mud at the wall" type suggestion - considering the impact of the dual card. Could Windows and Poser be offloading display and OpenGL rendering onto the 1050 leaving the 1080ti free for ray tracing only? One thought is these two functions are throttling the 1080ti. It is possible to control how CUDA is used in the Nividia control panel 3d settings so it may be possible to mimic this behaviour when using one card only.

Maybe? That's over my head to be honest. But I mean if I'm not using the 1050 Ti to render, meaning it's set in Render Settings to only use the 1080, then the 1050 should not be involved in the render at all. At least in my mind anyway. Also, I don't think it explains the behavior that the 1050 Ti is faster on its own, and the two cards together work correctly and are way faster than either card alone.

DarthJ posted at 7:16PM Mon, 21 October 2019 - #4367893

mr_phoenyxx 100 pixel samples ? That is crazy. No wonder the machine takes days to render. For quality I use 20 pixel samples, maximum 30.

I knew someone would say that. :)

Here is the issue though, this exact same machine before the update to 11.2 would render a 1600x1200 scene in about 6 hours.

My personal experience has been that with anything less than 100 pixel samples, I don't get a decent result. I would be curious to see all of your render settings to compare.

Well the plot thickens, and I may have been wrong about some of my determinations. See blow:

So I also use a GTX 1050 Ti, but I don't usually use it to render. It just drives my monitors. These tests were performed using the same test scene above (Femme default scene at 1440x1080), with the same render settings (GPU Medium Quality).

1. As you can see from the log, if I use the 1050 Ti and the 1080 Ti together then I get a much more reasonable result: 37.56 seconds.
2. The GTX 1050 Ti by itself actually renders quite quickly as well: 111.17 seconds.
3. The GTx 1080 Ti by itself though, then takes 691.01 seconds. So the 1080 Ti is operating much more closely to its proper abilities when working with the 1050 Ti.
4. Out of curiosity, I did a CPU render with the same "GPU Medium Quality" settings which gave 238.69 seconds. Much faster than the GPU render, though not up to the 1050 Ti speeds. But then I was using GPU settings that were not optimal to a CPU.
5. Using "Medium Quality" (just regular Medium Quality, not the GPU one) I get a much better result on the CPU of 43.98 seconds.

So I may be wrong that my CPU renders are out of whack as well as my GPU renders. I said that based on some test renders I did using CPU Ultimate Quality settings. Those tests weren't performed with the settings I would normally use for a CPU render. So that is my mistake I think. CPU renders seem like they might be fine on my machine given results above.

So this begs the question of whether it's an issue with 1080 series cards, or is it related to having multiple cards in your machine?

Can anyone confirm that they run multiple GPUs, not SLI, in a single machine and do not have any render time slow downs?

I've submitted a ticket about this issue. Meanwhile, any help or suggestions for things to try to help me narrow down this problem would be appreciated.

For instance, is your Poser still installed into a Smithmicro folder? Or have you installed it in a different location? Specifically, my Poser is installed into C:Program Files-Smith Micro-Poser 11.

Did a test similar to GWild. Femme base scene that loads with Poser now at 1440x1080. Used the "GPU Medium quality option" that is one of Poser's included options for render settings. I wasn't sure what GWild meant by "default render settings". See images below:

Render settings:

19, nearly 20, minutes to render this scene is ridiculous. Logs:

Result, which isn't a very good quality render. But then I was using "Medium" so I didn't expect a great result. I am including this in case people using the same settings get a far superior rendering result. I've also censored out sensitive areas just to be safe on the forum.

GWild99 posted at 4:36PM Mon, 21 October 2019 - #4367789

Ditto on super slow super any renders ... Smith Micro Poser 11 Pro was slow; Bondware is probaby 10x slower on CPU; and my GPU renders even slower. It'd be funny, except it isn't.

@Gwild Did you ever use Reality for LuxRender for Poser? Do you use EZSkin3?

@rhia474 If you were seeing slow down, you wouldn't need to time it. It's so drastically, blatantly slower that you can't help but notice. Also, oddly, I'm experiencing the opposite of you. I have found that pops, figures, and saved scenes all seem to load faster now. Just my rendering is so slow that Poser is unusable at the moment.

@Gitterati3D What CPU do you use? I'm running an i7-8700 K with 32 GB of DDR4 RAM. Historically I found my CPU renders were just slightly slower than GPU rendering with the GTX 1080 Ti. So I converted to almost all GPU rendering, as I could also continue to use the system when GPU rendering. Whereas CPU rendering pins the CPU to 100%.

@seachnasaigh I also have 2 remote machines. One is a Dell R610 with 2 x Intel Xeon X5670 6-core processors and 64 GB of RAM. The other remote is just a crappy little Windows 10 box that only manages to put out 1 or 2 renders a week, but there's no reason not to use it. Obviously both of these remote machines do CPU rendering. They have also slowed down just as much, if not more, as my main machine.

@ironsoul Thanks!

@SamTherapy Same here. Nothing has changed on my machine or in the things I am using. It's just the version of Poser.

@caisson I didn't really think it was a global issue, as I hadn't seen any threads about it. But sometimes you just never know, so I needed to ask the community. As for my logs, you can see them below. The pertinent part is at the bottom. 68643 seconds rendering, and it was only done 12 out of 35 tiles. That is far more slow than my original estimate of three times slower. This is more along the lines of 10 to 20 times slower. Completely unusable.

@HartyBart I did make sure that "use existing preference files" was checked during the upgrade to 11.2. The most recent update though, uninstalled Poser completely and then re-installed it. I believe it still offered me that option though, and I still checked "use existing preference files".

As to what I meant about the progress bar, please refer to the following image:

Please take note of the "Sample 4131/10000" part of that picture. Prior to the update that 4131 number would jump by anywhere from 5 to 30 each time it moved, just depending on what part of the scene was being rendered. Backgrounds and walls would jump by like 30 each time, while hair would only move by 5 numbers at a time. So like 4131 would become 4136, then 4141, 4146, etc when doing very detailed hair models.

After the update, that number moves by 1's and 2's when rendering the least demanding part of the scene: backgrounds and simple objects. The progress bar reflects what I'm saying as far as my render speeds being inexplicably 10x slower or more. The reason I say, "It's different" is that simple, grey backgrounds shouldn't be "moving the dial" by 1's and 2's. That progress bar should be lightning fast on a simple grey background. So I don't know if the weird bar behavior is an indicator of what's wrong (that my card isn't rendering large chunks at a time like it should), or if it's merely reflecting the much slower rendering progress.

Please keep in mind that I have tried CPU renders both locally and on **two **separate remote machines, as well as a variety of different GPU rendering options. Everything I do on every machine is an order of magnitude slower. Again, so slow that the software is quite literally not usable. It takes days to render the simplest scene.

Here are the render settings that I am using. Some of you are going to look at them and wonder if I'm insane. That's fine. I know they aren't the best, but they the settings that I find work the best for me after weeks and weeks of experimenting when Poser Pro 11 was released.

If you have additional questions, then please let me know. If you have any suggestions of things to try, then please let me know also.

By the way all of you reading this, I would really appreciate if you post even if you aren't experiencing this issue. I'd kind of like to get a feel for if this is a global issue for everyone, or if it's limited to only a few people.

Are there people that have not experienced this? Are there some, or a bunch, of you that Poser is rendering at the exact same speed? Or maybe even faster?

Please let me know!

SamTherapy posted at 10:41AM Mon, 21 October 2019 - #4367787

I don't use Superfly but I found Firefly to be significantly slower than before, when using settings I'd always used in the past.

I was concerned that Poser may trash my settings when I installed 11.2, so I did the old fashioned thing of screenshotting my settings for various types of renders. My final quality settings are now painfully slow. Admittedly, they weren't exactly sprinting along before, but now they're like a sloth on quaaludes.

Well that's concerning. I had hoped it was only SuperFly.