This is my first post, but I have been using Poser for over 20 years now to create some technical renders and see how the software is evolving. I have a workstation that is mainly used for creating assets for game engines like Unity and Unreal. So the following information may not apply to many people.

But the question needs to be asked as I just upgraded my GPU to an RTX 3090. I would have hoped for very fast render times, but it appears this is not the case. My cpu is an i9-10980XE which is a pretty fast processor.

The attached image was rendered in:

Firefly with CPU:  33.45s render time

Superfly with CPU: 180 sec

Superfly with GPU: 290.6 sec

Normally one would expect the opposite render times, but here it is.

Any thoughts as to what is going on?

There is some more information needed. Here are the settings, for Superfly almost all default setting and same for CPU and GPU.

First of all, from my personal experiences, hair props/characters may explode the render time.
Here is a portrait, created at the same dimensions as your picture, using Poser's default "Ultra non-branched (GPU)"

The hair is Aeon-Soul's Desir hair, which is time-consuming during the computations.
My GPU is an old RTX-2080Ti and it took 13 minutes using the Optix mode, probably near twice longer using the default mode.

No lights involved: only an HDRi on the background node.

From what I've read online, your 3090 should be at least twice as fast. Even though you have a pretty fast CPU, your video card should kind of shrink the render time down to a microscopic level compared to the CPU. I mean: the first time Poser12 supported the 3090, the crew did a test with a scene, it was something like 5 to 6 minutes for the CPU down to 10 seconds with the 3090... (these aren't the right values but that gives ou an idea)

That's why I'd like to say: check the following points.

First, check the video card's drivers. I never used NVidia's default driver, only the Studio version. The standard version is often a nightmare in terms of stability...

Then, something else I've remarked: if Photoshop is loaded, with a scene and not minimized and "empty", as it can use the GPU as well, I can see a real difference during the computations: consider this as well depending on the other programs you are using in your process.

Last but not least: I suppose that you're on Windows. Don't forget once you know that you'll made a series of computations to restart your Windows station. Don't turn it off/on: reboot, it will refresh its resources.
Example: for my job, I'm using PyCharm, which is a real resource-hungry program. After a few days of using up to 3 instances of PYcharm, 1'234'567 tabs/windows of different browsers etc..., rebooting my PC makes a really noticeable difference depending on what I've been doing during work hours (I'm working at home on a browser-based python-written framework)

The reason I found my results strange is that I used premade Blender benchmarks to test the new Video card.  In that scenario, there is no comparison. I also usually render in 4K resolution, the image I uploaded was a reduced res screen snip.

Victor benchmark

CPU  7:38 min

3090 cuda  2:20 min

3090 optix 57 secs

I expected same in Poser. I tried to use optix, but the preload time before it even started took longer than the 33 sec CPU render time.

I assume the Superfly settings are wrong.  They don't recommend branched path tracing with GPU, but if I don't use it I need a ton of rays to eliminate the noise.

Frankly in many cases I prefer the look of the Firefly render because it is noise free. I will use your posted setting of Superfly and see if the 3090 will be faster than the I9 CPU and report back

Here is my new results. Using these settings

Using the same 2000x2180 pix image as before, I get the following:

3090 Cuda: 194 secs, branch tracing off

i9 CPU: 631 sec, branch tracing off

i9 CPU: 218 sec, branch tracing on using 4 rays

3090 GPU Optix: 152 sec

As for visual comparison, I don't see any difference between CPU branch tracing on and off.

Conclusion so far: The render times of CPU with branch tracing on is in the same ballpark as GPU Cude and Optix. GPU is not 5 times faster as in Blender 3.21

So the whole reason for this exercise is to cut the render time for a 1 minute 60 fps lipsyn animation for a customer. A total of 3600 frames.

You can see that the 33 sec Firefly CPU render seems very attractive even though I think the ray-trace render looks better.

Any suggestions to managing the GPU render settings?

There's something you should experiment with your card. I did this for mine (11Gb RAM) and I found that a bucket size raised up to 512 was perfect. If I'm not wrong, lower  (such as 64) values are for CPU-based renders.

My scene contains a pair of shoes, a chair, the hair prop and a Vic4-equivalent (with 6x 4K bitmaps plus the bump versions), all with SubD at 1, plus a 2k HDRi bitmap for the lighting, and it took my old system 9 seconds to push everything to the video card.
My feeling is that given the fact that your system is really newer, something is slowing it down (HD access, PCI-Express bus, or whatever)
Note that I've installed my Poser and all its runtimes on a dedicated SSD drive.

Thank you Y_Phil. Now we are getting somewhere.

a 512 bucket gets me 122 sec instead of 152. And rendering a 1080 x 1024 pix lowers it to 36 secs.

I use SSD and MVme for all my 3D work, then do backups on HDD. I find my MVme drives get too hot for two reasons. They sit under video card and if you flog them with compiling a game file, the temps reach over 65C, even with the heatsink on top. I never really thought that the EVGA RTX 3090 would generate so much heat. 420W does not seem like much when you read about it, but when the heat blows in your face ( I should put the computer on the floor), it is annoying. Sometimes I long for my 1080 ti. I mainly got the 3090, now that they prices are not sky high anymore, just for the amount of Vram. I have some scenes with 6 or more avatars, and all that sucks up a lot of Vram during rendering.

Y-Phil mentions tweaking 'vols & buckets', and I fully agree. Also, that may take some experimenting to find the 'sweet spot' for each type of job...

This PC has a pair of now-ageing GeForce GTX 750 Ti cards driving four (4) displays. And, yes, a low-flying duty-cat who's just deleted a so-carefully crafted paragraph. For 'modest' scenes, Superfly, progressive, I've found 'vol-bounces & buckets' may both be set to 1024. If cards max-out and scene is blank or missing chunks, I back those off via 512, 256... down to 64, which is what CPU-only renders prefer, be they this PC's weary FX8350 (8 threads) or network-render 'Box' nimble Ryzen-7 1700x (16 threads)...

I have a bunch of 'presets', making it easy to configure render for type of job...

This set-up is no match for a modern GPU card with 'studio' driver complemented by Poser render support, but building 'Box' was a lot cheaper than upgrading to such... 

if you have assembled your PC yourself, dare I suggest another case? One with a better ventilation, especially with a 3090.
In my case, there are 3 12" fans on the bottom, pulling fresh air inside, 3 on the top, pushing the out  and on on the back.

The SSD on which Poser is installed is between the video card and the 3 fans on the bottom...

Not sure that one day  I will be able to understand what engineering means... I mean: that's a true  evidence: it's one of the worst position, especially since the video card manufacturers insist on having GPU's fan directed downwards

And for what concerns Poser's speed when a scene is loaded, here are some parameters you should consider:

I keep the preview resolution up to 4096, which is the size of most bitmaps I use. And ono the left (hardware shading supported) I keep these options and settings.
One parameter that I've never understood the meaning: PBuffer or Scene Window.

I never looked at that tab. It still had the original install defaults, so I changed them.

I did read this page:

http://www.posersoftware.com/documentation/12/Poser_Reference_Manual/Rendering/SuperFly/SuperFly_Render_Settings.htm

but it is not clear on how to optimize video card render times without actually setting each of the parameters to see if you like that quality.

In summary, using the same settings except for bucket size I am surprised that my fancy new video card is not that much faster than my CPU.

3090 GPU best times: 122.5 sec

i9-10980xe best time 218 sec.

each image look to me indistinguishable in quality, and GPU is not even 2 X faster.

So why in Blender 2.79, using the same Cycles engines, do I get at least 3 to 6 times faster render times? Even more so in Blender 3.x

This is why I consider this to be a technical question.

As you can see not a lot of room for extra fans. I have to keep the side open, back is off because I am doing some rewiring. The drive that gets hot has the silver heatsink just below video card.

Yes I am sucker for drive space. Three 6TB HDDs and 3 assorted SSDs in addition. Knowing that my HDD will still have data after 20 years, as opposed to bit rot on SSD after 5 years unpowered, means I will still use them. Also I use DVD and bluray disk for archival storage, that's why I bought this case. Had to shoehorn my CPU cooler as you can see. Tie-wrap holds up the video card to keep it from sagging.

Not sure if I understood correctly: "20 years" and "HDD" in the same sentence? OMG
I changed my hard drives every 2-3 years, when they're still ok, as in Switzerland the manufacturers must guarantee these items for 2years. I don't trust any brand nowadays.
For my external pocket-size drive: I trust them up to 4, 5 years, rarely more, even though they are disconnected most of the time.

For what concerns your case: 3 Sata drives plus 1 RTX3090 = quite a bit of heat.
The problem is: often, cases are thought for standard, office-like computing. If I were you, I'd invest in a better case, BeQuiet for example. I think that you really need something that pulls air inside + something that pushes it outside.
And check that your case has enough place around it.
Having built 6 PC's overtime, I know the time it takes  

Concerning the render settings: I'm doing test renders using this:

I try the OptiX mode, than I switch back to the standard GeForce mode if some props are causing crashes (it truly depends on  the materials)
Then, Poser's "Ultra NonBranch (GPU)" is a good start.
I don't know for Blender, but with Poser, don't check the "Branched Path Tracing" option using a GPU

Thanks Y-Phil. I will look at the various preset.

I never had any crashes with GPU rendering, even my old GTX 1080 ti has 11 GB of Vram. In Unity3D I use GPU for lightbaking, and sometimes it filled it up so program had to revert back to CPU.

Off topic: A 6TB 7000 rpm WD has power consumption of 8 Watt. My video card with 420 Watt under full load and CPU with 320 Watt under full load dwarf that. Since I need 5 1/4 cages for my Bluray burner and my HDD hotswap bay, in addition to the hard drive bays, there are virtually no other cases available with that feature other  than one Corsair case I found. I don't want all that stuff external, I have enough wires already. I am a retired RF engineer and I don't want bluetooth or WiFi connections to my work computer out of principle. No matter how good the protocols are today, but there is always outside interference.

I run my HDD for roughly 50,000 hours or until I need more storage. Because electricity is relatively cheap here in Atlanta, I run my computer 24/7. The killer of HHD are the startup current and the heat cycling from being turned off and on. By 20 years I mean I will put a drive on a shelf and sometimes I need old files from 10 or so years ago. After all I have been using Poser since version 3. They had a revolutionary user interface back then.

Ok, great!

---

Ah, some-one else with a 'KiloWatt' Corsair...

Does that have the USB link to mobo for dash-board ??

In the official Poser 12 launch webinar recording, direct from Charles Taylor:

39:53 min: the new Cycles 2 “Adaptive Sampling works best with smaller bucket sizes”.

40:27 min: “branched Path Tracing works well for CPU renders, but not GPU renders”.

After more tests I have finally come to the following conclusion. If you have a ton of cores on CPU, Firefly will most times give you the faster render time.

Firefly cannot really be compared to Superfly, as Superfly is a real ray-trace renderer and will always be slower.

Given the comparison between CPU and GPU on a complicated scene, the optix enabled render will always be the fastest.

Based on the image below, I have timed the following times

GPU optix 256 bucket: 343 sec

GPU optix 512 bucket: 331 sec - only saves 10 secs

CPU 128 bucket: 3469 sec

As can be seen GPU is 10 times faster! Total Vram usage 5.5GB

Elven castle is my own model imported into Poser. Ray traced reflections are well worth the effort.

My 1000W Corsair is not USB enabled. It is the cheaper version.

3d-designer posted at 1:34 PM Sat, 30 July 2022 - #4442060

Yes, I said the GPU is 10 times faster, exactly what I was looking for. But only for Superfly.

For Firefly, the CPU is not that much slower, but image is not as good. But it takes only 420 sec.

So if the image does not fit into GPU memory, than you are stuck with CPU rendering. Above image used 5.5GB of Vram. If you have a card with only 4GB, then you cannot render in Superfly, not sure. I know in Blender, it will kick you out of GPU render.

3d-designer posted at 5:08 PM Sun, 31 July 2022 - #4442119

Yes, I said the GPU is 10 times faster, exactly what I was looking for. But only for Superfly.

For Firefly, the CPU is not that much slower, but image is not as good. But it takes only 420 sec.

So if the image does not fit into GPU memory, than you are stuck with CPU rendering. Above image used 5.5GB of Vram. If you have a card with only 4GB, then you cannot render in Superfly, not sure. I know in Blender, it will kick you out of GPU render.

Now i know where I was wrong, sorry about that: I read that incorrectly lol
That being said: with my 3D girls: most are based on Sasha16, and most of the time, they have the same bitmaps for some parts f the bodies (bump, no-makeup skin parts used for the SSS effects, etc...), that reduces the consumption of RAM
Furthermore: I'm not sure that 4k bitmaps are useful when it's not for close-ups.
My previous computer was equipped with a GTX980Ti (6Gb of RAM). At the time, Poser was able to use it only with 2 Vic4 naked, but I wasn't optimizing at the time.

Thank you for all your advice Y-Phil. Now that I have been pointed in the right direction I have been using it more.

BTW, I added a fan at bottom and it lowered my NVme drive temp.

Now this is another topic, I have a ton of Vic4 girls I bought over the years. Unfortunately to use the GPU render I have to rework every texture to look good. Then of course all the eyes are screwed up - all white. So I still use Firefly mostly, the exception was making animations where the extra effort to rework textures is worthwhile to get the extra speed.

3d-designer posted at 10:18 AM Fri, 12 August 2022 - #4442683

Thank you. I did download Ghostship's material. But it still required extra work if you need to change eyecolor, as you can see in the above render. I also now have Snarlygribblies scripts to convert any material for Superfly. Not tried it yet.

Between 190 V4 morphs from various artists, over 60 from Syltermaid (there was a time when a ton were free an Renderrosity). I guess she retired. But most likely have less than you. Anyhow unless the scripts work, I am not changing all of them over. Some textures look fine in preview but render completely black in Superfly. Now you see why I only recently took interest in GPU rendering. With 36 threads my CPU was not a whole lot slower than my 1080 ti. Most times my video card maybe was twice as fast.

All you have to do is run the Snarlygribbly script if the V4 materials show black. It will fix it. No manual tweaking required. Regarding eyes, if you need to change eye color, the only thing you need to change, once you applied Ghosthip2's eye materials is the Iris color map. I have several characters I use based on V4 and I don't even start without running those two scripts--they are part of my daily workflow (then again, I also saved out P12 compatible copies of those characters I use most often).

I am not trying to convince you: each 3D artist has his/her own process. But as I like simple things, I'm arranging the material this way, for example with Vic's irises:

Same arrangement for the sclera and the pupils. That way, it's easy to change the eyes' corresponding bitmaps: a simple drag and drop and all eyes' parts are replaced.
Furthermore: not all Vic's eyes are useable. In fact, few are as most have burned-in fake reflections, something I hate as much as I hate burned-in highlights in most hairs.
That way, it's easy to setup a few MC6's with those checked:

For what concerns the skins: I've stored a few MC6's fully setup, but not all: I like to interchange the skins.
Example with two characters: between both pics, I've swapped everything but the morphologies:

The result: with two characters, you make four

Thanks for the info Y_Phil.  I sometimes save material packages for changes to materials I have done.

And yes I like simple. I was looking at some of the material nodes people use on some avatars. I don't understand why some are so complicated. It just slows down render time. A lot of times I just delete nodes until I see a difference in the render. Many times I can remove many.. For GPU rendering not may nodes are needed for a great looking render.

Some belated thoughts: Please check using eg Task Manager that your CPU renders are actually using all the CPU core threads they should.

As I mentioned above and YPhil & Co seriously extended, CPU and GPU renders may need very different 'Vols & Buckets' settings. For Superfly, Progressive on this now-ageing twin-GPU system, I set 1024 unless Poser crashes / renders blank, when binary chop until happy. For CPU-only here or on network-render 'Box', 64 of each seems optimal, as more disables use of multiple CPU threads...

FWIW, I like 'Progressive' as I can tell within seconds if render will look good. Or, yet again, yet again, 'Needs More Light'...