3D Software Help and Assistance. Ask Away.

[osanaiko]

Are there any edge cases I should be aware of (that any of you might have experienced) with CPU+GPU rendering?

In a few quick test renders I was able to get about 1/3 render times by using both the 5900x and GTX 1080 I have, versus just the GPU.

1000x1000, 150 iteration test render done with GPU, 2m 51s to 2m 59s on two tests.
Same render with CPU+GPU, about 1m 2s +/- 1s on two tests.

2000x, 150 iteration, GPU: 4m 37s (1 test)
2000x, 150 iteration, CPU+GPU: 3m 1s (1 test)

I've heard that I should only be using the GPU as it is faster for rendering, and I'm sure that if I would ONLY use my CPU that would be true (vs. only using GPU). However, the speed boost is considerable, and I'm wondering if there's something I'm missing here? I have capped the CPU at 22 threads (out of 24) so that my computer stays usable so that isn't a problem.

Of course there are some differences outside, CPU rendering puts my cpu usage (and temp. + power) to max at 100%, about 75c, and 142W. Vs. GPU only at about 9%, 62c, and 71W.

I have also intentionally underclocked my GPU (because I CANNOT AFFORD TO LOSE IT) so that may be the cause for the rather substantial difference.

Those numbers seem a little strange from my understanding of relative cpu/gpu rendering power. Are you using Iray render engine? or 3Delight?

It was my understanding that even a threadripper CPU would only add <5% to render performance for Iray renders. There's an extensive thread on the official forums with benchmarks from various GPUs and configs, and if i remeber correctly there were some Hybrid results in there too.

 

[acusrola]

Those numbers seem a little strange from my understanding of relative cpu/gpu rendering power. Are you using Iray render engine? or 3Delight?

It was my understanding that even a threadripper CPU would only add <5% to render performance for Iray renders. There's an extensive thread on the official forums with benchmarks from various GPUs and configs, and if i remeber correctly there were some Hybrid results in there too.

Definitely using Iray, no mistaking it.

Maybe the difference turns to negligible with bigger and more detailed renders. A 1000x1000 render at 150 iterations (limited to 150 for benchmarking) is hardly a normal image render. When I upped only the resolution two 2x (meaning 4x pixels total) the difference shrunk from 280% faster to just 150% - which is about half. A linear decrease, surprisingly.

 

[acusrola]

Rendering with CPUs generally kick out a lot more heat and tend to end up with noisier renders (in Daz, at least.), but 75c seems awfully low - likely due to how short your renders are coming out. I imagine, unless you have a surprisingly good Cooler/AIO, those temperatures will rise considerably as you render longer. Sustained high-load heat will never be good for CPU, Intel or AMD, nor is constant 90-100% utilization. Not saying that's the case here, but I'd experiment with longer/heavier renders and see what those numbers look like afterwards.

I did do a few other "production" renders earlier, CPU did cap at about 80c - I'm not that worried since the throttle limit is 90c and the CPU steadied at about 80 (78-83, iirc). Heat is, of course the enemy of CPU longevity, but the 5900x already runs rather hot. And to be more accurate, it is the difference in heat leves iirc, not actually running hot in and of itself. When the metal / silicon gets hotter it'll expand causing cracks in the material, etc. Multiple cycles will destroy it, no doubt, but I don't believe it is going to be an issue even in (pro) consumer grade stuff. My old i5 4670K lasted from 2013 to 2021 (and still works, it is just shelved).

The other reason you see others say only using a GPU in Daz is because Daz (or rather Iray) is optimized for Nvidia GPUs, which is naturally going to make it a bit faster. Of course, this is all a moot point if you're not planning on doing VNs and just want to make one-off still renders or something of the sort.

That does of course make sense, but from my limited testing that I posted above, the render times did improve dramatically. I even did a test of putting the CPU to "Eco mode" which limited the 140W power limit to about 80, which also limited the heat level. The render that took about 1m 1s before, now took 1m 11s or so to complete.

 

[amster22]

Definitely using Iray, no mistaking it.

Maybe the difference turns to negligible with bigger and more detailed renders. A 1000x1000 render at 150 iterations (limited to 150 for benchmarking) is hardly a normal image render. When I upped only the resolution two 2x (meaning 4x pixels total) the difference shrunk from 280% faster to just 150% - which is about half. A linear decrease, surprisingly.

You should also do a CPU only render.

What is obvious is that your CPU rendering is faster than the GPU one for some reasons. There is no magic. Some amount of work has to be done. If it is done by two units that have the same computing power, the time will be divided by two (more or less). If the CPU is much less powerful, the rendering will be only marginally accelerated. To have a 3x acceleration, your CPU must be more powerful than your GPU.

 

[DogzBallz]

You should also do a CPU only render.

What is obvious is that your CPU rendering is faster than the GPU one for some reasons. There is no magic. Some amount of work has to be done. If it is done by two units that have the same computing power, the time will be divided by two (more or less). If the CPU is much less powerful, the rendering will be only marginally accelerated. To have a 3x acceleration, your CPU must be more powerful than your GPU.

I only render on CPU, this is because I do not have an Nvidia card, I am working on a visual novel have done around 1000 renders and have only had one fall over, this was because I maxed my RAM, I was only running 16gb at the time. CPU rendering is obviously going to be slower than a decent GPU, but has the advantage of rendering more complex scenes without issue, due to easy availability of memory, I am now running 32gb soon to be upgraded to 64. I have never had an issue with noise in my renders. I have attached a simple portrait as an example, admittedly there are a couple of firefly's but easily fixed with minimal post work.

 

[acusrola]

You should also do a CPU only render.

What is obvious is that your CPU rendering is faster than the GPU one for some reasons. There is no magic. Some amount of work has to be done. If it is done by two units that have the same computing power, the time will be divided by two (more or less). If the CPU is much less powerful, the rendering will be only marginally accelerated. To have a 3x acceleration, your CPU must be more powerful than your GPU.

After doing a lot of testing it turned out that my aggressive underclock of the GPU was the culprit. BUT, it doesn't always work?

I managed to make a 1000x1000 150 iteration test with my CPU in 71 seconds, consistently. And with my GPU, after removing the underclock, I managed to render the same scene in 58 and 61 seconds. But it was extremely fluctuating, and I cannot find out why.

From Afterburner I can see that my GPU is basically not being stressed at all, CUDA (in task manager) seems to be at about 100% but core clock is at about 800~MHz. When I managed to do the 60 second render, the clock was at 1200 or so. I cleaned up my computer, in case there was some dust or something that could've managed to block airflow and heat up the CUDA cores (I don't really know the 1080 core topology, but whatever). I'm gonna be running a render test again soon enough.

 

[osanaiko]

acusrola it's an interesting puzzle.

I found the link to the Daz3d GPU comparison chart:

You must be registered to see the links

There wasn't an entry for the CPU 5900x. closest comparable was a ADM3700x, which is apparently about 0.55x your CPU's compute power.

The most recent benchmarks submitted for a GTX1080ti rendering the test scene result: 3.9 iterations per second
The most recent benchmarks submitted for a AMD3700X rendering the test scene result: 0.6 iterations per second.
Extrapolating, I'd guess your CPU alone would do about 1.1 iterations per second.

Neithier of these benchmarks are with latest Daz/Iray versions, which can sometimes make substantial differences in performance (*although recently most of the gains are in the RTX upgrades found on the NV2xxx and NV3xxx GPU series).

Regardless, it goes with my expecation that even a relatively old (but high end) GPU (the 1080ti) is approx 3-4x faster at Iray rendering than a CPU.

When you get into the RenderIterations/Second/Dollar (i.e. cost effectiveness) then the difference becomes even more pronounced (not withstanding the current GPU price gouging/scalpers and the fact the 1080 is no longer manufactured new).

The RenderIterations/Second/Dollar comparison is:

Ryzen 3700x = 06.906
NV GTX 1080ti = 20.560

 

[acusrola]

-snip-

Yeah it's 99% likely to do with how GPU's handle multitasks VS. cpus. In the end I managed to figure out why CPU rendering was so much faster, but now that I found that GPU rendering could be faster I cannot use it. Here's a screenshot of some sexy graphs:

Between the red lines is an iray preview done in DAZ "editor" or whatever, i.e. not pressing Render. White lines are the actual (benchmark) rendering - which I cut short as it would've taken about 3 to 4 minutes.

As you can see, the preview actually taxes the GPU, core clock jumps to about 1500 or so, mem. clock is about 4500, power % is at about 40-60 and power is at 50-120W.

When I press render, VRAM spikes up as the benchmark I use is a portrait of a model at subD of about 3 or 4. Core clock doesn't go up at all and stays at a bit over 721. Mem is about 2x idle, 810 (VS 4500 max). There's no power usage, but the GPU usage is blasted at 100%.

I don't think it is the scene, because I managed to render the benchmark 2.5 times (one time the clocks went up mid-render???). 2 times at about a minute, and once in 90 seconds. Doesn't really make any sense.

 

[osanaiko]

perhaps visit the Daz3d thread I posted above and grab the test scene, then you'd have a directly comparable result.

 

[acusrola]

Tried it, didn't work - I think it has to be a driver issue or something. Gonna do a clean install, hopefully that will work.

Google hasn't been very useful so far. I can't say that DAZ isn't using my GPU, because it is, it's just so fucking low...

 

[acusrola]

9 minutes and 59.41 seconds for a pure GPU render. Thank fuck the clean install worked. I have to wonder what caused the strange performance regression in the first place. Can't really think of anything other than something in the NV Control Panel 3D settings.

 

[osanaiko]

Hmm. I don't want to offend but i'll ask a silly question:

Check in your "render settings -> advanced" - are you using GPU for both interactive and photoreal settings? (interactive is the editor when in iray draw mode, and photoreal is the actual rendering pipeline)

Another thing to consider is memory causing fallback to CPU. although you stated your test scene is pretty simple so seems weird to exceed the 11gb memory of a 1080ti....


Just out of interest, I just ran the benchmark scene from the Daz forums and got the following:

GPU: RTX3060-12GB (stock clocks)
CPU: Ryzen 7 5800X 8-Core Processor 3.80 GHz (stock clocks, Corsair H60 AOi cooler (which is not very effective btw))

on GPU only: 216 seconds total, 8.32 iterations/second
on GPU + CPU: 210 seconds total, 8.56 iterations/second on GPU + CPU

So in my case, enabling CPU as well as GPU results in sluggish PC interactivity, noisy fans and lots of heat, and a <2% improvement in render time.

Oh, and i've got "Daz v4.15.02" and "Iray RTX 2020.1.3" versions.

 

[acusrola]

I tried mixing and matching all manner of settings, CPU on off, GPU on off, both on, etc. but it never rendered fast on the GPU. Fortunately, now with the clean drivers everything's working just fine again.

Really goddamn jealous of that 216 second total, I managed it in 599.41. Guess having those RT ASICs really helps... Gonna need to see if I can buy a 3080 or a 3090 this year.

 

[osanaiko]

I tried mixing and matching all manner of settings, CPU on off, GPU on off, both on, etc. but it never rendered fast on the GPU. Fortunately, now with the clean drivers everything's working just fine again.

Really goddamn jealous of that 216 second total, I managed it in 599.41. Guess having those RT ASICs really helps... Gonna need to see if I can buy a 3080 or a 3090 this year.

Good to hear you got it sorted out...

Going from my previous rig with GTX1070-8gb -> current rig RTX3060-12GB was huge, both from a render-time perspective and because I don't have to deal with texture size issues as much anymore.

Bang-for-buck-wise, the 3060 is a real sweetspot right now - you could get 2 or 3 for the (street not MSRP) price of a 3090 and have better Iray performance. Of course you need a motherboard to support the multi-gpu and it doesn't give you the benefit for gaming that a 3090 does, but if iray rendering is your thing then it's a much better value proposition.

 

[acusrola]

So I've run into the most fantastical of problems, a butt morph is - in addition to affecting the butt - affecting the TEETH of my character. The FUCKING TEETH.

So, I'm wondering if there's a way to fix this?

Spoiler: horrific

You don't have permission to view the spoiler content. Log in or register now.

I think I've done it in the past with a G3F dev load model or something. But that's another thing, I don't have the G3(F or M) dev models at all, I only have G8 and G8.1. I should probably reinstall the G3 starter essentials?

For reference, the morph asset is GC Booty Morphs (For G3F) which aren't found here edit 1: found here: https://f95zone.to/threads/7886/. Could this be some elaborate troll for dirty pirates like me?

Spoiler: Some are so much more horrifying than others...

You don't have permission to view the spoiler content. Log in or register now.

Edit 2: After doing some testing it seems that the booty morphs affect a whole lot of morphs outside the butt area, so definitely avoid it unless it can be fixed. The G8F version: https://f95zone.to/threads/gc-booty-morph-for-g8f.20486/ affects the mouth too somewhat, though to a lesser degree.

The GC Breast presets for G3F and 8F affect the face too, but it is negligible and I don't think can be seen in renders. Obviously the butt morphs are fine as well, as long as your G3/8F keeps her mouth shut lmao!

 

[theKimb]

I'm having an annoying problem. I did some renders 2 months ago and wanted to work on that same scene. I load that scene, change a few poses and hit render. But somehow my lights are being (almost) ignored. I have no idea why. I haven't changed anything, checked the values with even older scenes where it was correct (not working in older scenes too anymore). No idea what's going on.

I'm using Iray Ghost Light Kit 3

You must be registered to see the links

Spoiler: Older image with working light

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: New one with broken light

You don't have permission to view the spoiler content. Log in or register now.

 

[osanaiko]

theKimb It's an issue with the latest Daz release (actually due to a change in the way Iray handles light emission from objects with very low cutout opacity).

There's some work arounds being discussed in this thread: https://f95zone.to/threads/daz-4-20-issue.108814/

I haven't checked on the Daz forums for a while but i recall at least one thread talked about an update that would partially fix it. Or maybe that was an update to the GhostLight products...

 

[V-Ten]

Need some help.

I converted a G3 character to G8 using Riversoft Character Converter and it worked fine and I created this image with the converted G8 version of the character.


The next day i tried to reload the character and the morph is all distorted in some areas and I can't figure out what happened.

I discovered that I didn't even need to use the character converter in the first place, when I apply the G3 character morph and textures to a G8 figure it loads but with the same exact morphing issue.

I understand if the morph is somehow simply not compatible with the G8 figure and could move on, but it worked the first time i did it.... and now I've been trying for few days to figure out what is causing this. If anyone has seen this in there experience, any advise would be appreciated.

thanks

 

[Liphisbus]

Need some help.

I converted a G3 character to G8 using Riversoft Character Converter and it worked fine and I created this image with the converted G8 version of the character.
View attachment 1697058

The next day i tried to reload the character and the morph is all distorted in some areas and I can figure out what happened. View attachment 1697060 View attachment 1697061

I discovered that I didn't even need to use the character converter in the first place, when I apply the G3 character morph and textures to a G8 figure it loads but with the same exact morphing issue.

I understand if the morph is somehow simply not compatible with the G8 figure and could move on, but it worked the first time i did it.... and now I've been trying for few days to figure out what is causing this. If anyone has seen this in there experience, any advise would be appreciated.

thanks

Something similar happened to me, I loaded a scene and my character was all distorted, I found out that another character morph that I was not even using it caused the problem, I just deleted the data file and it fixed, so the problem "probably" might be a morph that you recently installed.

 

[V-Ten]

Something similar happened to me, I loaded a scene and my character was all distorted, I found out that another character morph that I was not even using it caused the problem, I just deleted the data file and it fixed, so the problem "probably" might be a morph that you recently installed.

Thanks for that advise, I did try and delete the data from some resent installs and didn't fix the issue. Perhaps something slipped through that I didn't catch... at this point I there's no way to know in that mess of data.