Which render-plugin for Autodesk 3DS Max for a non-GCN FirePro card ?

[Gryz]

I know a little about hardware for playing games. I know very little about rendering in programs like Autodesk 3DS Max and Maya.

The son of a friend of mine studies to become an architect. He is using 3DS Max to renderer nice pictures of the buildings he has designed. Weirdly enough, not many students do this. And the university staff consists of old-school people who grew up with pencils and paper. Regarding 3D rendering he is on his own.

He has a laptop with a i7-2670QM cpu (@2.2GHz), 8GB ram and an AMD FirePro M5950 vdieocard. The problem is that the M5950 card is from the last generation of cards before the GCN cards.

He is using the VRay plugin to render his pictures. A render with 300k polygons takes ~8 hours. VRay supports hardware rendering via OpenCL. But it requires a CUDA card or a GCN card. As his FirePro M5950 does not have GCN, his rendering is completely done on the CPU. (Surprisingly, his i7-2670QM at 2.2GHz seems to be not much slower than my non-hyperthreading i5-3570K at 4.0GHz. He's maybe 20-30% slower only).

He wants to speed up his rendering.
Buying a new desktop-PC is not viable. Cost is a factor (he's just a student). And in 6 months he will go abroad for his studies for a year. Buying a new PC now for just 6 months is a waste of money.

So the question is:
What render-plugin for 3DS Max should he use, that can do hardware-acceleration on his non-GCN videocard ?

I looked at a few render-plugins.
VRay only supports GCN (and Cuda).
iRay and Mental Ray are written by nVidia, so I suppose they only run on CUDA (or the CPU). (Correct ?)

Then there is a render-plugin called "QuickSilver hardware renderer". This one only uses DirectX and not CUDA or GCN. Only DirectX and SM3.0 required. I suppose he could try this renderer.

Are there any other alternatives that are worth checking out ?
There seems very little documentation on the net about these things to get started. And I have found no active forums about 3D-rendering that could answer questions. Any suggestions ?
Thanks in advance.

 

[evilspoons]

I've been out of the CG rendering "game" for a while (like ten years now) so I haven't kept up to date with the latest and greatest renderers, but the reason you're starting to see GPU-based renderers at all is their increased programmability - OpenCL and nvidia's CUDA. Before GCN, the AMD GPUs weren't that easy to write renderers for (as far as I can remember) so there really aren't any around. NVIDIA's CUDA was more powerful and there was a lot of money behind it, which is why Mental Ray is a CUDA thing. (Mental Ray also has a CPU mode.)

That being said, the folks at the CG Society forums will know if there's something your friend can benefit from: http://forums.cgsociety.org/

It may also be down to rendering technique. 8 hours sounds like a hell of a lot for 300k polys. Might want to tweak some settings, perhaps do multiple rendering passes or simplify lighting. Again, CG Society's forums should have some advice.

 

[mdram]

3ds and open ray support distributed rendering
maybe he can borrow some pc power from some friends

http://docs.chaosgroup.com/display/VRAY3/Distributed+Rendering

 

[ThatBuzzkiller]

QuickSilver hardware renderer isn't really used for photorealistic rendering so it might be good if that's what your friend wants to achieve ...

If your friend was looking for a GPU accelerated ray tracer, then he can check out Indigo Renderer or Luxrender ...

 

[Gryz]

Thanks everyone for the suggestions. If you have more suggestions, please keep posting.

I was afraid that QuickSilver wasn't gonna be high-quality enough. (I had a quick look only, and the rendered pictures I saw weren't as spectacular as the normal raytracing pictures). I'll look into Luxrender and Indigo.

Edit: it seems Luxrender supports GPU-based rendering on the FirePro M5950 !
http://www.luxrender.net/wiki/GPU#AMD.2FATI
"You will need an HD 5000 series or newer card, or equivalent FirePro model."
The M5950 is based on the Radeon HD 6770M. So that requirement should be good !
Another nice thing is that Luxrender seems to have its own wiki. I think my friend's son can use some proper documentation. Luxrender might be the best solution.


I don't think distributed rendering is gonna help much. Kids these days all have laptops they carry around all day. I doubt he'll get the compute-power when he needs it. And I doubt he'll be able to set it all up and configure his friends' laptops.

I thought gaming could require some heavy compute- and gpu-power. It seems gaming is peanuts compared to this raytracing stuff.

 

[Timmah!]

He should stick to the CPU-based Vray. Dont know about the settings he uses and the resolutions at which he renders, but 8 hours sounds like quite a lot with that CPU... talking from personal experience.... i recommend following this tutorial>
http://www.ronenbekerman.com/hdr-image-based-lighting-3d-scene-setup/


Most GPU renderers are AFAIK physically unbiased, which means they are fairly heavy on hardware and the laptop GPUs are not exactly optimal in this regard. For Cycles/LuxMark (OpenCL) or Octane (CUDA) you really need top desktop card to get some acceptable render times. You need like several Titans if you want to produce clean renders in matter of minutes and not hours.

Interesting thing to know btw about not many students in his class doing it like him....i finished my studies 7 years ago and we were doing all the stuff from the third year onward almost exclusively with computers...

 

[ThatBuzzkiller]

He should stick to the CPU-based Vray. Dont know about the settings he uses and the resolutions at which he renders, but 8 hours sounds like quite a lot with that CPU... talking from personal experience.... i recommend following this tutorial>
http://www.ronenbekerman.com/hdr-image-based-lighting-3d-scene-setup/


Most GPU renderers are AFAIK physically unbiased, which means they are fairly heavy on hardware and the laptop GPUs are not exactly optimal in this regard. For Cycles/LuxMark (OpenCL) or Octane (CUDA) you really need top desktop card to get some acceptable render times. You need like several Titans if you want to produce clean renders in matter of minutes and not hours.

Interesting thing to know btw about not many students in his class doing it like him....i finished my studies 7 years ago and we were doing all the stuff from the third year onward almost exclusively with computers...

You don't really need high end video cards to get "acceptable" render times, just midranged video cards today can run circles around what a CPU can do ...

An M5950 can still net huge gains in this case so long as his scenes are simple enough to compile ...

 

[Mike64]

Re Luxrender. It's a good program and a great overall project (ie, a fairly long-lived, more or less continuously, though not necessarily consistently, upgraded, open-source application with a fairly large base of users and contributors.) But it is more or less a constant work-in-progress and probably best for someone able/willing to deal with occasional hiccups that don't necessarily have clear solutions.

I'd played around with it a lot in the somewhat distant past, but not much of late, so I'm not familiar with the latest version(s). It used to be OpenCL-based, and ran much better on AMD cards than NVIDIA's, I'm not sure where that stands these days (ie, whether NVIDIA ever got around to improving their drivers' OpenCL performance.)

But in addition to the Wiki, there's also a Luxrender forum where you probably will (or a least might) find more information about the current state of GPU-accelerated rendering with the program. Here's a link to that specific sub-forum: www.luxrender.net/forum/viewforum.php?f=34

 

[kagui]

luxrender have biased render path, and you can tweak the render to take less time also is free so he can do a quick test .

 

[Timmah!]

You don't really need high end video cards to get "acceptable" render times, just midranged video cards today can run circles around what a CPU can do ...

An M5950 can still net huge gains in this case so long as his scenes are simple enough to compile ...

Depends what you consider "acceptable" haha. OFC when it comes to physically unbiased rendering, GPU renderers are faster than CPU ones. What Octane, Arion, Iray etc... can do in hours, used to take days with CPU based Maxwell few years ago - unless you had render farm. But Vray is biased, so its naturally faster and with modern i7 CPU it should be faster than the likes of Cycles on old and weak GPU... which that FirePro is, it has 480 shaders, thats like 1/10 of current fury x card, so its definitely not midrange...

 

[Gryz]

Thanks for all the input.
There is one thing about this 3D rendering that made me think.

When you want to know how a GPU is performing in gaming, there are dozens of websites that do benchmarks. Comparing different GPUs. In different games. With different resolutions. With different amount of eyecandy enabled. Some people even benchmark the impact of CPUs on GPU-performance. Or they benchmark CF and SLI.

But when I look at 3D rendering, I don't find any benchmarks. No numbers at all. I would suspect there would be some people who benchmark GPUs and CPUs, FirePro/Quadros versus Radeon/GeForce. With CPU-rendering versus GPU-rendering. For different render-plugins in different applications.

But to my surprise I have not been able to find *any* benchmarks at all. No numbers anywhere. This makes it very hard to decide whether a hardware upgrade is worthwile. Or what the best render-plugin would be. Or whether to invest extra in a CPU or GPU.

Have I been looking in the wrong places ? Do such benchmarks and websites exist ?

 

[ThatBuzzkiller]

Thanks for all the input.
There is one thing about this 3D rendering that made me think.

When you want to know how a GPU is performing in gaming, there are dozens of websites that do benchmarks. Comparing different GPUs. In different games. With different resolutions. With different amount of eyecandy enabled. Some people even benchmark the impact of CPUs on GPU-performance. Or they benchmark CF and SLI.

But when I look at 3D rendering, I don't find any benchmarks. No numbers at all. I would suspect there would be some people who benchmark GPUs and CPUs, FirePro/Quadros versus Radeon/GeForce. With CPU-rendering versus GPU-rendering. For different render-plugins in different applications.

But to my surprise I have not been able to find *any* benchmarks at all. No numbers anywhere. This makes it very hard to decide whether a hardware upgrade is worthwile. Or what the best render-plugin would be. Or whether to invest extra in a CPU or GPU.

Have I been looking in the wrong places ? Do such benchmarks and websites exist ?

If you want to find any benchmarks on 3D rendering then you'd have to look at specific sites that focus on such software ...

 

[zebrax2]

Anandtech do include LuxMark when benching GPUs which i believe is based on LuxRender

 

[Stormflux]

Is that 8 hours for a single frame or sequence of images? Unfortunately, there is not a lot that could be recommended for GPU Rendering. Most decent GPU Renders are CUDA based. A GTX680/770 is probably a cut off point for using GPU Render that will prove beneficial over a CPU.

8 Hours is terribly long though, even on that CPU. VRay is a good renderer to use and can be heavily optimized. For ArchViz stuff and 300k polys, I'd guess 30 minutes a frame on that CPU for a 1080p image with decent settings would be a good goal to shoot for.

http://docs.chaosgroup.com/display/VRAY3/Rendering+an+Interior+Scene

That's the main help page (link is a more specific relevant page) that he could use as a start.

Main thing though, is look into Adaptive DMC image sampling. And if using Global Illumination, using Irradiance Mapping for Primary bounces and Light Cache for secondary bounces. Defaults are awful.

Edit: Finding benchmarks are usually on the respective render plugins forums and aren't really covered in most Tech Sites aside from LuxMark, which LuxRender is rarely used in production. nVidia does have the monopoly on the scene though.

Which, if you were to go by LuxMark, AMD is King. Sadly, most GPU renders put their OpenCL effort on the back burner and CUDA is the staple. AMD recent efforts to get Blender Cycles up to snuff on their GPUs is a good show.

 

[Timmah!]

Thanks for all the input.
There is one thing about this 3D rendering that made me think.

When you want to know how a GPU is performing in gaming, there are dozens of websites that do benchmarks. Comparing different GPUs. In different games. With different resolutions. With different amount of eyecandy enabled. Some people even benchmark the impact of CPUs on GPU-performance. Or they benchmark CF and SLI.

But when I look at 3D rendering, I don't find any benchmarks. No numbers at all. I would suspect there would be some people who benchmark GPUs and CPUs, FirePro/Quadros versus Radeon/GeForce. With CPU-rendering versus GPU-rendering. For different render-plugins in different applications.

But to my surprise I have not been able to find *any* benchmarks at all. No numbers anywhere. This makes it very hard to decide whether a hardware upgrade is worthwile. Or what the best render-plugin would be. Or whether to invest extra in a CPU or GPU.

Have I been looking in the wrong places ? Do such benchmarks and websites exist ?

Octane render has its own benchmarking tool, you can find some results over
here:
https://render.otoy.com/octanebench/results.php

Its only for Nvidia cards though, since Octane is so far CUDA based only. I would assume those results are directly transferable to other CUDA based rendering apps like Arion or iRay, in other words say the GTX980 will be as powerful relatively to say GTX580 in Octane, as it is in Arion...

And yes, its incredibly annoying, the Anandtech is probably the only hardware related site in existence to actually test graphics cards for their computing capabilties, and even they could do it more thorough (in regard to rendering not just luxmark, which probably heavily favors AMD card)... but still better than everyone else, who just pretend the only use of GPUs is for gaming, so thats all they ever test. I dont know how many years since the CUDA and the whole GPGPU thing was introduced. Crying shame.

I lobbied for this kind of testing several times on local hardware sites (local as in my language, sort of) and i every time i was turned down with explanation, this is not what most people are interested in unlike gaming performance and they are not going to do another test for minority like me...

well dont do it then. stay average and do exactly the same as everyone else. Way to go.

 

[ThatBuzzkiller]

Its only for Nvidia cards though, since Octane is so far CUDA based only.

OctaneRender 3 promises to have an OpenCL back-end so that AMD and Intel GPUs can do some ray tracing ...

I have some faith in AMD since they were able to transform a large monolithic kernel like blender cycles into something that their hardware and compilers can cope with ...

As for Intel, I'm doubtful if their ever going to be interested enough in GPU acceleration for ray tracing to get it running for their GPUs. Not even Luxrender works with Intel GPUs!

 

[Mike64]

There is one thing about this 3D rendering that made me think.

When you want to know how a GPU is performing in gaming, there are dozens of websites that do benchmarks. Comparing different GPUs. In different games.
[...]
But when I look at 3D rendering, I don't find any benchmarks. No numbers at all. I would suspect there would be some people who benchmark GPUs and CPUs, FirePro/Quadros versus Radeon/GeForce. With CPU-rendering versus GPU-rendering. For different render-plugins in different applications.

But to my surprise I have not been able to find *any* benchmarks at all. No numbers anywhere. This makes it very hard to decide whether a hardware upgrade is worthwile. Or what the best render-plugin would be. Or whether to invest extra in a CPU or GPU.

Have I been looking in the wrong places ? Do such benchmarks and websites exist ?

Well, you have to keep in mind that the user base of rendering software is infinitesimal compared to that of games... There are a few benchmarks, but no, I'm not aware of any centralized sources of accumulated data or large-scale comparisons of different cards vs different programs and/or benchmarks. Presumably there just isn't an audience (or an ad-revenue generating audience, anyway) to support such a thing as an independent entity. And since software companies and card manufacturers tend to offer somewhat limited GPU-acceleration solutions, at least at this point relatively early in its development, I suspect they aren't terribly interested in broad surveys of all the available options.:\ And lastly, users of rendering software, for the most part, aren't especially PC-geeky and spending hours of their free time running benchmarks and then posting them to blogs and the like just for the hell of it...

But if you do searches for individual cards/programs using the word "benchmark" as one of the search terms, you will certainly find some info, if not necessarily data about specific cards tested with specific software, from which you can glean/extrapolate useful information even if they don't use the exact hardware/software configuration you're interested in. So to that extent, yes, it would seem you are indeed looking in the wrong places...

 

[kagui]

you can download luxmark (the luxrender benchmark) here:
http://www.luxrender.net/wiki/LuxMark#LuxMark_v3.0

and see users scores here:
http://www.luxmark.info/top_results/Hotel/OpenCL/GPU

but to anything else, the only thing i have found are some older blogs