Spectral Studio Unbiased Renderer CPU | GPU

Hello guys,

Here is another unbiased renderer to add to the list.
A CPU/GPU || Cuda < works with in just fine /OpenCL renderer
I was going though some raytracing forums and a saw this little engine on the way :eyebrowlift:

I will get right to it, its not free, its around $115 for the program and $10 for extended download services. Pretty good compared to those $800-$900 renderers that dont work half the time.

It is currently in beta with 0.03 just released a few days ago, it is a bit buggy but i have gotten some sick looking renders with it.

The trail has a 45 day limit, with no saving options in scene information or export.
However even with this limitations im already getting my wallet ready :smiley:

Here is the feature list:

Features summary

  • Run on both the GPU and CPU through OpenCL
  • Unbiased & physically correct rendering with accurate light simulation
  • Shading language based on the OSL specifications (Open Shading Language)
  • Physically correct materials supporting diffuse, fresnel IOR, dielectrics, and many more…
  • Standalone application
  • Wavefront OBJ file import
  • Physically correct sun and sky light, HDRI lighting environment
  • Unbiased rendering
  • Multiples rendering modes AO, direct-light, path tracing, metropolis light transport
  • Physical camera with DOF (Depth Of Field), response, exposure.
  • Mesh emitters

Links:

http://www.spectralpixel.com/
http://www.spectralpixel.com/index.php/features
http://www.spectralpixel.com/forum/phpBB3/

Videos

Renders

Looking at the few pictures in the gallery i’m not seeing anything Cycles couldn’t do. Considering the quality and speed of development Cycles is going through I don’t think I will be pulling my wallet for this renderer.

Well, this does appear to have caustics.

As well as cycles

The only cool thing I see is in the last render. Do you think Cycles could do that? I don’t even know what to call it. Are the things coming down caustics as well?

In seven hours? Yes, I think Cycles can do that. A pathtracer like Cycles has a hard time with caustics in general, but seven hours should be enough. Depends on the power of the CPU or GPU, of course.

The downwards caustics look like just caustics on the back wall.

Oh that makes sense, thanks. Why does cycles have a hard time with them? Does time really make a difference on how prominent they are?

It’s just the way the pathtracing render algorithm works. It’s basically like this: shoot rays from the camera into the scene for every pixel in your image. Let a ray bounce around until it hits a light source (in this case it contributes to the pixel color), or until max bounce is reached (this ray is discarded, does not contribute to the pixel). So you can imagine that a pathtracer loves big light sources, so that lots of rays find lights.

Caustics are the better and sharper the smaller the light source is. But the smaller the light, the harder it is for the rays to find it, and contribute to the pixel in the image that is part of a caustics area. So when a lot of rays don’t contribute to the caustics, you need a whole lot more rays shot into the scene to get a clear image of them. More rays means more samples, and that means longer render time.

So if you want fast images with Cycles, use big lights or better yet, the whole background as light source, so as few as possible rays get lost. And avoid caustics - or have a lot of patience. :slight_smile:

It has full OpenCL support and a MLT render kernel.

Then again the question is why I would pay 115USD for a buggy Version 0.03b - and better yet, 9 Euro for an extended download service that stores this 0.03b for the next 2 years for me?
So if my HDD burns down and I payed for my license but not for extended download it’s sorry pal, no new binaries for you?

Soon I have my 2 year anniversary for beta testing Octane, and by now its quite mature. So there you get a pretty much production stable GPU renderer for less money - but CUDA only.
I also wonder what those other renderers for 800-900USD that don’t work half the time are.
For RandomControl’s stuff the pricerange would match, but I’d hardly count Arion and FryRender as “not working half the time”
Bunkspeed’s Shot is just 500 Bucks and surely neither in that category, and Bunkspeed Pro is 3500 USD and quite a beast.

I’d love to hear some selling points for Spectral Studio.

@radiant if you are part of the team/company, i strongly recommend you to get in contact with a good artist and make some serious renders to promote your product. Even when we know that your unbiased render can make as good renders as many others, most artist only will look at your gallery to determine if your render engine is accurate, inaccurate, amazing or crap.
cheers

Nah, im not part of the company, just wanted to spread the word on it, they cant kick all the bugs out without some beta testers.
The renders are a bit basic as i only found this out a few days ago, will be spending more time on it modeling

This post isnt really a “This is better then cycles” its more of a, “here is another work around for your rendering”.
But with that being said, i do find the caustic coverage more effienct and better then cycles even after a few minutes and it has path+MLT which is a bonus.

I’m pretty sure cycles could do that in less than 7 hours, at least on my machine. :stuck_out_tongue:

With the caustic test, i just left that on over night to see what it might look like, i could have stopped it at 20-30 minutes with it still looking decent. :slight_smile:

I don’t know why they are bothering to be honest with you - charging for it especially.

7 hours? Really? Yafaray will do that in a few minutes. It’s not physically accurate? Guess what, your viewer/client won’t know the difference!

If you really, really, just have to have a physically accurate renderer, then Lux is free, mature and stable and Octane will do the job nicely for about the same amount of money - again, its mature and stable.

When is someone going to produce a good, fast, biased raytracer that can be used for animations as well as arch-vis (apart from V-Ray and Yafaray that is)?

The renders are horrid, I especially like the sportscar on the frontpage, with the unaltered ground texture from cgtextures.com/brick/brock rounded/ :smiley:

They should start with finding a better business model to attract testers than trying to get betatesters to pay them for a single digit version.

Lots of new pathtracers are born these days, really strange. :smiley:

I’ll stick to Cycles, rather then spreading energy and look into other (alpha/beta) renders. In the middle term it’s better to have 1 great and fully integrated solution, rather than many.

My 2 cents.

When is someone going to produce a good, fast, biased raytracer that can be used for animations as well as arch-vis (apart from V-Ray and Yafaray that is)

It’s not like you can just “introduce some bias” and magically get better performance. One thing the industry understands, is that artist time is worth more than computer time. All the hacks that V-Ray/Yafaray uses to compute faster global illumination are more complicated to develop and to use, take much more time to tweak to actually get better performance and have their own artifacts to deal with (especially for animation). And worst of all: They often require preprocessing which makes them non-interactive to use.
There’s are good reasons why everyone (including the V-Ray guys) is doing path-tracing:

  • The workflow is faster, simpler and interactive.
  • The algorithm is simple and fully data parallel (which can’t be said of some other approaches), therefore it scales well
  • It runs on GPUs, which are getting better and better suited for the problem. (The performance increase with every new generation is past Moore’s Law)

Apart from that, better sampling strategies and smart post-filtering are hot research topics and I expect to see some of that end up in Cycles sooner or later. (Someone want to implement ERPT? :D)

You need to understand that this program is in its infancy, it would be really irrational to compare it with program that have been in years of development. cycles was a buggy mess back in last Easter, now it is tolerable, why?.. due to poeple testing it out :rolleyes:

I agree having the price stated at such an early age of development, but with the pain in coding such a renderer, with the math, coding and debugging, it can be understandable.

I really wonder who else would invest into this soft. Sorry to say but watching their Forum/Gallery/about us sections -looks not very promising to me.
Lux, yaf, mitsuba, studioGPU, indigo, octane… -all free/cheap solutions, but why pay for this?

Edit: StudioGPU seems to be gone.
http://forums.newtek.com/showthread.php?p=1186899#post1186899

Someone want to implement ERPT? :D)

It almost same as “classic MLT” with same pro’s and con’s, main problem is to find good implementation for GPU with all that GPU trickery. I have almost all parts that can do result close to Bi-dir except wrong asymmetric BSDF (like ior depend glass shader) and initial sampling code, but stuck when figure my variant do not fit current GPUs and same time last Brecht render layer addition require to rewrite it. Anyway, scenes where MLT-like renderer “shine” tend to be complex and in 10+hours converging time, but we need good solution to render animated scenes in reason time. I think full motion blur, better memory and hardware support is more priority.