Quote from: Oshyan on June 10, 2019, 04:33:58 PM
I'm also curious why you haven't run the benchmark on your machine yet.
- Oshyan
I have. I posted it on Facebook. Even did a 3D depth pass of it. I didn't post it here because you specifically targeted much higher class machines then your system requirements publicly released when asking for system benchmarks. You said I could still submit mine but didn't sound like it would matter or be looked at much. Another reason I have big issues with "Minimum requirements" and that's it, and than being criticized for what the company has put forward. I'm not sure if you're aware of the Vista lawsuits where they put out system requirements where it did not function as it should, and than didn't offer these people support simply citing their systems as outdated when the box clearly targets it. From a business stand-point, transparency of functionality, and liability, it should probably be looked into. Maybe a real hardware evaluation through a beta test to target minimum and recommended system requirements. Seems the benchmark period for TG4 didn't effect what's available at all.
I'm honestly tired of everything starting with "Your system barely meets minimum requirements" or w/e I'm just don't. I think I'm just too poor to keep up with terragen I guess. Was a mistake. Which is sad since I can go use Maya, Blender, UNREAL, etc and render shockingly amazing stuff in a fraction of the time. Just no real terrain generation. Oh well.
Quote from: Matt on June 10, 2019, 06:39:30 PM
Hey guys,
RTP is not ready for previewing terrain displacements. Speed/quality comparisons for terrain displacements will clearly show a normal render to be faster for equivalent quality. I think we all agree on that.
The reason the RTP takes longer than the normal render is because it spends time anti-aliasing, and no amount of anti-aliasing can make up for a low micropoly detail level. If high micropoly detail is important in your test renders, you will not find the 3D Preview very useful in RTP or non-RTP modes at the moment.
The anti-aliasing in the RTP is non-adaptive, which is another reason why a full render can produce a higher quality result in a shorter time. Adaptive sampling will be added in future, and that will make it much faster.
I didn't see anyone acknowledge that the RTP anti-aliasing is much higher than an AA2 render. It's equivalent to AA6 with max sampling (slightly lower in earlier versions). If you wait for the RTP to finish then of course it will take longer than a test render at AA2, so it doesn't make sense to compare an AA2 render to a fully-resolved RTP render. But the point is: you don't have to wait for the RTP to finish if you are not interested in an AA6 render. And if you decide you want to see a higher quality, you can choose to wait to let it progressively resolve. On complex scenes you probably won't want to do this, but you'll be more likely to do this after we put in adaptive sampling because then it will have a similar speed to a normal render.
So, clearly RTP is slower for high quality renders at the moment.
The reason we have the RTP is this: In many cases it updates in a fraction of a second. Sometimes it takes longer, but it shouldn't take more than a few seconds to give you some useful feedback. Atmosphere, clouds, lighting, object texturing (and terrain texturing, although less detailed) are some areas where you can get some idea of the results of your changes much quicker than you can with normal renders. For low quality renders, the RTP is often faster. It depends on your definition of low quality.
If this level of quality is not enough for your purposes, I understand. RTP will need to improve before it can be used to effectively preview everything. I think it also depends on the complexity of your scene, and the crossover point at which normal renders are more useful than RTP previews might depend on your hardware, e.g. if both the RTP and the normal render are very slow for a particular scene on a particular computer, simply doing a test render is more likely to tell you what you want to know in a shorter time. I am not surprised there are some scenes where the stated minimum requirements are not enough to benefit from the RTP. But I believe that on simple scenes with atmosphere, clouds and lighting changes the RTP may still be useful. Am I wrong about this? If this really is not the case I am open to changing our minimum requirements and/or stating some exceptions for the RTP until we can improve them, but I would like to get more information about the hardware we are talking about and exactly what is happening when you try to use the RTP.
The AA doesn't appear to be near even AA2, realistically, when finished. Do you mean just texture space and clouds? Cause even that looks far sub-par to a standard MPD 0.5 AA2 render. Could you show specific quality examples at the same resolution? I would be curious to see. Cause the RTP, finished, on my system is a jagged mess of square edges... I'm sure using the HD preview before RTP would help there but I can't imagine it being much better. I'll test.
The issue with displacement, is it is PART of texturing, especially when using it to influence texture space, warp it, and evaluate shadowing and lighting within it displacement space. That's where it being advertised for surfaces is confusing to me. Maybe for 0.5 PF displacement detail surfaces with high min scale, it's great, but that doesn't make sense for most scenes when actually judging your shaders for a final render without having to be surprised and go for a round 2, 3 or 4 overnights. That's where I say just rendering out crops or full preview (full) renders if more beneficial.
And than with clouds, I consistently see issues with shadow intensity, and shapes, where shapes don't match, or it seems to just "miss" parts of the coud forms. Maybe this is
another issue between HD and standard preview render before RTP.
Also occasionally, as noted and seen in a image here, sometimes the terrain will explode during a RTP and revert to like 5% normal preview or something like equiv of MPD 0.01 -- obviously without making any changes to the scene, it just happens mid RTP iterations.