aaronrumple

Check Fresnel Reflections. (I'd also check reflect on back side and up the number of reflections)

Too much noise for the denoiser to make a respectable interpolation. Look at your noise channel to see what the noise levels are in that area. The denoiser is good at cleanup, but needs a certain level of info to do its work.

I'd probably go with the AMD Ryzen these days over my i7 5960x. But this is an interesting article: http://www.techspot.com/review/1155-affordable-dual-xeon-pc/

30 sec walkthrough at 30 fps? That's 900 frames. Your time can be calculated based on your single frame test renders. Benchmark your current system with Cinebench. Then find a similar system to what you want with the i7 7700 and the math is pretty simple on what your speed increase might be. It sounds like your current system would be part of the rendering farm.

I think GPU farms are still somewhat esoteric and limited for general production rendering. Great for RT previews. I think they can be great for specific workflows as well - but that would be a specific type/style of scene you do over and over with predictable input/output. CPU is solid and flexible if you're doing lots of different types of work - and as you mentioned - the least restrictive in workflow and plugins. If I were to expand my farm - I'd be looking at used server chips. These pop up when datacenters switch things out and then flood the used market with perfectly good silicon. It might not be cutting edge, but the speed to price point is great. My next farm would also be Linux based. My Linux servers are rock solid and simple to maintain. The cost savings on just windows and deploying backburner will be worth it.

What's your camera settings? Vray camera with exposure set at 1/100th second. ISO 100 and F8 to F11 have always worked well for me as a daylight model using Vray sun at the default exposure of 1. Typically, I'm using a reasonably low sun angle for either early morning or late afternoon. I always test with a material override of RGB128-128-128 on the building and a something a little darker for the ground plane (RGB64-64-64) I find that compensates well for parking lots and other dark planting material.

I just built a new rig around the i7 Extreme 8-core chip and the Asus x99 Pro Board in a Corsair case, power supply and cooler. I'm extremely happy with it. The only overclocking I've done is to hit the idiot proof tuning in Bios and let it do its thing. It clocks in at 1556 using Cinebench 15. I haven't even flipped the jumper one the motherboard that will take it even higher. It's running right how at ~4ghz using the stock voltage. I'll give it a go at 4.5 ghz, but right now just getting all the other software setup. I have a GTX 760 as the gpu right now, but will swap it out down the road for probably a 980ti or two. (I'm doing mostly CPU rendering with Vray LM/LC) The 760 is more than capable for using as a workstation. Watch out for the 2011 sockets. They are very delicate. The first two motherboards I was shipped from Amazon had the protective cap off and rolling around inside the packaging. The sockets were mangled. I suspect these were re-boxed boards that were resold. I ordered a 3rd from Newegg and it arrived in fine condition. Memory is the G Skill Ripsaw 2400 at 32 gig.

That's what RT is designed to do - use the full capacity of the GPU. The solution is a second video card. One to run the software display and another for RT. Selecting GPU's for RT is set in the options.

Using the Vray frame buffer and linear workflow in you non-RT setup?

You could be. You won't get an 'error' message. Rather you'll see Vray 'unloading geometry'. When this starts happening, things get really slow. However the renderings should finish given enough time.

Frozen layer?

1 CPU Xeon E5-2687 Cinebench 15 score of: ~1227 $1900 2 CPU Xeon E5-2650 Cinebench 15 score of: ~2043 $2250 1 CPU i7-5890 Cinebench 15 Score of: 1393 @ 3.0 ghz and 1747 @ 4.45 ghz $1000 The dual is clearly a better value in the two systems you describe, but Xeons are low on the price to performance scale. I've got an office full of i7-4770's at work. Never seen them crash. I have an i7-5890 at home. This chip is designed to run at 'overclocked' speeds. The older 4770 systems only benchmark at 731, but they are cheap and I can offload rendering to the other cheap (3x731 = 2193 for $900 in chips.) ) systems using backburnner and still keep working. Staying productive while still rendering is important. Personally, I'd build a couple of i7 (probably the i7-4930) systems for the price the dual Xeon and probably come out way ahead in both cost and speed. And have a better workstation experience overall.

Lots of variables there. Probably the 20 @ 2.4. Maybe. Are they both Xeon's? Same on-chip cache and all? Or is the single chip and overclock-able i7? Extra memory on the dual chip system? Memory bus speed the same? What's the budget?

Yes, yes, burned in gamma ad sRGB off were my standard settings in 2.0 with exr out @ 1 linear workflow and all. I guess I never noticed the big difference between VRay's and PS's sRGB curve until trying the 3.0 defaults of not burning in the gamma. You'll see in the first screenshot, the sRGB preview in the frame buffer is considerably different than the same file in opened in PS. With the gamma burned in and sRGB off, the preview closely matches PS. Since I don't like to do too much post production - it is more important that the frame buffer very closely matches the end result in PS. Without burning in the gamma - it is too far off.

Here's a side-by-side with "Color Mapping and Gamma" burned in. A much closer match from frame buffer to exr in Photoshop.