GPU Box

[glorybound]

I have been looking for a solution to my workflow bottleneck. I am getting great rendertimes on my main machine, however, a render can still tie it up things for a couple hours. I did some searching for GPU boxes and they exist, however, do they make sense? Or, does one of the BOXX render boxes make more semse? I was reading about Corona and they write that dollar for dollar, CPU's and GPU's are the same and they don't plan on going down the GPU path. I purchased two GTX 1080's and so far, they are doing me little good. If I purchased two Titan X's, it'll cost 5-6 grand and I am afraid it'll be a waste of money. I can get on of the BOXX render boxes for that price.

[Dave_Wortley]

You can build your own VCA with 1080s, best thing to do right now, not sure how many graphics cards VRay will actually use, some applications will only use 2 or 4 maximum.

[Nicinus]

Yes, this whole CPU vs GPU is a big dilemma. On the one hand I think there is just no way CPU will be able to compete in the long run, but on the other there are just no viable GPU solutions available at the moment. You are either restricted memory wise, or they miss plug in support, or features. The most logical one is obviously Vray RT if you are a Vray user, but they are to me really two different products.

I can't understand why there can't be cross over benefits, surely Vray Advanced could offload something to GPU?

[Macker]

Memory is the bottleneck here. Our scenes can sometimes max out 32GB ram.

[thanulee]

I really dont know the answer after 1 year that i am involved with gpu rendering. I think it really depends on what ur needs are. And i mean by that, "how much VRAM" ur usual scenes consume. And what type of scenes u are making? Mograph stuff are better with gpus. Interiors are a problem. Huge scenes are a problem too. Im in that dilemma as well now that vray supports ipr. Maybe 2 xeons are better than 2x1080?
I have invested in gpu though, i also use octane and testing redshift. And i use a lot of RT and i find it very very nice right now. Maybe u should just consider what ur needs are. And if ur needs is "faster rendering", either gpu or cpu u will need a lot of $$$

[glorybound]

I really dont know the answer after 1 year that i am involved with gpu rendering. I think it really depends on what ur needs are. And i mean by that, "how much VRAM" ur usual scenes consume. And what type of scenes u are making? Mograph stuff are better with gpus. Interiors are a problem. Huge scenes are a problem too. Im in that dilemma as well now that vray supports ipr. Maybe 2 xeons are better than 2x1080?
I have invested in gpu though, i also use octane and testing redshift. And i use a lot of RT and i find it very very nice right now. Maybe u should just consider what ur needs are. And if ur needs is "faster rendering", either gpu or cpu u will need a lot of $$$

Well, sad to say, but I have to use Corona for my interiors. I have been getting many jobs with a lot of stills, so I can't take 2-3 hours for each of them. Like tonight, I have changes for 6 interiors, due tomorrow morning. In Corona, it is taking 20 minutes per still and 2+ hours per view in V-Ray. I haven't had much success with RT, other than basic interior lighting and rotating my exterior HDRI's. Exteriors? V-Ray is still the bomb, however, if something else show up that can do it faster, I'll have to contemplate using it. I hope V-Ray isn't going to by just for you special effects people.

[Alex_M]

Bobby, I think there's something wrong with your render settings or the way you setup your scenes to notice such a big difference in render times. I find it hard to believe that something that takes 20 min. in Corona takes you 2 hrs in Vray. Generally I've found that Vray and Corona are pretty much the same speed, although I've not used Corona extensively, just on a few interior jobs. 6 times slower render times means there's something you need to check with your scenes. I've found that 95% of the time the default render settings in Vray deliver great results in terms of speed and quality.

[glorybound]

Possible however the only reason I tried Corona was the talk about how fast it is compared to V-Ray. My first try at Corona was when I couldn't get V-Ray to render a scene in less than 3 hours. I downloaded Corona, converted the scene and hit render. Literally, in a few minutes without touching a setting, I had a clean render. I tried RT and it never really cleaned up any cleaner than a few seconds in Corona. I might be doing something wrong, however, it is like Steve Job saying that we are holding it wrong (antenna gate).

[thanulee]

Hmm yes, seems a bit odd this discrepancy between an unbiased and a biased solution. Vray by definition, unless ur using bf + bf should be faster.

[dean_dmoo]

I've gone down the route of buying refurb dual xeon 2670 workstations with 2 years warranty. Have 3 of them now. About half to a 3rd of the price for a new one I think. Having these has freed up my workstation (which admittedly is about half the speed of the nodes) but it had made it a lot more efficient to work as renders kan just be placed in a que (DR+BB) and I can keep doing test renders and working on other aspects in the mean time.

I keep looking at GPU's but it just doesn't seem viable at the moment.

[tricky]

Another idea: for the clients who want changes on 6 images overnight, just say "Err...no! That'll take longer than that and it will cost you to make those changes".

[tricky]

...or another idea, can you Photoshop the changes? We do a lot of painting work for these very reasons: it isn't always sensible to re-render everything when a patch render or bit of painting would do it.

[PIXELBOX_SRO]

it would be cool to know of the new AMD VEGA architecture will work with RT
That little fucker has 1TB of RAM!!! More than youll ever need. I am afraid RT is meant to work mostly with CUDA....can someone from Chaos prove me wrong here?

i am easily maxing out 64GB so my next workstation will have at least 128 GB for sure

[Alex_M]

Yes, I'm eager to know how does OpenCL perform in RT compared to CUDA and if it supports all features of RT CUDA. If it does and the performance is comparable to CUDA, my next GPU will be AMD Vega.

[glorybound]

Apparently they have a meeting with a big time client.

Another idea: for the clients who want changes on 6 images overnight, just say "Err...no! That'll take longer than that and it will cost you to make those changes".