Vray - multi-threading

#20

Bit of a misleading statement.
Rendering Engines with inefficient threading aren’t the only reason, or even the chief reason, why that’s done.
Peaking memory and keeping as many nodes dark as possible is a much bigger factor in making that choice.

It’s also a bit of old news, mostly coming from when everybody and their dog used RenderMan, which has always had, and seems to keep having, the poorest and most wasteful multithreading possible.

Studios that moved to more modern and better scaling models and engines don’t do it nearly as frequently.

The scaling outlined in the original post is actually quite poor, so it’s either a particularly tricky scene, due to its nature or possibly even poor settings choice, or there’s a bottlneck somewhere (which might be anything from thrashing the memory to storage bound issues), or VRay has work to do (and apparently they have done a lot of work to scale better in recent times).

Lets not keep propagating relatively dated knowledge as verb, the most modern models and engines don’t have “a lot” of single threaded heading any longer, or not very often at least in the context of rendering most frames of a movie :wink:

#21

ya, I’d say that based on my render tests, that’s a pretty accurate statement. I won’t be adding this to V-Ray Tuner since, from my experience, I don’t see any scenarios where you spend any major amount of CPU time in poorly-threaded tasks.

#22

Yeah, I read a pixar article about making Monster’s University and they were saying their main rendering goal was to keep each frame under 20 gigs memory so they could pack 4 parallel renders per machine with their 96 gig ram render nodes. I guess I assumed if pixar is doing that with all their new raytracing GI stuff they have going on, I was figuring most other major studios using renderman probably did the same.

Then later I read an article and watched a video about Arnold renderer where they made a huge deal how efficient their engine is and how other engines end up making people render multiple jobs in parallel to try to recoup that lost efficiency from single-threading.

These articles weren’t very old, so I was assuming they were somewhat accurate

#23

It’s not dated to the point of being non-sense, and anything you read about rendering engines that are also commercial products, or nominated for sci or tech awards, is to be taken with a shovelful of salt, but rendering in general is possibly the discipline I’ve seen advancing the most and in the most revolutionary fashion in the last five years, and keeps doing so.

Also nobody really comes out with a lot about the management and IT side of things, because it’s highly local and not really good marketing, but when you reach the thousands of nodes scale it has a considerable impact on a facility’s choice, but that choice might have to be presented otherwise to the public buying or awarding prizes to their products.

Raytracing is one of the few problems in CG that can scale to truly impressive widths, but for some engines that scalability requires a pretty radical model change (see PRMan 19 in example, and we’ll see if that will have been enough, the fact Disney is going proprietary for DA instead of putting the chips on 19 isn’t terribly auspicious, but might just as well mean nothing, or even be a political choice).

I’d say of all disciplines and of all times this is one to keep a constantly watchful eye on rendering and assume knowledge can obsolesce fast. Between major paradigm shifts in representation, in programming, and in hardware it’s pretty varied and easy to lose track of. Exciting times and all, I guess :slight_smile:

#24

Well I still run into this issue with MR - especially with heavy geometry that has rendertime active polysmooth nodes in order to keep scene file sizes and network traffic manageable. I generally have seen a 15-20% render performance boost rendering files in parallel.

It’s especially a problem in MR with the “fast” SSS shaders that use lightmaps which become more of a problem as output resolutions increase to 4k or higher. I know there’s new SSS shaders for MR now or you can override the light map resolution for the legacy shaders, but I still have to wonder how often people will encounter single-threaded bottlenecks, especially dealing with older files for long-running projects.

#25

Mental Ray is anything but a modern engine. iRay might have stood, or still stand, a chance, I don’t know, but the old engine is like the old PRMan, I don’t think it can be redeemed by anything short of a complete reboot that keeps only the name around.
In that case, and with some shaders that rely on accumulating a crapton of samples before actually doing anything with the shading itself, yes, you could still be bottlenecked down the line.

You got to be one of three or four people I’ve heard in the last year even bothering to mention MRay though :slight_smile:

#26

Yeah, MR doesn’t have the most bleeding-edge features at the moment, but it’s still capable of good work. We’ve all heard that about many software over the years and it’s always a temporary thing for the major players.

In some small studios/industries we have to ride things out and can’t justify switching software every time there’s a new kid on the block. It would be nice if the time/manpower/money was available to do so, but there’s too much work to do and often enough artistic freedom in shading to not have to adhere to the rendering requirements of matching the lighting from footage, even though I still lust after many of the rendering features other engines have right now.

Given MR’s integration with most of the major software, the backing from Nvidia, and the progress they’ve been making with MILA and the recent better maya integration, IMO they’ll pull through in the end. Every renderer engine’s goals pretty much are the same anyway. They’ll all get there.

#27

ya, but follow the money. A GPU company isn’t necessarily interested in the future prospects of a software renderer.

#28

I’m not sure I would call vRay, Arnold and PRman new kids on the block either :slight_smile:
Chaos Group is a triple digit employees company and has been for a while, and at this point it’s probably more used and more backed than MRay.

MRay is still around uniquely and solely because it comes bundled. Had it been an option it would have sold nothing, and could you save even just 200$ on the cost of a seat by giving it up probably a good half or more of AD’s clients would instantly save the 200$ and renounce it to spend them on something that actually works. It’s not like that poll didn’t crop up in a couple places to come up with a 90% “screw MRay” result :slight_smile:

As it is it’s already disappeared from entire creative fields.

An engine so slow and unpredictable, even “for free” is simply a fake saving. When another engine renders the same stuff 10 to 50 times faster and gets the results iterated to quality into a fraction of the time you’re saving a penny to spend a pound by sticking with the old.

Honestly, get in touch with Chaos Group or Solid Angle and get a trial for one project. Lets see how much more stuff you can render and how much quicker when you push it through a decent product. I’m not having a go, Sentry, I am most sincerely recommending you do.

#29

yeah I fully agree, but then again every renderer is heading towards massive parallel computing, and ultimately on the GPU. CPU’s aren’t going to own rendering forever.

In the meantime at least MR offers rendering compatibility with all our legacy work, vs ditching a lot of work it en mass to switch renderers. I certainly can’t defend MR as being the most modern or best integrated right now, but for our needs, other renderers are going to have to offer something that’s 20x better, not 2-3x better sway me over.

It’s similar to asking you to ditch maya and switch to houdini. It’s easier said than done when you have 15 years of maya files that you continuously work work on for clients.

#30 
  Thanks. I plan on evaluating vray closely this year. I want to fully explore MR's MILA first to see what I think of it. We don't have a lot of staff so most of my time is spent actually working on productions. The budget isn't an issue, but the R&D time in recreating a bunch of shaders that we rely on and then learning the nuances of a new renderer is.
 
 I mean I can put it this way, to some extent it's almost arguably more worthwhile for us to drop $30k on additional rendering computers so we can focus on production than it is drop $10k on vray and then invest a lot of R&D time into switching to it in order to get the gains from it. Then given how much is currently going on with MR, I have to wonder if once we switched renderers, MR will have narrowed the gap anyway.

Then again as our (small) render farm expands, MR's license costs make less and less sense and vray's cost looks better and better. 

And right now vray is the renderer I'm leaning towards the most because I feel like it has the most industry support, tutorials, etc and the price is right.

What I'd like to find out at some point is if vray can render single 20+ million polygon objects. MR last I tried with maya 2013, it can't and starts omitting various triangles as if they were deleted. I know that sounds like a silly thing, but that limitation has created headaches on a few occasions with things like large connected artery trees that have lots of specific curves and branches.
#31

Other renderers DO offer something 20x better very often, that’s what I’m saying :stuck_out_tongue:
MRay just doesn’t render the same amount of stuff in the same amount of time (let alone anywhere as reliably) as Arnold for almost any and every scenario.

Why do you think MRay has always been patchy in film, adoption wise, while Arnold swept in and picked up every other major house before it was even publicly available? :slight_smile:

Try something else man, you have no idea how much of a difference there is. You are talking double digit multipliers, yes.

#32

yeah I should probably make evaluating new renderers a priority. It’s just such a PITA to migrate over, but that’s probably what it’ll take to move forward in both output and quality.

#33

I second every word ThE_JaCo has said about the subject. Not that Jaco needs me to backup his words at all, but I felt I had to comment on this as I’m right now stuck on a mray show. Even worst, I had to go back to mray after working with arnold for almost 2 years.

I used to work on mray only, and I was strugling against changing, and then I had to. It was such a breeze of fresh air. The worst part on mray, other than it being slow, unpredictable and unintuitive, is how every single thing in it is a hack. And the fact than some mray nodes aren’t compatible with other mray new nodes drives me completely crazy. And even the most simple tasks are somewhat intrincate and silly, you end up connecting 4 nodes to make a map just work as a gobo.

And I didn’t even mention how much more capable are arnold and vray churning out nice results with hair, displacement, and everything in it without having to re write shaders, or connecting very complex networks of hacks, or it won’t even render at all.

So yeah, making the change is a pain, but once you’re settled, you’ll never look back. And maybe some of the lost hair will grow back.

#34

Yeah, and I have to admit some of the resistance from me is just stubbornness of dealing with MR’s intricacies all these years and being bitter about learning all that stuff just to ditch it out of frustration.

Things like FGshooter being a solution for final gather is where I started saying “are you’re kidding me?” - great that there’s a solution, but yet it’s another band-aid for something that seems like it should work right in the first place.

#35

I understand completely, I’ve been in your shoes for longer that I care to admit. I know ditching years of tricks and knowledge makes you feel bitter now, but imagine that all that brain activity can be emptied to now use it for more creative and pleasent activities. So, you’ll be changing bitter knowledge for relief and more creative output. Nothing to lose here. :scream:

Take it from someone that is now back and stuck with mray, I can’t wait to ditch it out again.
Cheers.

#36

Admitting you have a MRay addiction with all the problems it brings with it (hair loss, bad breath, gas, mood swings) is the first step. You’re on your way, man, you’re on your way.

#37

Just wait until you see what support looks like for a software renderer that wants your business. You hear a lot of stories about V-Ray adding features within a week of asking for them and that’s been my experience. I have heard someone say “that’s not sustainable with growth” but it beats the hell out of “likely never”

#38

Ah but that’s the thing. Vray is not NEARLY as hard to learn as MR. You’re probably thinking “I worked so hard to get where I am with one renderer, now I have to learn a another?”
Put that out of your mind! I was proficient with vray after the very first commercial I used it for. The second time I used it was on the Halo 3 release trailer, and it hauled ass - I was stunned - and looked awesome doing it. But after using Mray for 9 months, I still didn’t know how to use most of the shaders, and experimenting with them, trying to get them to work, left me bald. Vray is vastly more intuitive. Depending on what you’re doing, you should be up & running in no time.

I’m likely to catch some flak here; Arnold is, overall, not as fast as Vray, when you look at the big picture. Almost everyone, including experienced professionals, find it very slow from the start, and are told ‘you have to learn how to use it’. While partially true, vray rips through renders right from the start. I’ve heard about a few things that Arnold does extremely well, like volumetrics (vray is still a newborn in that respect) and massive assloads of geometry, but the lack of any Irradiance cache, light cache or baking, leaves you without some seriously powerful time-saving tools. Arnold is brute-force-only, all the time (afaik - I’m still checking it out, when I have time).

Even their own docs claim that it’s designed to save artist’s time, not render time. That’s fine for big studios, but not something I’d use at home or a small studio.

I wouldn’t give “big studios adopted it before it was commercially available” too much weight. Sony bought it primarily to save money on Renderman licenses, and secondarily because it could raytrace, which Renderman was crap at. Sony wanted as much as possible to be in-house software - their compositor / lighting tool (katana), image viewer, even their IM messenger was proprietary, and they were working on thier own cloth. Apparently there’s a nice tax incentive for software development.

I’ve been through Sony’s arnold training twice, which was focused on minimizing noise and optimizing speed; and I can’t get amazingly fast mind-blowing renders out of it, whereas with vray, with as much or less training, it was killing it.

Back to the OP - I’m still getting benefits from splitting the renders up, in vray 2.4. Even with very basic scenes; an F1 car with just diffuse, and glossy reflection, an hdr, brute force gi, no deformation and no motion blur. I do have one scene that showed no benefit, but 2 out of 3 makes it worth trying, to me. I’d love to compare speeds with vray 3.0 some time - if splitting renders becomes moot, then, all the better.

#39

I’ve been checking vray out, even though the demo’s low resolution limit makes it almost pointless to really test anything out of any worth since you can’t see how well it actually scales at production resolutions.

Our models are heavy and even vray will still sit there using 1 core for a bit before starting to actually render even when just using a single vray light and stock vray shader. I’m obviously not an expert at vray optimizations though.

I like that vray’s render pass system seems more straightforward and likely more capable than MR’sin terms of ease and speed. Things like RGB passes showing the RGB mattes in reflections and refractions for no real extra rendering cost along with your beauty pass is pretty nice.

However, I’m still digging into MR more too. I do really like it’s new layered shader system, though it seems slightly prone to noise. My impression so far is it strikes me as more powerful than vray’s current shaders, though vray’s materials seem more simple to work with with already having the majority of settings already available.

I’ve confirmed that MR now can render single 20+ million poly objects correctly with maya 2015. This was a problem with maya 2013 and earlier and was one of my primary reasons for even thinking about looking into trying other renderers as our models get heavier and output resolutions getting higher (preparing for 8k).

I’ve also been researching arnold too, but I don’t think it’s the right renderer for us. I really like what it can do, but it’s not a good fit for us for a number of reasons.

Anyway, that’s what I’ve learned so far even though my evaluations are still a work in progress.