ART VPS Render Drive/Pure Cards

[Vince Paske]

I was wondering what sort of experiences anyone out there has had with these systems. I use Max6 default scanline with radiosity for interiors and occasionally exteriors. I'm still learning Mental Ray, and I'm liking what I see.

 

Are these ART systems as fast as they say they are in real world conditions, for architectural visualization?

 

Are you happy with the quality? I heard you had to change the mesh of your model to refine the radiosity mesh. This is a pretty steep setback for me because I work in a design environment and make a lot of changes with my linked dwg. Going through and adjusting the mesh everytime would be a real drag.

 

I like the idea that one Render Drive takes the place of several dedicated PC's. It seems the cost of a dozen PC's, including the OS would be greater. (not to mention the service you have to do on your network machines...patching, plugins, etc.).

 

Any other parallel processing systems out there?

 

It would be great if a few images were posted in the "Finished Work" forum with render times, and perhaps a camparison with other renderers.

 

Thanks!

 

ps...I did search through the forums and found the previous threads not very helpful and over a year old, looking for new, up to date thoughts.

[3dp]

modesty forbids me

look at http://www.arcimage.co.uk to see renderdrives in in action on stills

and yes they are as fast as they say

i would not be able to turn out the work i do without them on high res stills

then again i also run multicpu workstation/rack renderfarm too for video work

heres the story

 

http://www.cgarchitect.com/upclose/article1_DP.asp

[Vince Paske]

i would not be able to turn out the work i do without them on high res stills

then again i also run multicpu workstation/rack renderfarm too for video work

heres the story

 

http://www.cgarchitect.com/upclose/article1_DP.asp

 

Thanks David! You do beautiful work.

 

Have you experimented using radiosity in conjunction with the Art systems yet? I take it you render frames for animations on your Art system, and use your multicpu workstation/rack renderfarm for editing video.

 

I agree totally from your interview that visualization should be more integrated with the design process, not just a final product, which is why I'm exploring the Art systems. Faster is definately better and too much is never enough. Thanks again.

[Devin Johnston]

I've also looked into the Render Drive option, but the draw back is you have to learn a new way to apply materials and lighting. I've been using radiosity ever since Viz 4 came out, I've invested to much time into learning how to use it to just give it up for some other proprietary way of lighting. Someone needs to develop a system that will allow distributed radiosity processing directly from the max environment over an already existing network or a comparable render drive system, that would be worth something!

[Vince Paske]

I've also looked into the Render Drive option, but the draw back is you have to learn a new way to apply materials and lighting. I've been using radiosity ever since Viz 4 came out, I've invested to much time into learning how to use it to just give it up for some other proprietary way of lighting. Someone needs to develop a system that will allow distributed radiosity processing directly from the max environment over an already existing network or a comparable render drive system, that would be worth something!

 

I completely agree, I've also invested a lot of time learning Viz/Max radiosity and don't want to abandon it. ART says it's supported, but that means you have to use your workstation to calculated it and then send it to a Renderdrive (or Pure card) to render. That's not a real show stopper, but what is is that it doesn't recognize your radiosity meshing parameters (which is one of the best ways to control radiosity) and that you have to adjust your model to refine the mesh. This doeesn't seam like a very good option working in a design environment using the File Link Manager to bring in dwg's.

[3dp]

hi again

yep i don't use the radiosity implementation generally

these days for externals the lighting rig is 1 sun 1 sky using renderdrive skylight shader and maybe 5 - 6 bounce lights from ground level

this enables me to turn out a average full res a3 image in 2-3 hours of render time but by dragging a lighting rig which i have created i can get the lighting to match a background plate or video in maybe 10 mins

internal are a little trickier most of the stuff on the webpage is done the hardway or is a composite of lightscape/scanline

what i have developed now is a system of max radiosty pass1 and then substitute lights for renderdrive lights with ies then render a second pass which provides the scallops with a nice defination it's a little more labour intensive but it gives much much more control on outout plus i can raytrace all the metals glass and refelction in away scanline never can

as for renderdrive materials lights materials

it;s like brazill fr vray etc whatever you invest in if you go 3rd party you will have to learn new tricks but to art's credit the learning curve is small and it's pretty seamless in max (actually it's pretty boring with the lack of complecated dialogue boxes to navigate)

true there not for everyone and all solutions but for high res architecture stills with a artistic flare that " this is radiosity so it must be right" i found and still find for speed quality and the flexibilty it's the best sledhammer to crack a nut

if there was something better i'd use it

[Craig Ramsay]

I agree with David. Remember that to be a good 3D artist you should be able to light with radiosity and without it, just look at the Smoke 3D tutorials on the site to see how good "fake" lighting can look... there are just some situations where radiosity/GI is going to be out of the question. Anybody can make a model, put in some IES lights and press that radiosity button but your work is going to look the same as everybody else.

 

So really you don't need to lean new lighting techniques to use a renderdrive you just have to be confident enough not to be relying on radiosity all the time which is going to help you if you have a renderdrive/pure card or not.

 

Sorry for the rant, it was a late one last night

[Vince Paske]

Thanks for your perspective guys, it gives me a better base to make a decision.

I haven't found radiosity in Max/Viz to be "perfect" and have had to fake a lot of lights to get the desired output too. Metals especially, I've always had to use Free Direct lights to boost them up, although I'm having better luck with the Architectural metal materials. I usually have a group of IES lights and an omni light set at specular only for typical flo light fixtures to get the highlights looking right.

I've seen people use radiosity where I work, throwing in a bunch of ies lights and expecting the image to be perfect, and it never is. Then I get a phone call with a "why isn't my __ looking right? or "how do I ___?" So, you're absolutely correct when you say it's the artist, there is no magic button labelled "make pretty picture" on anybody's keyboard. It takes time and some bold experimentation to develope a good techinique, and for interior renderings - I haven't found one standard technique that works for everything, it always seems to like a combination of things I've learned before. I guess that's why I love doing this stuff.

Thanks for the lighting technique tip David, I'll give it a whirl if I get a render drive. Hope you get some sleep Craig.

[luckytohaveher]

The issue of rendering is any system is exceedingly complicated. It is a dynamic relationship between speed, accuracy, ease of use, compatibility, and, as always, cost. As a digitally enabled architect I have been working with Max/Viz for years in various venues. I have rendered on stations, farms, Renderdrives, radiosity, and Vray. They all have advantages and disadvantages.

 

Having rendered on Renderdrive for a couple of years in architectural practice the Renderdrives are in my opinion a 'world-changing' item. The Renderdrive 3000, in my work at least, is the equivalent of a 20 station renderfarm. However, because it can work 24/7 it is more like 35 stations when night time rendering is employed. You don't have to relight or change your materials. You don't have to do much of anything. Just plug it in, copy over the plug-in, and hit 'render'. That's it for a baseline. However, after you use it--things change. What would you do with a nuclear weapon? When area lights, raytraced shadows, and complicated materials become 'free' in rendering terms wouldn't you use them? So quality goes WAY up for almost no effort. If you work in an architectural office, it is the only way to go because there is not usually a dedicated renderfarm. When using a PC station renderfarm, they are used all day for Autocad and outlook so you have to do it all at night or weekends (it SUCKS and its not 'free'...)!

 

Radiosity IS an option in Renderdrive with the RP SSI material (RenderPipe Selective Secondary Illumination). It requires minimal effort and produces radiosity, vray, and Brazil type solutions quickly in a fairly predictable manor. It can not save the solution across frames for animation and because it is computationally expensive so it doesn't really work for animation. For still though, there is NO comparison. I have rendered full 'radiosity' type stills in 6-12 minutes (not hours).

 

So what is the downside? Cost, compatibility, and scalability. A pure card costs about the same as two pcs and is about the same speed as eight when raytracing. Since it is dedicated, it is pretty much up to how much you can utilize it. Some plugins don't work -- anything with custom algorithms and without polygons won't work. There doesn't appear to a 'hair' solution that works. Also, once you use custom RP materials, lights, etc. you can't scale the RP hardware. If you have one or two pures you can't throw it on your network renderfarm because Renderdrives are there own deal. However, I recently found http://www.rendermax.com . They have a RENDERDRIVE RENDERFARM which SOLVES the scalability issue for minimal cost (the prices are exceedingly low). They in-effect offer a test-drive for cheap. Simply send them your file and they will render it.

 

Bottom line: I am sold on RenderDrive and will never render on a generic cpu again. It just is not worth the pain and suffering. It isn’t free and isn’t easy, but nothing is. Creating the ‘ART’ we create IS worth it… Good luck on your choice -- "Do you want the red pill or the blue pill."

[Vince Paske]

Hi Ted,

 

I was wondering if you could post a few images to the "finished work" forum. I'd really like to see an example of an interior scene using those RP SSI materials. If you could give me an indication of how long it takes to do a 720x480 interior scene with lots of detail and lights, using radiosity, I'd really appreciate it.

I do a lot of animations, both interior and exterior. I currently use a 6 machine "farm" and flip on about 30 machines at the office at night if I really need to crank something out. I'm assuming that you can render a range of frames, or specified frames like you can in Viz/Max with a render drive?

[luckytohaveher]

All of my images are copyrighted by my current and previous employers so posting would be a no-no...

 

If you have 6 machines during the day and 30 at night, why would you want RenderDrive technology? I would recommend VRAY if you want advanced lighting and have 6 dedicated machines. Slave rendering is FREE and frame-splitting up to 10 machines is included in the professional license. That would be about $800 and be fairly fast. Or you could buy 6 PUREs (16 cores x 6 pures = 96 cores) at $18,000 or so and have about 50 2Ghz P4 machines worth of DEDICATED hardware. I guess the big question is 'What is the Goal?"

 

Ted

[Vince Paske]

Ted,

 

The goal is to save time, cut down on maintenance (updating, patching, etc..), and save space in the server room.

 

http://caselaw.lp.findlaw.com/casecode/uscodes/17/chapters/1/toc.html

 

section 106, 106a and 120

 

I understand what your concerns about copyright are, but I really doubt anyone would come after you for posting images in a forum like this. Take a look in the Work in Progress and Finished Work forum's - I don't think anyone has been sued for posting, mostly because it isn't for self-promotion. It's up to you wether you want to share or not, this a great forum community here and I think a lot of people would benefit from seeing what you did using the Art render farm system.

[Brian Cassil]

I think I'm a little confused? I'm looking into the pure card solution and I have a few questions:

 

Is it not possible to use rendering plugins like final render, brasil, or vray with art vps? From reading the posts is appears that there is some sort of built in software that you must use with the card/drive.

 

If that's true is it a fully functional GI solution. In other words can you render with HDRI, set a certain number of bounces, do caustics, sub-surface scattering, MTD, and good quality AA? Is it upgradeable? In other words, with a software solution there will always be something better coming out in a few months. If this is purely hardware driven, are you stuck with the technology and if something better comes out you have throw the old one away, and shell out thousands of dollars more to "upgrade".

 

I heard someone mention that there are limitations with "special" materials. What exactly are those?

 

Any help would be greatly appreciated since I'm looking to get a high speed rendering solution pretty quick.

[Vince Paske]

Hi Brian,

 

There's some good info at:

 

http://www.4dartists.com/

 

The hardware upgradeability is something I'm wrestling with too. CPU's are just getting faster and faster, makes me wonder how long before a standard PC CPU will out gun the art vps. Also, when will they replace the AR350? What sort of upgrade package do they offer?

[Craig Ramsay]

Maybe you guys could get your answers here: http://www.render-art.com

 

There is a forum dedicated to pure and renderdrive

[luckytohaveher]

Let's make a few clarifications:

 

The hardware solution of Renderdrive ( http://www.art-vps.com ) is a dedicated hardware solution based on the renderman language. The RenderPipe plugin converts the scene from Maya, 3DSMAX, or 3DSViz to renderman format and then massively parallel processes the frame on custom dedicated chips with full raytracing (only!). The resultant image is then sent back to the program just like the standard scanline renderer.

 

Due to the dedicated nature of the chips they are very efficient per area at creating a solution. The Pure has 16 cores equal to about 16 Ghz of generic CPUs calculating a raytracing solution. The RD5000 has 42 or so. The question is how much would a renderfarm of 42 Ghz (+/-), all working together with networking, hardware, software, plugins, Windows XP Pro, Windows patches, Windows patches, and Windows patches cost you in dollars and TIME? Supporting a 20 station renderfarm is not free. Also, due to 'sending' overhead the speed, especially on high-res frames, may not be that efficient.

 

VRAY, Brazil, FinalRender, Mental Ray DO NOT work with RenderDrives. They are inherently incompatible. They use custom programming to replace the scanline renderer using the generic CPU (Intel/AMD) to complete their calculations and thus images. The key difference is that the generic CPU is fundamentally inefficient at these calculations. The custom programming only uses a very small portion of the silicon of the generic CPU. To be FAST you need 10 or 20 dedicated generic CPUs calculating a solution. In EVERY use Silicon -- SIZE is directly proportional to COST...

 

The third rail in all of this is quality. RenderDrive ARx50 chips are based on physics. They offer physics based light, material (optional), and camera (optional) environment. So 20 Ghz (+/-) of Renderdrive isn't exactly equal to 20 Ghz of generic cpu power. ALL generic cpu renderers 'fake' the lighting and camera due to the HIGH cost and calculation overhead of 'real' lighting. The other issues is the adjustable quality settings of the Renderpipe plugin. Quality settings of 10-30 are good for previews, 50-60 for animations, and 70-110 for high-res prints. Every 30 points approximately doubles the rendering time. A preview of 30 may take just a few seconds or a couple of minutes. Try that with a Brazil/Vray/MR/Finalrender... Got an hour? How much is your time worth? Also, because RenderDrive is SO FAST, you are able to instantly get feedback on lighting, materials, and other aspects of your design. How much is that worth?

 

As far as upgrading goes Renderpipe has added 'frame-splitting', RP SSI (faked GI), and HDRI over the last year or so. These are BIG items that are VERY powerful. The RP SSI material takes a great deal of time on interactive surfaces, but doesn't every GI solution. Most importantly, Mike the head of technical support for ART-VPS is unmatched! In my 25 years of computer experience (Back to the CPM {pre DOS} and Apple IIc days) I have yet to meet anyone who can even come close.

 

Like I said before: "You want the red pill or the blue pill?" I am sold on Renderdrive. I've tried everything -- Renderdrives solved MY problems.

 

Ted

 

P.S. 6 Pures (running in your 6 existing machines), with frame-splitting, would be 48 processors and 96 cores... That would be INSANE... With Dual Full-Duplex gigabit connections (4000 Mbits/second), that would be TOTALLY INSANE...

[Craig Ramsay]

Due to the dedicated nature of the chips they are very efficient per area at creating a solution. The Pure has 16 cores equal to about 16 Ghz of generic CPUs calculating a raytracing solution. The RD5000 has 42 or so. The question is how much would a renderfarm of 42 Ghz (+/-), all working together with networking, hardware, software, plugins, Windows XP Pro, Windows patches, Windows patches, and Windows patches cost you in dollars and TIME? Supporting a 20 station renderfarm is not free. Also, due to 'sending' overhead the speed, especially on high-res frames, may not be that efficient.

 

All very good points Ted, I think you really hit the nail on the head with the statement above... if you make the investment then I think the renderdrive will save you money in the long run

 

Craig

[Vince Paske]

The question is how much would a renderfarm of 42 Ghz (+/-), all working together with networking, hardware, software, plugins, Windows XP Pro, Windows patches, Windows patches, and Windows patches cost you in dollars and TIME? Supporting a 20 station renderfarm is not free. Also, due to 'sending' overhead the speed, especially on high-res frames, may not be that efficient.

 

Windows patches, Max/Viz patches and new releases and lets not forget about the plugins. All of that costs more time than whats accounted for and its never considered if a massive job presents itself and you just converted over to a new version of Max.

 

P.S. 6 Pures (running in your 6 existing machines), with frame-splitting, would be 48 processors and 96 cores... That would be INSANE... With Dual Full-Duplex gigabit connections (4000 Mbits/second), that would be TOTALLY INSANE...

 

Hmm, now there's an idea. I deal with insanity on a daily basis, so that may be right up my alley. Thanks Ted!

[Craig Ramsay]

Can you frame split with pure cards? I thought it was just renderdrives that you could do that with. Maybe they have updated the software.

[luckytohaveher]

Can you frame split with pure cards? I thought it was just renderdrives that you could do that with. Maybe they have updated the software.

 

The plugin supports eight devices on a single frame basis or ACROSS frames on an animation basis. That could be eight (8) RD5000s, RD3000s, or PURE, or any combination thereof. It also supports automatic OR user definited weighting for a split. One of my PUREs is on a old station so I weight it 15% less, then the different splits all come back at the same time (when you take 'OBJECT CACHING' this becomes an amplified differential)... Frame splitting is functionally similar to "Region-Net-Rendering" except without the queueing, management, overhead, complexity, and it is essentially seemless. I haven't been able to "Region-Net-Render" across frames successfully. The turn-around in that case is as fast as your single fastest cpu machine. We had a case where a single frame took 23 hours on a cpu machine and only an hour on a PURE. Of course, that file was absolutely over-the-top.