Computer buying help thread for October

[F J]

ehh, all of this has already been goin' on for quite a while now by chaosgroup n randomcontrol (at least these r the ones im aware of), also with the GPU Cloud capability..

[AJLynn]

Aha! I've found where they're keeping the good GPU architecture renders: http://www.randomcontrol.com/arion-gallery-architecture (Of course - it's the one that's got Fran's stuff.)

 

Ry, Folding@home is easy for GPUs because you can parallelize it across many threads, the math is simple and you incur no penalty from the treads having no influence on each other. In this, it's like brute-force-only (e.g. monte carlo) rendering - the thing is, when you can do more complex computation, e.g. adaptive sampling and other optimizing algorithms that can run on the current mainstream renderers but that software like iray and Vray RT-GPU can't do (it's a fundamental limitation of this type of economy-of-scale computation that anything you do that deviates from a pure monte carlo approach actually decreases its efficiency) you need less work to do the same job and my pretty pathetic CPU can do the same work as one of the best nVidia GPUs. (And I'm out of practice and wasn't really trying.)

 

This is why only comparisons of real, useful images with render times are proof of concept.

 

The car thing is potentially interesting. I tried it out at the Fermi launch event back in March. It's hard to tell how much of the technology used is applicable to arch vis - I had the impression that what it's doing is putting cars on geometry stages in prerendered HDR environments used for reflections, but I think it's promising if it's an indication of what they're aspiring to for the next generation of real time.

[revuprender]

Hi Travis,

 

I should point out that RealityServer uses iRay for its GPU rendering engine. RealityServer also has CPU rendering mode and a generic GPU rendering mode as well. RealityServer itself is not the rendering engine, think of it more of a platform. It does web services for cloud computing among many other functions. Besides being a web services platform RealityServer also has some advanced clustering tools so it can cluster together GPU's across many servers and it does this through the ethernet connection.

 

I do not think they have plans to sell RealityServer to the average end-user since it is a big platform that typically large hosting companies will purchase and run on their servers as a service. I hope that helps you understand this technology a little more.

[revuprender]

Ratnakar,

 

Let me explain a bit about my background and why I got involved in all of this. My background is in architectural design and I was working on projects with very complex geometry inside Revit and when we would hit the render button it would freeze and never get past 10% completion. Around that time I started to realize how big of a gap there was in the AEC industry and other industries when it came to technology. My goal is to remove the creative disconnect between what a designer actually thinks of in their mind and what ends up designed on the computer screen or even built. Currently there is a huge disconnect there and we want to bridge that gap.

 

Over two years ago I started looking at ways to speed up the rendering directly inside of Revit because most architects I worked with did not use 3DS Max and I just felt like the Revit rendering capabilities were not feasible at all. I started to look at what the gaming industry was doing and then I think the only GPU rendering engine even available was Gelato http://en.wikipedia.org/wiki/Gelato_(software) which some of you may remember was a project that Nvidia made. I think Nvidia took the entire team that made Galato and moved them over to make what became CUDA. Basically I felt at the time that if we could make a GPU rendering engine for Revit then it would help architects to be able to render walkthrough animations.

 

Look at this animation here I made in Revit about six months ago:

this took about two weeks to render these two simple animations

I was using a system with (2) Intel Xeon X5550 Nehalem CPU's and 24GB of RAM. I am showing this example to help show how unpractical it is to render animations inside Revit. The frames for this animation was not good quality but it still took this long for 1200 frames. Most architects just give up trying to render walkthroughs in Revit because it takes too long and this is supposed to be an important part of the design process to walk the client through the unbuilt project.

 

Before iRay was launched with the GPU version and RealityServer 3.0 I was working with some programmers to see if we could make our own GPU rendering engine. However then I saw iRay coming out and figured it would take us years to come up with a GPU renderer that could do the same. It is really crazy now that when I started down this path that there was really just that one GPU rendering engine Gelato that I could find and now they are popping up left and right from everywhere.

 

Well so I started talking to Mental Images because I knew we could take advantage of their technology faster then we could make our own from scratch. Now it has been over a year of development and we will finally be launching a product at Autodesk University which will enable architects to take the entire Revit model out into RealityServer and render it in seconds/minutes whereas it would take hours/days inside Revit. While RealityServer was built to be a production display tool we are quickly turning it into a very powerful collaborative design and design review tool.

[revuprender]

"i m not making a system of six xeons.. it was of four opteron each one at a cost of $300... less than 1/10 of a decent quadro.... GPU rendering does look promising"

 

I think you might have the numbers off a bit. You can get a GTX 480 card for about $400 and a Quadro 4000 FERMI card for about $800. No doubt four Opteron's would be a beefy system.

[F J]

Revup, before u start feeling like im tearing one of ur legs off let me say that i understand ur pain, as a while back i too was confronted with the same type of problem..

instead of Revit i was using MicroStation which was perfect for architectural technical design, but, eventhough it was looking promising rendering-wise, it accomplished this slowly than desirable..

im sorry to say that in my opinion u've embarked on an unnecessary quest..

dont get me wrong, i hope ur business turns out to compensate the investment..

u speak of a "huge disconnect" but i think somebody already connected it for ya..

rendering that walkthrough was just silly.. ya couldav simply told ur architect friends to do this: http://area.autodesk.com/blogs/eddie/revit_to_3dsmax_with_vray_workflow

 

yes im aware of the difference/convenience between doing this inside the main software n exporting, but i think that if u'r gonna be channeling all ur efforts into comin' up with a platform based on technology others have already thought about u'll always end up being one step behind..

 

just my 2 cents..

[revuprender]

It looks like that is just taking the Revit model out and into Max and rendering with vRay. What is so special about that? What we have come up with is a workflow for RealityServer to run directly on top or side by side with Revit. This means you move the Revit model and it updates automatically inside the RevUp RealityServer window instantly. Again most architects do not own 3DS Max or know how to use it to make animations. With our product they do not have to buy any new software. There is also no having to re-set up the scene like you have to when going from Revit to Max.

 

When people see the way this product works at AU, they will definitely want to use it and see a huge benefit with it.

[rats]

Hey..... U r working for all of our dream future.. revup... Very good luck to u.... ur efforts will pay up sooner.. May God help u in ur mission...

 

but... what u r doing is every architect's ( designer's ) dream.. to visualize everything in realtime.. without anyother;s help.. but with current technology it still seems a distant future..

 

what u believe .. of ur dream ll pay off. sooner than i invest in a sytem.. is still feels like a day dream.. i wud still request u to.. please show us some practical productive good quality implimentation of ur technology..

 

GPU technology is promising.. but in today's context.. not practically adaptable.. ( coz.. no GPU renderer can give me my 8 million poly scene rendered in less time with great quality .. n that too affordable.. which has hardly a dozen of proxy cars n another dozen of proxy trees... ).. n i ve even bigger projects in line. to work on.. so once again.. good luck for ur efforts.. but u can't convince us to give up CPU stystem n Adapt to GPU...

 

N yes.. I am also an architect.. working in field of designing n viz for past 10 years..

 

N to AJLynn.. i very wel know that thre doesnt exist a system of GPU as processor... but i definitely wish it gets developed. .. as then i can adapt to octane or arion.. to render great n fast .... ( iray feels way slow n backword... looking at these two renderers..)..

[AJLynn]

I think I understand what Ry's trying to do, and it makes sense. You can set up a walkthrough animation in Revit without going to Max, but it's limited in how many of the usual animation-render-speeding-up methods you can do, and if you want some real mental ray quality it's going to take a very long time to render. This is a different situation from what most of the regulars here are used to, so while we're saying "why not just use Max and do it the way we've been doing it for years, which works fine and which we already know" that's not a path that's as readily available to his target market. (There are many people using Revit but not Max, and for the most part they're not the people who can properly optimize and farm an animation render anyway.) So a service that can take the setup straight from Revit and return an animation in reasonable time (and can work on a large variety of projects without needing to do any high level optimization) has value.

 

But we're all thinking in terms of what you can do out of Max. The users here are more likely to be the ones who can bring the Revit model into Max and make a mental ray or Vray animation properly, and have the equipment to do it, and are less likely to need such a service. If we see some nVidia people setting up a straw man argument between a crippled mr render and an iray render, we'll call it out immediately and add that it's not even a good render anyway. So Ry can have a system that works well for one sort of work, but the same base technology applied to this other sort of work might not realistically be helpful at all.

[AJLynn]

Ratnakar - I understand what you're saying, and I agree that as a desktop application, for an experienced user, iray leaves a lot to be desired (or a lot to be proven anyway) - though it's very convenient that it works fairly seamlessly with mental ray setups. Arion shows more promise for the single user but does add steps in needing to get your work into a side app and doesn't look like it would be as useful for animating (e.g., look at their gallery and they show a pretty low res render that took 2.75 minutes per frame, and they probably needed some high end hardware for that) so if I were in a production environment right now I'd definitely be sticking with mental ray - but I'd consider Ry's solution if I were going to want to spend my time on architecture instad of rendering but needed a fast way to make visuals for a Revit project.

[rats]

somewhere i read.. Nvidia is opening CUDA for CPU developers also... did anyone know of...

 

 

also by the way... archicad seems far better than. revit...

 

Edited November 14, 2010 by rats

[AJLynn]

CUDA for CPUs? I don't see why it wouldn't work, but I also don't see why you'd do it. Programming for CUDA is just C programming with a bunch of capabilities removed, so you could compile it to, say, x86 code, but look at the limitations you'd be stuck with - many independent memory spaces (garbage collection alone would drop your efficiency), restrictions on recursion and function pointers and nonstandard float handling, etc., etc., and you'd lose the advantage of many threads which is all you've got to offset the limitations. (CUDA doesn't make sense, for efficiency, unless you can break your process into dozens to hundreds of fully independent threads.) There's a reason CPUs are not designed the same way GPUs are.

[revuprender]

Thank you taking the time to listen to what I am really trying to do and understanding that there is value to it. I would not have spent over two years of my life working on this startup company if I felt that there was not a huge value to this.

 

RevUp RealityServer is just part of the solution. I did not mention this before but the first thing RevUp Render came up with is a Cloud Workstation. Now please consider this because the solution is simple but it is two products together that create a really BIG solution. The designer does all of their modeling, design work and everything except rendering on the RevUp VCS (cloud workstation) which you can see the kind of hardware being used on the website but it is very high end server hardware. When the user wants to render they simply send the model data out to RevUp RealityServer. The reason this is such a beautiful combination is because none of the workstation resources are being used and the designer can keep working with no slow down in performance.

 

There are many benefits to this solution but one is that it can provide designers from anywhere in world access to a workstation with hardware that they might be able to get access to normally. We use special remote connection technology so the end-user has a "like local experience."

 

One thing that drives me crazy is seeing architects, designers etc.. with workstations sitting next to them on their desk with a 1400 watt power working very long hours. There is proven research now and case studies of exposure to this kind of equipment at close range is very dangerous for your health. There was a school in California with poor wiring, powerful WiFi, and other items that were leaking out EMF's like crazy. The teachers and young students started getting cancer at a rate that was shocking. Europe is always way ahead of the United States with this kind of stuff and they are taking WiFi out of most public places. I myself have WiFi running right now in my home, but I try to remember to disable it when I am not using it.

 

That is just one reason I am a big believer in moving high end graphical workstations into the cloud. Other reasons are many but some are that the end user never has to worry about hardware failure since load balancing can move them to another server. The end-user also does not have to worry about upgrading since they always will be using the best hardware.

 

This solution might not make sense to some of you if already have high end equipment right now. For architects though who are sharing BIM models across multiple locations it makes a lot of sense. Instead of creating a problem and then trying to fix it with a band-aid, why not just not create that problem in the first place? A lot of A+E firms are taking their BIM model and sharing it across multiple office locations and then using bandwidth accelerators that are very expensive and only work so well to synchronize the model. Instead I believe they should just not create that problem and use the most simple and logical solution which is to keep the BIM model in the same central location and have cloud workstations that the remote users log into. If you are sharing BIM models then this solution makes a lot of sense to you.

 

I could keep going on but I just wanted to express this because this thread is about hardware and this applies even more than the GPU rendering stuff we are doing. We have been talking mostly about software and how it runs on the hardware.

[revuprender]

Thanks Andrew for your response. I feel that you guys now understand what I am trying to do with this technology. I think it was just misunderstood at first because RealitServer is meant to be used for final productions and this application certainly can be used like that. However our first goal is just to get a single still frame rendering out of Revit very fast and this has accomplished that. As you said there is no extra setup and it is really the same as just hitting render inside Revit.

 

I am attaching two renderings. The first renderings was done directly inside of Revit 2011 64bit using mental ray. Revit 2011 does take advantage of all the CPU cores so there is no limitation there and this was done on a dual socket Xeon X5550 2.66Ghz system with 24GB of RAM (that is a total of 16 CPU Nehalem cores! This first image showing the bridge took over 6 hours and 19 minutes to render inside of Revit.

 

The last image rendered almost instantly with RevUp RealityServer on just (2) GTX 480 GPU's. Then it took about 2 minutes to finish rendering to this quality. Imagine how much better the performance is when using several Tesla Fermi cards on the cloud. :-)

 

I know these are not the stats you wanted originally but now that you understand exactly what our goals are right now then you can easily see the very valuable use for this.

[AJLynn]

Well, I remain unconvinced of the usefulness for pro 3D types. Some observations:

-The first sample has no business being a 6 hour render on an 8-core box (that's not a 16-core box, it's an 8-core box). Given a skilled user and a real copy of mental ray (I'm not sure how much optimization you can do in mr for Revit in the new version) it's a 2 minute render. (If you can send me an FBX export I can demonstrate.)

-The second one is done on $1000 worth of GPU in 2 minutes; given $1000 worth of CPU and a skilled user with a real copy of mental ray, it's also a 2 minute render.

-Neither of these samples is of a type that most people here would use for clients.

 

Still, we're not the target market (non-Max-using, but Revit-using, designers). I can see this being useful for the target market - post on a Revit board and you'll probably get oohs and aahs. Do you have any samples you can use for wowing CGArchitect readers?

[Jeff Mottle]

Still, we're not the target market (non-Max-using, but Revit-using, designers). I can see this being useful for the target market - post on a Revit board and you'll probably get oohs and aahs. Do you have any samples you can use for wowing CGArchitect readers?

 

That's not true. There are tons of CGA readers who ARE the target market. It's just that most of the most vocal members of this forum are only visualization artists. There are over 200,000 unique visits to our site every month and our demographic is a broad as the number of people.

[AJLynn]

You know best. I think this part goes outside the scope of the Hardware board, though - but certainly deserves discussion on the Revit board...

[revuprender]

Andrew, I hear what you are saying. The settings in Revit are nothing like Max and it is very normal to spend 10 hours or more rendering one single still image of an interior with a lot of lights. Right now I do not have any Max models that I can show since we have focused on this workflow for Revit. Hopefully we will be able to do some nice tests with Jeff and John from Renderstream.

 

When I said 16 cores I was referring to logical cores. Each CPU has 8 threaded cores so 2 CPU's is 16 cores. Depends which way you want to look at it. For example BOXX often says they have 24 core systems with two Xeon 6-core CPU's since each core is hyper-threaded. If I look in the device manager on that system then I will actually see 16 processors listed. I can send a screen shot if you would like to see.

[F J]

Revup, i think what AJ meant was that u can spend 5secs exporting ur model as FBX so that he can show u an easy alternative workflow to counter the 17min/frame of ur walkthrough..

[revuprender]

Hello Francisco, sorry I can't release the model of that project because it was actually a real project I worked on for architect Peter Chermayeff and he would not want me to do that.

 

I am familiar with the Revit to FBX to 3DS Max workflow since I have used it many times myself. However it is not a simple workflow and certainly not seamless since there is a lot of setup needed in Max once you bring your Revit model in. The technology we have created is a very seamless workflow from Revit to RealityServer and it is actually just as easy as clicking a menu button.

 

Andrew also said that the solution I have makes sense for some people because most architects do not own 3DS Max and more importantly do not know how to use it.

[AJLynn]

Right, I think it's important to distinguish between people who are happy to do a two-program workflow and properly set up a mental ray, Vray, etc., rendering, and those who want to use one program with a solution that requires less intervening work. A product can have more value to one group than the other (or even a lot of value to one and very little to the other, in more extreme examples).

[F J]

ok, but even locked inside Revit wont those folk still have to incorporate into their workflows the new dealings with iray n its material system which is whats gonna give life to the objects in the scene? that also sounds like addicional learning for folk who cant be bothered goin' out of their way to gather extra knowledge, n not just press one button et voi la..

[revuprender]

They do not have to learn any new extra knowledge. Part of the technology we have created is for the Revit materials to be converted automatically to compatible RealityServer and iRay materials. This was maybe one of the most challenging parts of what we have accomplished.

 

As you can see from the examples I posted that the RevUp RealityServer rendering looks much better then the rendering that was done directly inside of Revit.

[F J]

oh wow, i did not know that.. lol

i guess u have it all covered then :X

[rats]

Hey.. guys.. check the homepage.. GPU farm for $2 an hour.. with 8 M2050 Teslaes at disposal.. whay say.... revup...

 

i m sorry to say.. but revup.. i m confused... as to how much u r aware about good quality prodection render.. n the complexity goes into it..

Edited November 18, 2010 by rats