View Full Version : ART VPS hardware rendering card vs. Satellite rendering

Hans Ulrich
10 October 2005, 12:17 PM
ART VPS hardware rendering card vs. Satellite rendering
I have questions about ART VPS rendering card and how it´s compared with using satellite rendering. The price would be about the same buying a Pure P1800 rendering card and buying 3, good satellite PC´s with dual core. So how would the P1800 work compared with satellite rendering using 8 CPUs?

I have neither used ART VPS´s cards or satellites redndering but are very interested in cutting rendering time.

Pure rendering cards and RenderPipe renderer ( (
How is the RenderPipe renderer?
Is it hard to learn?
Do you have to have programming skills to achieve top results?
How is it´s advantages/limits compared with Mental Ray?
How´s the materials and lights?

The images I have seen made with RenderPipe have had various quality. Many pictures are amaizingly good but some other pictures looks quite bad. On the forum: ( the quality varies.

One drawback is that RenderPipe only works on older versions of Maya and Max.
One advantage is that the Pure card seems to handle heavy scenes extreemly well in the viewport when working with them.
The new RenderPipe v.5.0 (now beta), has advanced Global Illumination, called SI (secondary illumination). How much better is this new one.

Reading on the forums about satellite rendering, it don´t allways seems to work 100%. And how does it work when tweeking a rendering during the test/setup phase? I believe it could be very usefull for the final renering but how much does it help on images taking about 30seconds to 5 minutes to render? Does the satellites have time to help or is the startup phase (reading textures etc.) too long for smaller renders?

Are there other, simple ways to render a single frame on several computers?
I want the renderer made in the Render View to render on all connected computers.

Feed back
It would be great if RenderPipe users would share their experiences on the renderer and the hardware. Especially how it works with GI and rendertimes with GI. It seems to handle Raytracing extremly fast, how is it with GI?

To see your RenderPipe rendered images with rendertimes would be nice.

I believe allot of people are interested in this quite new hardware. But it´s hard to really compare it with the traditional software solutions just reading about it on Internet.

So please, if you have used renderPipe, any Pure card or Render Drive, please share your experiences!

10 October 2005, 03:08 PM
Hi Hans,

Yes, the quality of the renderings varies at (, and that is because people are at varying levels of proficiency. Render-Art is an independent site, and not affiliated with ART VPS, and we encourage people to post images so that we can discuss them. Some times it’s the magic wand, and sometimes it’s the magician, and I think you find that with any rendering system.

For around $5,500 you can get a PURE with a PCI-X bus interface, drop it into an existing computer and give yourself 24-30 times the raytracing speed of software. The speed of the computer can affect rendering speed, so having lots of dual-channel memory and a fast hard drive helps. The PCI-X version card is a better value than the P1800, as it has twice the chips (16), faster interface, and isn’t twice the price as the P1800. The P1800 is still quite fast with 8 chips at around $3,400US. Dual PURE is $6,659, and I would still go with the PCI-X if you have the interface. Be sure your computer has space for a full-sized PCI card, and clear space on the motherboard, too. Heat sinks and ribbon cables can get in the way.

With the PURE cards and dual CPU systems (or dual-core and/or HyperThreading) the render plugin does not use up all of your CPU time, and on my dual Xeon system I can render at full-speed and still model without a noticeable slowdown. So there is little need to have another computer, or a render farm, just for rendering. That said, I get a little better performance out of having the PURE in my single-CPU P4 with the dual-channel memory than with my dual Xeon, and on large scenes I can use the memory for modeling so will offload to the other computer. I can also render from my local machine to the remote machine within max, just as if the card was in my local computer, and there is no need to move the card to render in any copy of max on the network; everyone can share the resource, as long as it is available. A gigabit network helps for any network rendering, but isn't required.

Rendering to a remote computer takes virtually zero CPU time on the local machine once the scene is sent to the other machine.

RenderPipe – the rendering plugin for max, Maya and Catia – runs of the current versions of those programs as far as I know. ART VPS has been a little behind from time to time with Maya as I recall, but the 3ds max plugin works fine since the code base for max hasn’t changed the last few versions. I used the AV4 plugin with no issues in max r8 beta.

The PURE card does not do any viewport functions at all; it is merely a rendering device.

The new AV5 Beta with GI (and many other enhancements) is due out soon, so until I get that I can’t comment on how fast that works. The samples so far are wonderful, but no benchmarks yet. RenderPipe has had Secondary Illumination for quite some time, which works well but slows things down considerably, just as adding GI in Mental Ray slows things down. I tend to use the 3ds max Radiosity function with PURE for large scenes, and SI for selective areas or on smaller scenes.

As to some of your questions....

Easy: RenderPipe is a very simple interface, and not hard to learn in that respect. Much easier than Mental Ray, certainly. The less you need to fake things, the easier the interface. You can see the latest interface and features at:

RP Lights and cameras are much simpler than standard objects, and give additional features such as lens flare and area light capabilities.

You don't have to use the RP Materials, Lights or Cameras to render a scene. Architectural materials will be supported in AV5. You should substitute Raytrace materials for RP materials, however.

Programming: No programming skills are needed whatsoever. If you want to get into programming of shaders you can, but there is no requirements for that. A large subset of Renderman features are supported.

Mental Ray: Easier interface in RP. RenderPipe does not support microtesselation displacements, but you can achieve that affect with a modifier in max. RenderPipe supports RPC. RP supports Radiosity and Secondary Illumination, but not true GI until next version. RP is much, much faster, but RP raytraces Everything. For simple scenes, MR and scanline may be faster because of this. When you add any more complicated functions (reflections, refractions, motion blur, DOF, inverse-square light attenuation, area lights, and so on), RP is the way to go in my opinion.

With RP you can use RenderMan shaders, which opens up a lot of possibilities. ART VPS's HDRI Skylight is a Renderman shader, for instance.

Satellites: I have done a lot of rendering on farms up to 15 computers, which is a royal pain in the neck. Managing that takes a lot of time and energy, and can be quite stressful when things aren’t going right. Buying and upgrading those machines over time takes a lot of money. With a RenderDrive or PURE card, you only manage the machine running the network server/manager and that copy of max (in my case), and not 15 machines. PURE does wonders for my blood pressure. We’ve had our PURE cards for many years and it still beats software, so I feel that it has been a good return on investment.

Tweaking Phase: RP Produces a full-frame preview at about 25% completion, which is a great time saver when you are in the ‘tweak phase’ of a project, and adjusting lighting, etc. With Scanline and MR you have to wait. With Maxwell you get the full-frame progressive update also, but that is not hardware accelerated.

Rendering a single frame on many computers: In max you can Split Scan Lines in a network rendering, which works great with RenderPipe. I will do this even with only one machine rendering as it protects you in the event of problems with the machine (lose power, for instance), and I can check the strips as they are completed to ensure that the rendering is producing what it should (I didn’t mess something up). I cannot comment on Maya’s capabilities with splitting renderings.

RenderPipe does allow you to send any scene to multiple rendering devices at the same time. It will split the scene based on the speed of the remote machines (RenderDrive vs PURE) and wait for completion. So larger portions are sent to RenderDrives, and smaller to PURE. This should work in max and Maya.

Feedback: The renderings that I put onto the ( site all have the render times listed, and are mostly high-resolution images. ART VPS has a page with speed comparisons at and I have found their benchmarks to be an accurate representation of the speed you can expect. I’ve also done some comparisons to mental ray using MR’s samples, and have them posted on the Render-Art site.

I think I got everything. You are welcome to post questions at (, and hopefully some of the members there will post here, too. :)

Have a great day!

- Jennifer

Hans Ulrich
10 October 2005, 07:13 PM
Thank you for your nice reply! :thumbsup:

I have a small question.
You wrote:

"The PURE card does not do any viewport functions at all; it is merely a rendering device."

Besides the rendering, does it not work as a regular graphic card?

10 October 2005, 08:09 PM

PURE does not work as a regular graphics card at all. Think of it as a rendering co-processor. You still need a good graphics video card to handle normal video.

nVIDIA has the Gelato software that uses your graphics card for rendering. You'll also need the Amaretto or Sorbetto plugins for Max or Maya. I've been working with their latest beta, and it is unexpectedly difficult to get it to render the way I expect. It needs more time, I feel.


Hans Ulrich
10 October 2005, 08:42 AM
Ok, I see!
Thank you, very much again Jenni!!!

Anybody else who has anything to say about Art Vps or Maya Satellite rendering?!!!
Anyone who have tested AV5 Beta?
I think Art Vps´s products sounds very good. But it still dont seems to be used that much.

10 October 2005, 12:37 PM
Hi Hans,

AV5 should be out soon, and we'll be discussing it and sharing samples at the site, so check in there when you get a chance.

I think a hardware rendering system has a more difficult time with getting out there to the people because you cannot simply download or 'share' a plugin and really give it a real try. Software will always have that advantage, at least until ART VPS makes a RenderDrive available on the Internet for testing!

The license I have for MR doesn't allow satellite rendering, so I can't comment on that. Between the hardware (computers) and the licensing for MR, the cost builds up quickly.

The PURE doesn’t fit all needs, certainly, just as MR doesn't fit all needs, but if you are in need of beautiful raytraced images, it is the way to go in my opinion. Perhaps I’m impatient, but I don’t like to wait any longer than I need to!

Have a great day!

11 November 2005, 03:52 AM
also from what i have been told by the Arts VPS team, the renderer they use is based on the Renderman specs and can use a .rib file too, with a small amount of tweaks.
This is wha they told me.

"Although we support the RenderMan interface, we are not 100% drop-in
compatible with prMan specific tools like the RenderMan Artist's Toolkit.
Specifically, Slim shaders will not work without editing them because of
the restrictions in our shader compiler.

Most RIB files will work, although some features of the RenderMan
Interface 3.2 are not implemented, such as Subdivision Surfaces and blobby

In summery ,we're not a drop-in replacement for prMan but it should be
really easy to retarget any RenderMan tool chain to our renderer."

Which for me is pretty cool, because i am learning Renderman at school :D

11 November 2005, 10:34 AM
The renderman support on those cards is incredibly crippled. Large amounts of the RiSpec are limited or do not work at all. The shading language is crippled and has many strange caveats like not allowing temporary assignments for the result colour etc.

Hans Ulrich
11 November 2005, 01:19 PM
Thanks for your replies!
I’m thinking, for a start to speed up my renders, to buy a computer with dual AMD Opteron 265 Multicore CPUs. This will give me 4 CPUs and surely speed up my render time compared to my old P4 2.8.

I will mainly use it for architectural visualizations, stills. I will probably stay with mental ray for a while.

Later I might get a Pure render card. I would really like to try one out!
Any one who will sell me a used one (or a Render Drive)?!

11 November 2005, 03:04 PM

RenderMan is a specification, not a renderer, and every renderer that uses the RenderMan specification has different subsets that it supports. Even PRMan, Pixar's renderer, didn't support Raytracing until recently, so just because the interface is supported, it is extensible and each renderer can and is different.

There are some very good free RM renderers out there if you are playing with writing shaders. Check out ( for resources. If you are using max, Maya or Catia and need Raytracing power, then PURE is a lot cheaper and easier than a renderfarm. :)

Have fun!

11 November 2005, 04:14 PM
i didnt need the renderman lesson but thanks ;) I guess i should have said RIspec or Renderman compliant, but most people glass over when i go off on RISpec info:P I started on 3delight and hand writing my rib files, or generating them at home and editing them in VI, because i cant get Liquid to work, i am not the best code writer. I finally took the plunge and ordered the edu priced Renderman PRo Server and RAT, im stoked to fire it up on my stystem!. Future looks director an all, we have very boring dinner conversation. I am just looking for a way to get my renders done faster. How is the heat generation from the card? it is 16 cpus so i imagine that the thing puts out a decent amount of heat.

11 November 2005, 04:19 PM

That was more of a comment in response to another person's comments, and a general statement for anyone interested. :)

Have a fun day!

11 November 2005, 04:23 PM
oh and it was some good info about the RISpec compatability, i was hopeing it was better so i could have my own little render farm.
I remember taliking about Renderman stuff with another student at school, he is a games artist and he looked like steam was gonna shoot from his ears. I think i ended with its all very hard, and you will never need to deal with it.

11 November 2005, 04:37 PM
Another thing for people to look at would be the RenderMan for Maya plugin, Renderman is pretty fast in the scale of renderers, and you get really good results with the RMfM plugin. So you can get tons of the benifits of Renderman loke speed and nice DoF and motion blur, but you loose out on all of the fun code stuff like building shaders from scratch and .dso and i dont think RMfM generates .rib. Anyhoo, sorry fro the thread hijack.

Hey HAns think of the electricity bill that would happen with multi computers, that is another bump in favour of the pure card.

11 November 2005, 04:56 PM

I use 3ds max, and the RenderMan interface to PURE is transparent and not something anyone has to understand in that environment. The is true same for Maya and Catia. The RM spec is used for compatibility between platforms. You can program your own shaders, if you want.

I do use a RenderMan HDRI Light shader all the time, and occasionally will use a RM surface shader, but usually don’t have to go beyond what is available within max and the custom shaders from ART VPS. I use Backburner for the renderfarm function, whether it is for one computer with two cards, or if I split our two into different computers.

There is no heat generation issues on the PURE cards that I have seen. I use standard cooling fans in the case, nothing special. If you run dual PURE cards, just make sure you have a 350+ watt power supply if you have a lot of other stuff in your system. A standard PURE card has 8 chips with 16 cores. The new PCI-X version has 16 chips for 32 cores.

Another fun fact is that on dual CPU systems you can render and model with little affect on modeling speed. No need for another computer at all, as long as you have the memory for both processes.

Have fun!

11 November 2005, 05:41 PM
Do a search for more info on the renderdrive - I've heard nothing but complaints about them. Hardware rendering seems like a good idea, but with pc's becoming cheaper you'd be better off building a render farm for the money. Plus, the cards are not upgradeable and you're stuck with their renderer/interface... at least the other way you can choose different rendering engines for different scenes/needs. Also, if you're so sold on hardware rendering, there's always Nvidia's Gelato.

Someone somewhere wrote they convinced a friend to buy one and it turned out really bad. If you're using Maya, check out Turtle by Illuminate Labs, and I saw today that Renderman for Maya has been released, although this is probably overkill if you're just doing architectural renderings.

Ohh, found it - here's a link to the chat thread:

11 November 2005, 05:46 PM
Plus, let's say on the off-chance you have a 2400 frame animation to render - you'll have to do it yourself since none of the render farms on the web support the Art VPS system... meaning your PC will be tied up for a long time rendering.

It seems to me like if it was really so incredibly faster than everything else out there, everyone would be using one.

11 November 2005, 06:13 PM
Way back when, Art VSP was kind enough to lend us a Renderdriver for a 2 week period to test on our system. Unfortunatly, because of those tests, we were able to decide NOT to purchase the system. The cost of the machine was very high and it did not prove to be even a 1/5th as fast as they predicted it would be. After some quick math we found that a few extra dedicated computer for distributed rendering (MR, Vray, etc..) was significantly faster, 1/5 of the price, and a LOT easier to upgrade. But as I said, this was a long time ago. Since then things might have changed. This Pure and Renderdrives may be faster. But I also know that regular computers are also a LOT faster. I would urge you to ask for a trial version if you can. Then compare that to the speed of something like this:

My guess is that the quad AMD would leave it in the dust...

11 November 2005, 06:51 PM

The 'company of the friend of this guy' sounds like a larger company, and a better familiarity with the tool may have produced what they needed. It’s hard to say without seeing the renderings. Being under a deadline is not the best time to switch renderers, IMHO, but in the heat of things I can understand. I don't like 'friend of some guy' analysis of any situation. You can do cruddy work with any rendering technology; don't necessarily blame the failure of someone to render a nice scene on just the tools that they use. I just recently tried one of my scenes in the beta of Gelato, and it looked terrible. All that means is that I don't know what this tool needs compared to what I had before. I drive by a REALLY BAD rendering of a new strip mall almost every day, but that doesn’t mean that VIZ or max, or whatever they used, is worthless. Switching to a different renderer takes a bit of time to figure out what it needs as compared to what you had. That said, the RenderPipe lights, cameras, etc, are pretty simple, as is the render interface, compared to anything else I've used.

PURE / RenderDrive are Raytracing renderers, and that's what they do well and what they do quickly. Some things don't need that power, don't need raytracing, and in those cases software may be faster. A lot of scenes don't need mental ray, Maxwell or VRay, either. My own personal experience with PURE, mental ray and VRay is that PURE is 12-15 times faster than my dual 2.4gHz Xeon system for raytraced scenes with the same or better results, plain and simple. If you don’t need raytracing (and the other things it does well), then I wouldn’t recommend these devices as it raytraces everything. It is not for all applications.

Once GI is supported directly (in addition to the already-supported max Radiosity and their Selective Secondary Illumination) in PURE/RD, then the complete feature sets will be more comparable with mr and VRay’s GI, and I can compare rendering times with these different GI tools.

A software rendering system is certainly less expensive on the small scale, until you start buying node licenses and extra computers to do the rendering. You get mr for free (one node) in max and Maya, and can get the free VRay version, and this is a big part of why it is more popular. And those that don’t want to support software development can get the (for lack of a kinder word) ‘shareware’ version from their favorite file-sharing software. Having easy, cheap access leads to higher use.

I have no patience for software raytracers at this point, nor do I have patience to manage a renderfarm. Been there, done that, and my blood pressure is much better with managing one or two computers rather than the 15 that I had at one point.

Once I get my quad dual-core AMD I’ll let you know which is faster! :)

Have a fun day!

11 November 2005, 07:34 PM
A software rendering system is certainly less expensive on the small scale, until you start buying node licenses and extra computers to do the rendering. You get mr for free (one node) in max and Maya, and can get the free VRay version, and this is a big part of why it is more popular. And those that don’t want to support software development can get the (for lack of a kinder word) ‘shareware’ version from their favorite file-sharing software. Having easy, cheap access leads to higher use.

Actually Vray node licenses are free. The distibuted rendering licenses are limited to 10 nodes per license, and that is 10 computers (not 10 cpus). So lets say you have 10 quad AMDs, that would be 80 processors for only one license of Vray at $799. On the other hand the computers would cost around $100k. So software cost is virtually nothing (for Vray). MR is relatively cheap as well... compared to lets say Renderman.

Like I said, we had a very different experience with the Renderdrive, which is why I urge the user to try it themselves. Again, this was 5 years ago, when I worked for another company.

And wait... did I hear correctly? Art VSP STILL does not have native GI? They promised it 5 years ago as being right around the corner. Boy it really looks like we made the right choice back then.

11 November 2005, 08:08 PM

That's good to know about VRay licenses. Mental ray is per CPU, and last I heard was $1,500 per node, but that was a year ago. I hope that isn't per core! :curious:

ART VPS has full support for max Radiosity, and has had a per-material 'Selective Secondary Illumination' material for years, both of which I use based on the situation. They have definitely been behind the curve on full GI, and it sounds like they are getting close with the latest beta. For users GI may seem easier than Radiosity, perhaps, but it certainly is not faster to render. Radiosity just takes some time to set up, and an understanding of what it needs from a modeling, lighting and settings standpoint.

A quad dual-core AMD system look like it will come in around $1,500 for the motherboard and around $1,000 per CPU, plus memory, etc, etc. A Boxx mid-range dual AMD dual-core system is around $6,000; they don't list the quad units that I found.

Have a great day!

11 November 2005, 08:32 PM
A quad dual-core AMD system look like it will come in around $1,500 for the motherboard and around $1,000 per CPU, plus memory, etc, etc. A Boxx mid-range dual AMD dual-core system is around $6,000; they don't list the quad units that I found.

Yeap... the $10k price came from a Boxx rep at Siggraph while I was looking at the quad AMD. He said, they expected them to start at $10k. Either way, they are not cheap. But if you outfit it to perform like a Renderdrive you can forgo the expensive GPU... all you need is the 4 procs, maybe around 4 gigs of ram, and a fast HD. Oh yeah... gigbit network too. Of course you can also do this:

That would give you 4 CPUs (2 dual cores) at around $3,300 will all that you need. 20 of those would give you 80 CPUs at a cost of $66k. Could save you some money.

11 November 2005, 09:08 PM

Myself, I think the RenderDrives are too expensive when you compare the cost of just putting a PCI-X PURE into a fast system. It is basically a lean-and-mean Linux box, now running single and dual 64-bit CPUs.

If you put a PURE into a slow system it can greatly affect the performance. PURE is actually faster in my single P4 system with dual-channel memory than in the dual Xeon system w/o dual-channel memory. But it uses 100% of the CPU. I haven't tried the 64-bit drivers yet. With RenderDrive you should at least be assured that the system is running as fast as it can.

Thanks for an interesting conversation! :)

11 November 2005, 03:52 PM

That's good to know about VRay licenses. Mental ray is per CPU, and last I heard was $1,500 per node, but that was a year ago. I hope that isn't per core! :curious:

Actually in Max 8 you get unlimited MR render nodes using network rendering (or something like a 999 CPU limit)... so the per node cost isn't so much a big deal now. You still only get I think 8 CPU distributed rendering, which is only good for stills anyway.

As there seems to be no arguing from jenni_o about this, I would also reccomend you get a test system and compare it to other rendering engines, especially for architectural work. As a raytracing system maybe it's great at tracing 255 reflections/refractions in a crystal, but in arch viz you may need at most 8 or 10. I'd also be interested in when the last time they upgraded the actual chipset is - I think it's always been based off the AR350's, so now that it's PCI-X it has a higher data bandwith but the procs are the same speed.

Everyone wants to convince you that what they're using is the best solution - even if it's not the best solution. That's why the 'Why is software x better than software y' arguments always get closed - people want to convince themselves they use the best product and don't always want to hear otherwise. I use Brazil, Vray, and Mental Ray, sometimes based on my mood, usually based on what the output is. They all have their + and -'s... and I couldn't honestly say one of them is -the- rendering solution. But, being software based, I have the option of buying all of them - for less than the price of one renderdrive. And still have money left over for a rendering node. Try them out, see what you like and what works best, and go that route.

11 November 2005, 04:41 PM
I use Brazil, Vray, and Mental Ray, sometimes based on my mood, usually based on what the output is. They all have their + and -'s... and I couldn't honestly say one of them is -the- rendering solution. But, being software based, I have the option of buying all of them - for less than the price of one renderdrive. And still have money left over for a rendering node. Try them out, see what you like and what works best, and go that route.



Hans Ulrich
11 November 2005, 05:34 PM
Nice to see this thread is getting more posts :thumbsup:
And thanks again folks for all the info!

I use Maya and Mental Ray. I wood like not have to learn a new renderer. For Maya, Turtle might be an alternative. What I’ve heard it should be easy to learn.

I have never used Render Man and it sounds kind of complicated.

Seems like Pure/RenderDrive are very fast with the things they are made to be fast at (raytracing). Like render a nice car. But might be slowed down when using GI, as in a architectural visualization.

When GI is fully supported there might be some other render times. Hope it will happen soon...

I don’t think I will need a whole bunch of computers. 4 CPUs might be enough for me. To avoid using multiple computers to render one frame I believe a Dual Dualcore (total 4 CPUs) will be a good solution. And a Quad Dualcore (8 CPUs) would be very nice. The later one would however be very expencive! About 3 times more expensive and twice as fast.

A good point is the freedom to choose renderer.

Reading on the cards seems too good to be true.

As a starter I probably will get myself a dual dualcore system and stick with Mentay Ray or Turtle.

But however if someone will sell me a pure card I would buy it for evaluation. It seems as it would be easy to get a used one since there are so many people complaining about them. Or maybe they are not so bad since there not seems to be that many for sale…

Thank you for the information!

CGTalk Moderation
11 November 2005, 05:34 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.