Uitgeverij ONTMOETING

Render programs

Index page

Do you need a Render Farm?

 
Large studios and 3D artists need a fast computer system to render all their work in as less time as possible. But it should't be restricted to those larger firms. Even one-men companies and artists need to put the renderers through their machines and that takes a lot of time, especially if one has only one computer, other creative work has to wait.
 
Additional machines are vital for this sort of work. And the good news is that the hardware is available, just choose the right one with an attractive price tag and don't forget the licenses fee. We consider two types of render programs:
 
1. conventional CPU based render programs
2. GPU CUDA (NVIDIA) based render programs
 
The first option only needs a fast processor with as much cores and enough RAM as efficiently possible. And of course a good graphic card that supports the program to the screen. Most of the render programs are of the first category (for instance Artlantis and Maxwell Render).
 
The second option is the rather new possibility where the render program is exclusively dependable on the number of cores in the Graphic Card, the speed of the processor(s) and the amount of RAM in the GPU. Only a few programs fell into this category (for instance Octane Render, Bunkspeed, iRay and Mental Ray).
 
The hudge features and possibilities of a GPU CUDA based render system is obvious to everyone involved. The artist is able to put together one render machine with two or more fantastic GPU's on board to make one big multi-core parallel processor engine. All renders fly through this engine if it was a Render Farm with hunderts of single computers in a data centre.
 
 

So few render programs based on CUDA GPU's on the market?

Everyone agrees that it will be 'on our wishlist for future updates' but that is easy to say in a Render Forum to impatient waiting customers. Who sometimes do not know waht they are really asking for. They probably has to change and renew their computer hardware outfut to incorporate the new GPU's with the new PCI & third generation of slots in the motherboard!
 
When or even IF that time comes, that is never specified by the render brands administrators. There is a Dutch expression for that "stretching and stay alive". Or in other words waiting for better days or perhaps take on your whishing well or magic stick. In the meantime there is a growing group of professional customers that become aware what is actually going on. And they are not amused at all by the misty answers, following numerous Render Fora.
 
 

The reasons that big brand names in Render World do not have such a render program on their shelf :

 
1. we'll wait till things changes in our direction.
 
2. we have invested so much time in the conventional program, that it whould be a waste of time and money to change right now. Or throw away our customer base by changing from CPU to GPU.
 
3. the parallel processing hype will blow away and something others will take it's place, for instance faster CPU processors or else.
 
4. one big world leading firm did just right that (the nr. 3) and that is Intel. They told the whole world each day after day that the CPU will eventually overrul all the parallel GPU-stuff. "It is easier to maintain and everyone could incorporate and use it in all kind of (other) programs too".
 
5. See "Intel, NVIDIA draw battle lines over parallel processing" in Computerworld june 2010.
 
http://www.computerworld.com/s/article/9178519/Intel_Nvidia_
draw_battle_lines_over_parallel_processing
 
The title of NVIDIA to the world was rather clear: "debunking the 100 x GPU vs. CPU myth, evaluation of throughput computing on CPU and GPU". NVIDIA was discussing the Intel paper about a performance test CPU versus GPU stating: "Our data shows that GTX280 only has an average of 2.5X performance advantage over Core i7 in the 14 kernels tested".
 
And of course in the meantime the GTX type mentioned is a rather slow and dated GPU nowadays. But the Intel researchers were not amused when they were asked to produce the necessary evidence and research setup. They did not do that at all. Without this vital information it was not possible to repeat Intel's test to confirm the paper's findings. Or to understand how the used programming code was optimatized for the chips. NVIDIA Andy Keane (general manager of GPU computing) said: "It's just bad science". Scientif proof is only proofed if it is declaired and documented enough to check the findings.
 
6. But time changes and even Intel was no longer able to go any further on that road and they told the world that parallel processing could be the future with their newly developed Intel Xeon Phi Co-processor, which they believe that it delivers a greater aggregate performance for highly parallel processing while elimating dual-programming architecture. Go Parallel seems the new direction (Intel introduced the Xeon mid 2012).
 
 
Maxwell screen during network render option with several Nodes.
Each node needs an extra software license.
 
 
7. In the past the GPU was the computer part that brought all those ones and nulls of the processor in the right porportion to the screen. But that was then, nowadays with NVIDIA CUDA cores a GPU is more or less a fantastic working horse that is able to calculate miljons of expressions with lighting speed.
Quite another task then the CPU, that has a much wider range in handling series of instructions in sequential order. The new GPU is no substitute for the CPU, the CPU remains the main processor in the motherboard that controls everything. The CUDA GPU is a parallel processor that can handle rather simple instructions parallel (at the same time). More cores in the GPU calculate directly into more throughput per second. Or faster rendering.
 
8. Although the modern CPU are copying the parallel processor's way by adding more cores to their processors. But that number is not at all comparable against the big CUDA GPU's.
 
9. All those marvels of parallel processing demands one other vital part in the process: an Application that is optimized to run on a CUDA CPU.
 
And that whole process not only can be time consuming but it IS an WILL be time consuming and very expensive. Figure out that the standard development room had to be expanded to accommodate new fresh developers just graduated from university. Under the leading professor Tanenbaum of the VNU university in Amsterdam the university is selected as a CUDA Teaching Center educating students for parallel processing
New CUDA centers continue to spread the world about GPU-related parallel courses.
 
http://tweakers.net/nieuws/80471/vu-amsterdam-aangewezen-als
-nvidia-cuda-teaching-center.html
VU Amsterdam aangewezen als NVIDIA CUDA Teaching Center (Mrt 2012)
 
http://www.cs.vu.nl/~rob/CUDA-teaching-center.html
 
The course material is:
class 1: Introduction, performance metrics & analysis.

class 2: Many-core hardware.

class 3: Cuda basics.

class 4: Advanced Cuda.

class 5: Many-core processing for the LOFAR telescope.

 
Also, they give hands-on experience with CUDA and/or OpenCL in the parallel programming practical.
 
 
10. The first applications that take advantage of the huge speed of parallel processing are found in the financial analysis and scientific field. A higher performance outweigh the added cost for code optimization. And in the Render World Octane Render managed to put version 1 on the market at the end of 2012. Further Bunkspeed renderer. NVIDIA is issuing fine render programs (iRay and Mental Ray, besides several other well known programs where the engines are put into) that depend on CUDA cores. Every brand can buy them and add their own interface to it. In Render Principles you can read all about it.
 
It is very hard to program efficiëntly for GPU CUDA and special skills are needed. Many months of testing and correcting will and must follow after ... years of work. A lot of money is involved to get it right and in the right timeframe and that can be difficult for a lot of brands.
 

11. Again: who needs a Render Farm?

Several ways to approach this question: build your own or let it build to your specifications. Or buy pre-built standard boxes that are put together to form one big render engine. Or shift to one of the more then twenty Render Farms on the internet to let them calculate your renders in more or less time. Of course by a Render Farm that's suits your render application!
 
Each solution has it's own advantages and disadvantages. Only renting virtual time elsewhere seems the right choice: only buy something when needed. No costs whatsoever when you do not need the big engines. Watch out for the blurred calculation prices, it sometimes looks like the Mobile Telephone world. Calculators to depect your price concluded as if you are rendering beforehand with your own computer. But the answer is simple if we incorporate the new parallel CUDA processor capabilities and dito compatible render applications. Then the balance is easy pointing in another direction, GPU rendering.
 
Or, if someone draw designs that must remain secure whatever the cost, then a Render Farm inside is the 'only' solution. Or if there are so many designs to render that it is logical to incorporate all these machines inside the studio or company. They are 'always' available day and night, when needed if one has foresee in a proper power backup aswel.
 
 
 

"1700 CPU's to your desktop at only 5 cents / GHz"

 
Render Farming is the solution to a graphic job to spread the work over to many machines at the same time. New is that the configuration and feasibility of such a system is on a low budget possible. The question remain: is the studio big enough to put all those machines together to make a new troughput in rendering?
 
The movie- and animation world could not exist if there where no Render Farms. How much faster is possible? Dramatically, if we speak about tens or even hunderts of renders. Just put in one render and the need of a Render Farm has gone. Most of them are not able to devide a page into more sections that could be calculated by different machines to stich together later on. So one page in one machine is not uncommon. Render programs that are network compatible (for instance Maxwell, V-Ray, Cinema 4D, 3ds Max etc.) can split the calculation in seperate processes. And that translates directly in a faster workflow.
 
In the Render Farm the number of cores and the processor speed concludes if this is faster or slower then the stand-alone machine at your desk. Render Farms magnetizes their customers by telling how much faster 100 (animation) or more renders could be calculated at the same time. But if this is a daily practice, then the studio already will be prepared for such sort of work. Common practice is different and one has to deal with customers that are changing till the last deadline minute.
 
 

Index page


to the top