so technically is it possible that u can build a render less machine i am thinking
some specs for rendering
Intel Xeon E5-4650 x4 processors physically
Intel® Server Board S4600LT2 motherbord that supperts 4 of them
and 32 gb of ram
and GeForce GTX 690
my boss wants no rendering time like max to 5 min like if a render is 25 min the it should be like 5 min tops can it be done or not
:surprised
Hey man,
If your talking Mental Ray/Vray/Renderman type rendering, then no. Those chips are designed for scientific/financial applications.
-AJ
yes i am talking about Mental Ray/Vray/Renderman type rendering
if not then how can it be build
Hey man,
Can you provide a little more info about your projects? Are you rendering large single frames that take a while, or animations with lots of frames that don’t take long each? Whats your budget?
-AJ
yes i am rendering large single farm that takes a while a lot longer but the machine that i already have its xeon processor that is more then 5 years old it is like 2.0 and 8 mb l2 cache no l3 cache ,
Hey man,
Have you looked at spiting your images up into tiles and sending the individual tiles to different machines on a farm?
-AJ
i don’t have a render farm i only have one another machine and am already using that as second rendering machine i use both methods split image or send single frame on that and in the meanwhile work on my machine for some more final changes
did i mention i have reduced the rendering size to 1920x1080 from 3096k res, at least i need that to look the images clearer look good in a 42inch led screen when presenting the image,
and after the presentation i need quick iteration like in 2 days max
Hey man,
If your comfortable with using tiles, than its probably best that you get a small farm of either dual or single socket machines going. Most software comes with rendering clients as part of the license, so that shouldn’t be an issue.
-AJ
but is it effective for me like doing all the rendering preparation to send one single frame on a render farm and what about http://www.boxxtech.com/
Intel I7 Quad Core Enhanced Performance Processor 4.5ghz 1
16gb Ddr3-1600 (2 Dimms) 1
Nvidia Geforce Gtx 680 2gb - (special Order) 1
120gb Ssd Sata 6gb/s 1
24gb 2.5 Ssd Cache Hard Drive (paired With Non-ssd Drives Only) 1
1tb 7,200rpm Sata
Hey man,
Your really going to have to provide more information about your projects. What software do you use? What resolutions do you render? How long do your renders take?
When I would do renders for large banner ads or billboards, I would split my images up into 4x4 tiles, and send them to a farm of 10 machines. Sometimes the images could top 16K pixels across, and take an hour to render across 10 machines. Its a pretty common way to work.
You also don’t need a GPU in your render slaves.
-AJ
mainly for architecture renderings and 3ds max and vray, but the thing is that if am going above 3000k render i would be happy for a rendering farm since i am only rendering at 1920x1080 for final render and 1280x720 for semi tests and in house iteration for some quick design changes and do u know how painful it is to setup a network render every 5 min later plus u need 4 to 5 systems which cost around 14k to 16k dollars 16 thousand dollar spending on 4 machines and making it a render farm is good or spending 9k dollars on one single machine i am only 1 guy here have to do every technical thing my self :banghead:
Hey man,
I’m not sure where your getting your numbers from, but in the US, a really screaming single socket render slave can be had for under 2000 USD. A mid range dual socket node could run 4000 USD. The quad socket nodes aren’t meant for 3D rendering.
If your using Backbuner for your rendering, all you need to do is save your scene file and refresh the job to run a new iteration. Using Cinema4D and Net Render is even easier.
-AJ
dude have a look at this rig
http://www.boxxtech.com/Products/3dboxx-8980-xtreme
and custom it with gtx680
Hey man,
At the scale your buying at, getting 2-4 single socket nodes is usually the best option. It might also be a good idea to get a small rack, and put your nodes in 2U cases.
Check out the Cinebench scores for the processors your looking at, and compare that with the price. The single socket nodes give you a much higher bang for your buck.
A pair of these 1400 USD Dell systems will equal that 10000 USD Boxx system in rendering power.
http://www.newegg.com/Product/Product.aspx?Item=N82E16883155601
Remember 3D raytracing does not require a graphics card, and most renders don’t use any more than 8GB of RAM. You also don’t need a SSD or a large HD.
-AJ
There are a few ways to improve rendering performance for single frames. If your workflow doesn’t require advanced rendering features like motion blur, instances, displacements, fur, then you might have the best luck with a renderer that uses the GPU for acceleration. Product renderings and architecture renderings are good candidates for GPU rendering.
If you don’t want to change your workflow dramatically for whatever reason there are two ways to improve the rendering performance for single frames. One is to build a very high performance single machine, and the other is to distribute the rendering of the image to multiple machines. Some renderers can send individual tiles to other clients on the fly, so if you start a render you see the tiles pop up as soon as they are done. If the renderer doesn’t support that the tiles can be sent off as individual renders and then reassembled after all of them are complete. Each method has pros and cons to consider.
GPU rendering: good bang for the buck and relatively inexpensive, doesn’t scale up to huge operations, only a few renderers to choose from, very limited rendering features (see first paragraph).
Distributed tile rendering: a good balance between expense and flexibility, most renderers support distributed tile rendering, can scale to huge operations, requires some tinkering and networking to setup.
Uberworkstation: very expensive and not very good bang for the buck, doesn’t scale to huge operations, doesn’t require any changes to the workflow.
If it were my money I’d find a way to make the distributed tile rendering work. Maybe upgrade the workstation if it’s old but also get a bunch of commodity machines to help with the rendering (four to eight cores, onboard graphics, 16-32GB of memory, small hard disk).
They’re still chips like anything else. They’re just used mainly in financial/scientific applications more often because of deep pockets and needs.
The main issue would be how many cores does Mental raay/vray/renderman support right now and is maya/3dsmax/zbrush compatible with windows 2012 server + an additional CPU license to support 4 CPU’s
For what you get, this machine with E5-4617’s CPU’s could be had for a somewhat reasonable price:
http://www.thinkmate.com/System/HPX_XS5-4460
I know… lol. 
I think English is a second language for the OP, so I was trying to keep things as simple and direct as possible.
I’ve read its the massive thread count and memory bandwidth of those machines makes them ideal for certain calculations, usually running custom written and complied software for solving physics and chemistry type problems. The biggest one from Intel has 160 threads on 80 cores.
I also know there are certain applications in the financial and engineering world that can run over 100K per node, and its best to get as much power on a single seat as possible. I’ve also read about them being used for industrial control systems, like a machine that needs to input and process huge amounts of data very quickly, and spit out the result.
I don’t think however that the quad socket boards are very efficient at ray tracing with the commercially available apps. I saw a cinebench score of 27 for a 48 core 12K USD system. I think a high end dual socket Intel system will score about the same. I’ve obviously never done anything with a quad socket system though, so I’m just passing along things I’ve read on the internet.
-AJ