Autocad support cuda (tesla cards)

[Anonymous]

Any help would be greatly appreciated, I have researched endlessly to find out if autocad supports cuda GPU acceleration aka Tesla Cards. I can't find the answer anywhere. If anyone knows could you please post the answer as My rendering animation times are outrageous!! and I have a nice new computer with core i7 and 12GB ram. My .dwgs are close to 100MB.

And I am pulling like 20+ hour render times for 30 sec. videos! It sure would be nice to cut that down to 20 min render time l.o.l. any help appreciated, thanks.

[pendean]

Hardware Discussion area:
http://forums.autodesk.com/t5/Hardware/bd-p/116

[dgorsman]

Rendering stills or animation is primarily a processor task, not a video card one.  I would think that if AutoCAD did support those monsters the information would be easy to find but again, that would only help for the animation playback.  Think about render farms - they have banks upon banks of processors with no video cards whatsoever...

[Anonymous]

cuda capable cards use excess gpu power as extra cpu power. At this point in time I don't know of any commercial software that is taking advantage of this capability. Programming to utilize is at least a little more complex than a rendering farm since the chip architecture is not even remotely similar to the cpu's architecture.

[dgorsman]

A quick read on both CUDA and Tesla cards make no mention of this, only that CUDA technology is a parallel processing architecture and Tesla cards are for massively parallel graphics applications.  Maybe I missed something? 

[Anonymous]

Go to Nvidia's website. They emphasize the use of their cuda chips in lieu of conventional cpu's for various types of analysis.

See specifically:

http://www.nvidia.com/object/cuda_finance.html

[dgorsman]

Ah, I think I misread you earlier.  Like most new video hardware the computer system doesn't borrow "unused" power from the GPU to support the CPU.  By providing higher-level functions in the GPU it lets the CPU send more raw data, passing some of the processing which used to be necessary in the CPU, thus freeing the CPU to do more.  Like sending a model of a car to the GPU and saying "Render this for display" rather than the CPU figuring out what needs to be displayed then sending the almost finished data to the GPU.

[Anonymous]

No, the webpage says it is doing the analysis there, not just screen rendering. Ever since the advent of 3d cards any 3d capable card has done what you specify, to some extent or another. There were multiple fields for which they claim this to be an applicable advantage. I chose to send you to the finance field webpage since that is where it was most obviously not just screen renders. What is really happening (according to how I interpret Nvidia) is the cpu is sending raw data to the gpu, just like it was screen data, but the gpu is responding with analysis results, which the cpu then uses as though it came from a different cpu core.

[Anonymous]

First off, I want to say thanks for all the input from everyone but there does not seem to be a plain and clear answer to whether or not the tesla card would be able to speed up the render by offsetting the load from the CPU (or however you want to word it, I think everyone knows what I am getting at here). I found a really good deal on a tesla c1060 card and bought it, It is scheduled to arrive Friday, as long as everything goes smoothly, We should have a definate answer by Friday night or Saturday. I really appreciate everyones help and input on this but in the end, there is only one way to really know... Trial and error. l.o.l. If it does work and speeds up renders by the approx. 250 times Nvidia claims.... well..... just think of the possibilities.

                           thanks,  Justinhhhfan

[Anonymous]

Got my Tesla today, It does work but the exceleration rate of 250x was highly generous. I already have a top notch computer and this increased the speed probably 10x. Some programs are unaffected, you have to go into nvidia control panel and enable CUDA GPU hardware exceleration and set the program to utilize the Tesla and not just the regular GPU. Some Graphics benchmarks are tremendously improved but autocads benefit has been minimal so far. I think the 4GB of onboard memory is what is helping it, I will try to tweek it and see where I get. Thanks again.

justinhhhfan

[dgorsman]

That analysis being done in the GPU is for the display, not for further use by the computer.  So the CPU is sending mostly raw data and a function to perform on that data, plus various directives on how to format the result on screen.  Its a one-way process, that processed data doesn't go back to the CPU.  They get the performance by unloading that data analysis function from the CPU to the GPU.

[Anonymous]

Not according to Nvidia. see http://www.nvidia.com/object/GPU_Computing.html

They claim they move the computation to the gpu for floating point work at which it is more efficient than the cpu, then bring the results back to the cpu. GPU's can now be programmed with C++, and they have been optimized for double precision floating point whereas cpu's are optimized for branch instructions and integer functions with floating-point math processes being a secondary add-on.

Prior to this new CUDA family, the description of the process matches the one you describe. Off load video processing to the gpu but leave computation on the cpu. Unless you were using the computer for 3d screen processing, enhanced gpu's often were often slower than older simpler gpu's at drawing 2 dimensional screens. (see Tom's Hardware for a review of this problem.) With CUDA you can program the system to take advantage of the increasingly powerful gpu to take over functions that it performs extremely well and gain back some speed when the chip isn't bogged down with generating screens. However, autocad barely uses a second cpu core, so expecting it to be programmed to use a gpu core with a vastly different architecture is something that we probably won't see happening for quite a few years.

[sam_m]

As far as I know CUDA is only supported on 2 Autodesk products at the moment - Ansys and Moldflow.  Equally believe it is only supported in their stand-alone packages (ie FEA within Inventor is not supported).  I have not seen any comment about Inventor (my cad package) or AutoCAD (and it's varients) taking advantage of Cuda (yet).

The initial report of Ansys and Moldflow suggested an aprox 2 fold increase in performance - significant, but not the massive change that was originally touted by Nvidia.  But, that was initial tests, so could get better with time as programmers understand how to take advantage of it better.

Due to some Autodesk software using it already I'm hoping that it ripples through the product line and we can all benefit, but could be a few releases...  But, apart from heavy parallel mathematical tasts like rendering and FEA I can't imagine it being used for the core use of Autocad or Inventor - at the end of the day, these products are still single-threaded and do not take advantage of multiple cpu-cores, until they can be implemented into the workflow then how can they do the same with piggybacking a GPU?

On a side note, I believe DirectX 11 also has some GPU co-processing features (basically Cuda but will also work with ATI cards) - so if/when Dx11 is implemented then we might get some benefit anyway...  But, as above - if it's problematic implementing a multiple-core cpu then how can then start to play with the gpu too...

Hope that's helped