Pro vs High End

[Xittenn]

I've built my rig up to workstation class and am finally ready to buy my top end graphics card to replace my HD5770. I was expecting to be doing a lot with programming/rendering and I had hoped to have accomplished much in terms of having my own personal platform to start developing algorithms while in school. Although my platform is pretty well roughed out it is nowhere near complete and I'm not sure if I can do school and program like I have been.

 

So now I'm left with a dilemma. I had planned to purchase a Tesla C2070 but this would be an extreme waste if I never used it within the next four years. Mid way I have looked at the FirePro v8800 and this is a fairly powerful card without sacrificing reliability. The new HD7970's are out and the power is >Tesla C2070, but obviously is no where near reliable. I want to have a workstation to do work on related to my studies, but I see that this will probably not matter until grad school. And I think I could always suss out a solver on my workstation and then run it on a time shared server or something.

 

I would substitute for a microscope but anything worth my time makes the video card look like nothing more than a new purse. Just curious if anyone had any other thoughts or suggestions on the matter?

[Xittenn]

Is there any reason why I should not go with the AMD Radeon™ HD 7970 Graphics? I mean it's a popping 970GFlops DP, which I think satisfies my needs and then some. Just curious is all . . . . .

[lightburst]

People wouldn't usually need an actual workstation GPU unless you work at some [insert cliche profession].

 

Though it seems you really would only use the GPU for legitimate stuff, not like the most of us would use it, so why not get the workstation GPU. Don't let the raw computation power fool you as mainstream cards' drivers were made to process different kind of data than their workstation counterparts.

 

Of course, programming that would require something like CUDA would need as much raw power as it needs. Ultimately, you shouldn't opt to buy a new card unless there is a demand for it.

[Xittenn]

I will be working with OpenFOAM or OpenCL, and my own variant of CFD/CE with my own pre and post processing meshing suite. I don't know much about the drivers, because it isn't something I'm all that interested in, and so I'm not clear on what changes are made between the two that affect performance. I assume much of this has to do with vector manipulation and optimizing the usage of the accumulators.

 

 

I guess what I'm saying is, are the drivers tailored to the application, or the specific softwares themselves? Is it that the driver is tailored to CFE/CAD/CAE, or is it that the driver is tailored to Solid Works, AutoCAD, CATIA and so on . . . .

[Xittenn]

Link to post on Stackoverflow

[Schrödinger's hat]

The line between not enough power and far more than you will ever use for any given problem is so thin that I would think it is hardly worth spending much money on performance.

Especially if you are just dealing with toy projects for learning, you will just be able to turn down the number of steps/elements in your simulation.

Save the larger investments for when you actually know enough that you are using your workstation for real work (or are you at this stage now?).

[lightburst]

http://www.viddler.com/v/9cb51810

 

I think that answers the last part. I do still think that most people wouldn't need an actual workstation card unless you would consider your work to be so critical as to justify the price tag, as most of it is from making sure the card is stable and with 24/7 support.

[khaled]

Xitenn, Graphical processors are not tailored for any software .. I think you should pay more focus on the design of your graphics processing model,

 

Linerize your algorithms and optimize your methods .. that's how these software work like if they have no load on the GPU

 

One more thing, you don't have to use wires to build a computer .. meaning, make use of the updates on tech and libraries

Edited April 1, 2012 by khaled

[Xittenn]

Someone answered the question on StackOverflow and they hit on the key point for me which is the drivers are optimized for big name software suites, and not for the application purpose itself. I am a registered business and I do have a long history of working with applications like AutoCAD, SolidWorks, XSI and so on . . . That said my focus is no longer in that direction and as noted by the respondent on StackOverflow I will most benefit from the HD7970. I think this card will fulfill my needs while I am completing my degree in Math/Comp-Sci and will be great for gaming in that time as well. I will be sure to blog about the outcome of this as I start to progress with my activities, in particular as I start to take Duality--my engine--online. This should be happening fairly quickly as I now have access to ATL and a variety of other tools, and will be procuring Intel's compiler suite shortly (fortran). I hope to have videos within the next year!

[Xittenn]

I just want to point out that it was in fact not me that gave khaled the -1 point, just to be clear!

[khaled]

I just want to point out that it was in fact not me that gave khaled the -1 point, just to be clear!

 

GPUs are not tailored for some software, they are enhanced by optimizing some issues that occur on some well-known software .. but they're not tailored for any of them

 

I've seen absurdity from semantic ambiguity to use specific jargon for everything .. I no longer care if they -1 me, it won't break symmetry or make the wrong come true