Ok, I should know about this stuff since I do 3ds/Revit/AutoCAD every single day, but alas, I can't wrap my head around why we need workstation cards, they are so ridiculously slow, when compared to a "comparable" desktop card in price.

Here's what I do know.
Workstation cards are the same hardware as Desktop Cards, except they have a different BIOS that limits them to only use workstation drivers. Those workstation drivers are optimized for stability and for 3D applications that deal with a lot of textures and triangulations. Specifically they reduce visual aliasing artifacts and improve visual performance with overlapping planes, fragments, and vertices. Additionally they also carry the best customer support with them.

A list of what the workstation cards are tuned for: (to be honest I'm not sure, but I think some of these are supported by the desktop cards too, just not sure which)

Quote :

"* Unlimited fragment instruction * Unlimited vertex instruction * 3D volumetric texture support * 12 pixels per clock rendering engine * Hardware accelerated antialiased points and lines * Hardware OpenGL overlay planes * Hardware accelerated two-sided lighting * Hardware accelerated clipping planes * 3rd generation occlusion culling * 16 textures per pixel in fragment programs * Window ID clipping functionality * Hardware accelerated line stippling * Long fragment programs (unlimited instructions) * Long vertex programs (unlimited instructions) * Looping and subroutines (up to 256 loops per vertex program) * Dynamic flow control * Conditional execution * 16x FSAA dramatically reduces visual aliasing artifacts or “jaggies” resulting in highly realistic scenes"

Here's a comparison of two equally priced cards ($180) and their specs:

Quote :

"Quadro FX570 PixelPipelines 12 Raster Operators 8 Texture Units 16 Memory Size 256MB Memory Interface 128-bit Memory Type GDDR2 Memory Bandwidth 12.8GB/sec Texture Fill Rate 3.68 Billion Triangles/sec 137 Million Features FULL DirectX 10 FULL OpenGL 2.1 NVIDIA Unified Architecture Shader Model 4.0 Rotated-Grid Full-Scene Antialiasing (RG FSAA) (+1) NVIDIA PureVideo Technology GeForce 9800 GTX(G92) Core clock 675MHz Stream Processors 128 Raster Operators 16 Texture Units 64 Memory Clock 2200MHz (effective) Memory Size 512MB Memory Interface 256-bit Memory Type GDDR3 Memory Bandwidth 70.4GB/sec Texture Fill Rate 43.2 Billion/sec Features FULL DirectX 10 FULL OpenGL 2.1 NVIDIA Unified Architecture Shader Model 4.0 NVIDIA PureVideo Technology"

To get the same level of performance hardware-wise in a workstation card, it costs nearly $800, that's FOUR TIMES (400%) the cost. To me it's just not worth it, for the "reduced jaggies" on my 3D model. Because it sure as hell ain't because of customer support.

I'm sure Noen has an answer for me.

[Edited on October 28, 2008 at 10:15 PM. Reason : .]

10/28/2008 10:12:08 PM

For a good comparison, the GTX260 has about 1/4 the performance of the Quadro CX, the exact same card. But the GTX260 is sufficient at 25fps on most applications. I would save $1800 and go with the GTX260 and get 1/4 of the performance if I was a business.

The only differences in the Quadro and FireGL cards and their desktop variants are the bios on the video card and the drivers for them. Nvidia has been hardware locking their desktop variants from modding the bios since the 6800GT incident.

P.S. Certain ATI cards can be modded to be FireGL cards ....specifically the HD2900

[Edited on October 28, 2008 at 11:32 PM. Reason : :]

10/28/2008 11:27:09 PM

well i'm not talking animation or advanced 3d modeling here, i'm talking about 90% of the time when i'm using 3ds/revit/autocad, we also don't use anything over mid-range cards, and in nearly all the entry-level to mid-level workstation cards, the desktop variant is nearly 4 times faster.

i'm certainly aware of softmod'ing (in some cases hardmod'ing) but we are not going to do that in a business environment.

[Edited on October 28, 2008 at 11:56 PM. Reason : .]

10/28/2008 11:55:26 PM

I was under the impression that workstation cards were optimized for rendering a lot of shapes at once while consumer cards focused more on texture rendering.

10/29/2008 3:43:30 PM

they are but only because of the drivers, the fact is, my desktop card has so much more power and can do so many more calculations per second, that it still outperforms the workstation card on rendering... albeit when i pan around it may not be as smooth, but the response time is definitely quicker, A LOT quicker, that's what pisses me off, you pay FOUR times the cost for just the driver/bios difference.

[Edited on October 29, 2008 at 4:01 PM. Reason : .]

10/29/2008 3:58:00 PM

It is more than the bios and drivers.

Quadro certification (and FireGL) is rough. Think of it in the same way you think of ECC memory vs. non-ECC, of Xeon processors versus Pentiums.

The reason GL cards have such slower specs for the PRICE (you can get a super baller GL card too), is that they guarantee the visual accuracy of what is being drawn on screen. You lose speed for precision. The cards themselves are essentially binned. Each GPU is fault free, the components are higher quality (generally), and many times the PCB layout is actually different for GL cards compared to their consumer counterparts to allow for more error correction, detection and avoidance.

-------------

That said, outside of high end CGI work, and maaybe post-processing for video, I just can't justify the price difference for a pro GL card. 10 years ago the market was entirely a different beast, but today there is too much parity in functionality for the disparity in price.

With the power of CPU's and GPU's now, you can do almost everything on a consumer card with settings tweaks to get VERY near the quality of a pro card.

Specific applications where you still may need a GL card:

-Surface modeling for manufacturing
-Solid modeling for manufacturing
-High fidelity, cinematic special effects
-Academic/Research data visualizations
-Realtime rendering systems

10/29/2008 5:05:32 PM

thanks for that.

10/29/2008 5:28:56 PM