Welcome to Fractal Forums

http://www.nvidia.com/object/product-quadro-4000-us.html (http://www.nvidia.com/object/product-quadro-4000-us.html)

adoring those? ;)
lol, it is always disappointing to see the limits of those cards. you will just have realtime interaction up to a certain iteration count :(
im off of nvidia, but wold love such card, especially when it could handle double precisipn!
thanks for notifiication

btw, those cards are basically the same hardware as consumer cards. the price difference is due to the optimized open gl drivers. normally they are only faster, when using one of a hand full of professional cad software (catia & co).
i doubt that they benefit hand coded gpgpu programs.

The Quadro cards are not more powerful than consumer GeForce cards (a good number to look for is the number of GPU cores). Historically, the Quadros provides quad buffered OpenGL support, which was (and is) needed for most professional 3D shutter glasses.

A couple of weeks ago, I got a Tesla C2075 at work (http://www.nvidia.com/object/workstation-solutions-tesla.html) which sure is a nice card, but it doesn't run Fragmentarium faster than my GeForce 570 (in single precision). The biggest advantage is 4x faster double precision (doubles running half speed of single precision), and 6GB error correcting ram.

I have converted a few simple Fragmentarium scripts to double precision, but annoyingly many things are not supported for doubles in GLSL (e.g. trigonometric functions, passing varying doubles between shaders). 

The latest Kepler based nVidia Tesla cards are said to have some 3.6 TFlops of single precision power. These are dual GPU cards.

The double precision accelerated Tesla boards featuring Kepler GPUs are to be available a bit later.

GeForce GTX 690, with 3072 CUDA cores. That's a supercomputer!

At the moment, the fastest dual precision hardware off-the-shelf is AMD's 79xx line (DP is quarter speed of single precision, and even the consumer models are unlocked to that maximum speed). But the two remaining GPU makers keep leapfrogging each other anyway.

In terms of DP FLOPs per watt or per money, I think IBM's special purpose "Blue Gene/Q" processors would win right now. But these machines are not exactly mainstream.

yep, I'm running two 7970s @ 1 GHz. Need work on some liquid cooling setup to keep them happy going full throttle but the flops are there.
That's 2 TFlops double precision, 8 TFlops single precision for $900.

The problem is that those cores have far less memory (registers+cache) in the Kepler arch than with Fermi. So yes, you get better power efficiency for small/simple kernels, but for complex GPGPU applications Fermi is still better.