It’s complicated.

The Quadro cards are designed for professional applications: CAD, 3-D visualization, 3D content creation, etc. The demands of those markets pretty much make developing modern GPUs for them impossible, since they’d have to sell for $10,000, $20,000, or more per card, just as they did in the 1990s, before Nvidia and ATI got into the professional GPU market.

So here’s their trick: they develop a new “general purpose” GPU… the ones you play games on, the ones I use for the 3D CAD I do at home, etc. They work out the bugs in the drivers for the chips, they may work around individual chip bugs, though not too often, as Nvidia simulates the crap out of their chips before making them, to avoid that sort of thing.

When they’re ready for a new professional graphics card, they’ll take one of the existing technologies for consumer cards and tweak it. Maybe. They’ll definitely deliver a card with more memory, so more memory address lines are enabled, but there aren’t necessarily any significant differences, though it’s usually an official different chip version. But they’ll also move to the professional drivers, based on very stable professional drivers from other cards, plus any GPU-specific code that’s been hashed out in the consumer market for a year or two.

And so, your professional-market GPU is usually a year or two behind the curve of the general-purpose GPU. But critical components of the drivers have had a year or two of debugging in a mass market environment. If you’re seeing problems with your GPU in a professional environment, that can be real money… no amount of Nvidia support is going to stop that if you do hit bugs.

So why buy one? Some of the 3D functions, particularly 64-bit floating point operations, are probably faster than the current consumer cards. Why? It’s primarily the drivers. Some operations are intentionally crippled in the consumer versions, so that they don’t perform as well as the professional models, but pretty much restricted to higher-end CAD stuff. This was demonstrated maybe a decade ago, by showing that, while a same-generation Quadro card was clobbering a GeForce card that should have been a good match in some CAD demos, recoding a few OpenGL operations using OpenCL or CUDA, the GeForce card ran just as fast. And more recently just by playing around with different OpenGL functions: GPU performance: Nvidia Quadro vs Nvidia GeForce.

Of course, in practical terms, you live with the code your CAD tools comes with, so if you’re not seeing the performance you need with your consumer GPU, it may be time to upgrade to a professional card. Also, because of the known stability of the drivers, some CAD software manufacturers only support the professional cards. If you’re having crashes or display bugs on a consumer card, they’ll blame Nvidia or AMD, not their own application. For serious professional work, the assurance of full support for that very expensive CAD application you’re buying every year (probably at prices comparable to a new Quadro GPU) is well worth the price of professional hardware.

Some professional cards have better 3D visualization functions. Most of them are designed to fill just a single slot, which is needed if you have the need to run three or four cards in a single PC. And of course, the extra memory you may find on such cards makes a difference with complex 3D work.