Using a Tesla M2090 for CAD

[Munchskull]

Originally Posted by Knufire

Nah, don't bother. Not worth the hassle or power draw. That card was Fermi based, which is a design several (four?) iterations previous to your current card. Fermi was also the last design Nvidia pushed out before realizing that they needed to care about power efficiency, so those cards were power and heat monsters.

What if power and heat were not and issue. Would it add quadro level 10-bit performance?"

[Knufire]

What are you looking to accelerate specifically? Hunting down information on exactly what is GPU accelerated and the performance gains from going to a workstation card in Solidworks has been a PITA for me.

The most obvious difference for me having a "certified" (read: workstation) GPU was enabling RealView Graphics, which I found horribly annoying and turned off anyway.

[Jared]

You won't get a benefit. The GTX 1060 is way more powerful than you would likely need for any FRC related CAD, but it's not a workstation card.

If I were you, unless I had a specific purpose for the M2090 I'd sell it and buy a quadro/firepro graphics card for SolidWorks. SolidWorks and most other CAD software uses OpenGL and video games almost always use DirectX. OpenGL stuff works way better on a workstation graphics card - you do a ton with even an entry level workstation card. SolidWorks is much more stable on a workstation card than a gaming card.

Something like an M2090 is really hard to take advantage of unless you really know what you're doing. MATLAB has pretty good support for doing lots of matrix operations with CUDA cores/xeon phi hardware, but something like SolidWorks can't really take advantage of the technology.

[tragic]

As others have stated your GTX 1060 will crush that Tesla in compute performance. I have a bunch of fermi series card lying around but I would never use them for anything except for diagnostic purposes. (testing to see if a different GPU is pooched.)

To really put this in perspective that Tesla has 512 CUDA cores while your GTX 1060 has 1280 CUDA cores. Not to mention increases in instructions per clock, much higher memory bandwidth (this huge to increasing performance as the GPU is often starved of data for large amounts of time), driver run time overhead reduction among other things all while using less than half the power and emitting less heat. Your 1060 is probably well over 5 times the performance of that Tesla.

[TAlholm]

It would be difficult to justify using the M2090. It is based on the Fermi architecture, which bears the nickname "Thermi." In Visualize or simulations, you MIGHT notice a performance increase, if even that. Then again, for FRC purposes, you will not need it.