Originally Posted by CTRLurself
The 580 is faster than all new graphics cards when it comes to drafting and CAD work. They deliberately crippled new cards double-precision performance so they wouldn't compete with their Quadro and Tesla cards as much. The 580 is nearly as fast as the Nvidia Titan when it comes to 3D modeling work.
Kepler based cards that do not
have "crippled" double-precision performance are the Titan, K6000, and tesla cards.
Everything else has had double precision "crippled." In other words, a quadro K5000 doesn't have any more double precision performance per cycle than it's "geforce" counterpart the GTX680. In other-words, this "premise" that they did this "crippling" to reduce the competition of GeForce and Quadro cards is entirely misguided. Most of the new quadro lineup shares the same deficiency.
The important distinction between quadro and geforce, is drivers that support proprietary APIs developed for specific applications. (AutoCad doesn't use a proprietary API)
Furthermore, AutoCad performance doesn't scale up particularly well with high end GPUs ANYWAY:
Notice the very high end GPUs not doing much better than low end GPUs (nowhere near the on-paper differences in compute capability between the low and high end options)
While you're there reading those articles, note that they point out that the only reason to use a workstation card for Autocad is if your work absolutely must take place within a "certified hardware" list. Otherwise, regular GeForce cards will work fine.
Look here: http://www.tomshardware.com/reviews/geforce-gtx-titan-opencl-cuda-workstation,3474-3.html
Note: AutoCad uses the directX API for 3D interactivity acceleration (an API which is supported through geforce drivers and has no need of double precision capabilities)
I wouldn't be recommending the GTX580 for this. Pointless legacy hardware implementation that will just use a ton of power unnecessarily. The advantages of the GTX580 over modern cards just aren't there for AutoCad.
For AutoCAD and PhotoShop I do not
recommend spending more than ~$250 on a GeForce GPU as the useful performance scaling will have already started to taper off well below that point anyway. If you must be on a "certified" GPU, then I would recommend the K2000 for ~$400 as the best fit for your budget. The GTX760 is a lot more GPU for $250 than the K2000 is, but sometimes certifications are more important than higher FPS rotations.
Your build should be focused on the best possible CPU performance and plenty of RAM, possibly
on a workstation class motherboard (C22X series, or soon-to-be-released haswell-E) with ECC memory and a workstation GPU. (if you need to be in an environment that AudoDesk and Adobe will fully "support" if you need technical assistance)
If you want to go the "enterprise/workstation" route that would be fully "certified/qualified", the build will look about like this:
CPU: E3-1240V3 :$260
MOBO: Supermicro X10SAE: $200
RAM: 2x8GB or 4x8GB 1600MT/s 1.5V or 1.35V ECC from Samsung/Kingston/Hynix etc... ~$160-320
GPU: K2000: $430
SSD: Toshiba/Samsung/Intel 256GB: ~$150+
Storage: 1TB FZEX: $85
Case: Fractal Design or Corsair mid/full towers ~$100-150.
PSU: 400-650W depending on long term plans for peripheral connection requirements. (it's not the wattage that is important here, as this machine will only use like ~200W, it's the connectivity for drives and things you might add later on). Something like a Seasonic G series IMO: $70-100
That's right around the $1500 mark for the core of the machine.
If you don't need/want the qualified GPU, switch the above build to the GTX760 to save about $200 (apply towards more RAM IMO). I recommend the EVGA ACX and GeForce Windforce models but most any GTX760 is pretty much the same in use unless overclocked. The GTX750Ti for ~$160 should also be on your list of "budget" alternatives, as it offers better GPGPU performance than a GTX770 under many workloads, and as such, may someday outgrow the GTX760 in productivity applications as they are better optimized to leverage the GPU.