As GPUs (or graphics cards) evolve quickly, it can be challenging for a Fortem partner or customer to pick the right model within their budget. This is our CTO’s personal recommendation as of April 2017.
Any NVIDIA Geforce GTX 9×0 (e.g. 970) or 10×0 (e.g. 1070) series will work great as long as it has at least 3GB video memory (VRAM). While they are primarily designed for video games, they work great with our software. The GTX 1070 and 1080 offer the best bang-for-the-buck and should offer good performance for years to come. I have a GTX 1070 at home, and I’m very pleased with it. I’m experimenting with Deep Learning during my spare time and the extra VRAM (8GB+) makes a huge difference there.
As for manufacturers, EVGA, MSI, ASUS are all good brands.
We cannot recommend the NVIDIA Quadro cards. These cards are designed for Workstation applications that use old-style OpenGL. Our software is optimized to act like a video game, so Quadro cards are overkill and on rare occasions, cause trouble. They are very pricy, they tend to be one generation behind their consumer cards, and they have subtle differences in their OpenGL drivers that sometimes cause problems for us. 3 years back, one of our customers was pressured into buying a Quadro card for $1500 instead of the $250K GeForce card we had recommended. They ran into a OpenGL driver issue we couldn’t reproduce in our office. In the end, they had to buy an extra Quadro card, ship it to us, and we had to spend 3 days diagnosing and working around the issue. Please don’t make the same mistake.
We also cannot recommend AMD GPUs and Intel Integrated graphics at this stage. While these cards and drivers generally work with our software, we don’t test on them as much, and these companies haven’t made as big an investment as NVIDIA in new technologies like Deep Learning, that we foresee having a big impact on our future products.
I hope this helps, don’t hesitate if you’d like us to double-check the card you plan to pick.