I am about to buy and build a PC to practice Deep Learning. No gaming here ;-)
I plan to start with one GTX 1080Ti (I will probably add a Titan Xp later, maybe when available with Volta architecture)
I am a bit confused as there are plenty variations of the Nvidia GTX1080Ti design, by numerous brands, with more or less overclocking, all aimed at the gamers. (And it looks the “Founders edition” GPU versions are a bit more difficult to find/be available). So it’s quite hard to make a choice.
All the more, usually these brands (for example EVGA) only provide drivers/tweaking tools for Windows (nothing for Linux…)
–> Are there risks of incompatibility issues of using the drivers provided by Nvidia, with these modified/customized/overclocked cards of these manufacturers ? How to choose wisely ?
To be clear, I am not specially looking for an “overclocked” version. I would prefer a model with a better cooling, without going crazy with the price, and be sure it’s quality made and ok to run with Linux. After all, model training will last hours/days, so the cooling will be much more important to maintain a good performance than seeking for a few more Mhz
Any successfull return of experience to share ?
For instance, here are some of the models I am considering to buy :
- CORSAIR (and MSI) HYDRO GFX GTX 1080 TI
- EVGA GTX 1080Ti FTW3 HYBRID, or
- EVGA GTX 1080Ti FTW3 GAMING, or
- or simply the GTX1080Ti Founder edition from Nvidia online shop to avoid configuration hassles
Thanks in advance for your feedback !