TL;DR - Runs fine on 240 W PSU; does NOT require an external power connecter (gets all power from PCIe x16 slot); averages 70 °C with 60 W power draw when training TensorFlow models; trains those models about fifteen times faster than i5 CPU.
I'd imagine most people are checking out this card to improve [1] their gaming experience or maybe even to mine cryptocurrency. I, however, can't tell you whether or not this GPU is good for those applications. I've never really done any mining, and I generally don't play games that are graphically intensive.
What I can tell you is that this card is great for machine learning. I do neural network related research using TensorFlow, and this card was a fantastic and inexpensive [2] investment. Assuming the model I'm fitting is complex enough to utilize the whole GPU, the training time is nearly fifteen times shorter than with my i5-2500 CPU alone.
This card performs surprisingly well in my very limited setup. I'm stuck with a hand-me-down HP Compaq 6200 SFF (Slim Form Factor) PC, meaning I can only fit half height graphics cards. My PC also has a weak, 240 W PSU that has no spare power connections available. As a result, my graphics card has to get all its power from the PCIe x16 slot. This GTX 1050 Ti fulfills both these requirements without any fuss what-so-ever. The hottest I've seen the card get during machine learning is 70 °C, and the power draw seems to average 60 W (it sits around 35 W when idle).
While Nvidia and Linux generally have not gotten along too well in the past, I haven't had much trouble using this card with Arch Linux (and Windows 7/10). Regarding specifically GNU/Linux, the open source Nouveau driver will probably be enough for most applications. Despite that, I'm fairly certain the proprietary, but still free Nvidia driver is required if you want to use this card for machine learning on Linux. Just be sure to get it from your package manager rather than the Nvidia website so that you can easily upgrade it in the future.
*1 - (03/14/18): Corrected typo.
*2 - (03/14/18): I bought this card on 09/21/17 for $160. I would no longer consider it inexpensive at its current price of $250.
| Memory Speed | 7000 MHz |
|---|---|
| Graphics Coprocessor | NVIDIA GeForce GTX 1050 |
| Chipset Brand | NVIDIA |
| Graphics Card Ram Size | 4 GB |
