官术网_书友最值得收藏!

Querying your GPU

Before we begin to program our GPU, we should really know something about its technical capacities and limits. We can determine this by doing what is known as a GPU query. A GPU query is a very basic operation that will tell us the specific technical details of our GPU, such as available GPU memory and core count. NVIDIA includes a command-line example written in pure CUDA-C called deviceQuery in the samples directory (for both Windows and Linux) that we can run to perform this operation. Let's take a look at the output that is produced on the author's Windows 10 laptop (which is a Microsoft Surface Book 2 with a GTX 1050 GPU):

Let's look at some of the essentials of all of the technical information displayed here. First, we see that there is only one GPU installed, Device 0—it is possible that a host computer has multiple GPUs and makes use of them, so CUDA will designate each GPU device an individual number. There are some cases where we may have to be specific about the device number, so it is always good to know. We can also see the specific type of device that we have (here, GTX 1050), and which CUDA version we are using. There are two more things we will take note of for now: the total number of cores (here, 640), and the total amount of global memory on the device (in this case, 2,048 megabytes, that is, 2 gigabytes). 

While you can see many other technical details from deviceQuery, the core count and amount of memory are usually the first two things your eyes should zero in on the first time you run this on a new GPU, since they can give you the most immediate idea of the capacity of your new device.
主站蜘蛛池模板: 开远市| 印江| 汨罗市| 莱芜市| 安仁县| 平定县| 土默特右旗| 衡山县| 保康县| 邵阳市| 高台县| 萍乡市| 贵州省| 沿河| 塔河县| 南乐县| 大埔区| 霍邱县| 枝江市| 洛南县| 防城港市| 金坛市| 新密市| 原阳县| 利辛县| 义马市| 苏州市| 凉城县| 阿克苏市| 海林市| 长葛市| 和平区| 香格里拉县| 乐平市| 尉犁县| 宝坻区| 彩票| 霍州市| 分宜县| 武义县| 泾川县|