A computer was set up for the laboratory yesterday.
Accessories
BIOS
The heat sink limits the height of the memory module
Configuration
- SSD: Samsung 960EVO 250GB
- HDD: Seagate New Barracuda 4TB
- CPU：i7 8700k
- CPU fan: big frost tower (in fact, the power of the CPU does not need such a large radiator, it is a little bit of the memory, see the picture above)
- Motherboard: ASUS PRIME Z370-A
- Chassis: Xianma Houdao G1
- Power supply: Great Wall Dragon 1000D
- Memory: Kingston 16GB DDR4 2400 * 2
- Graphics: Zotac GTX1080Ti-11GD5X * 2
Install
There is nothing to pay attention to when installing the machine. It is enough to install normally without water cooling.
For Xiaobai, I give the following order to help install the machine:
- Power supply plug-in cable, first calculate the number, plug all the required cables into the power supply, take my machine as an example: power supply of the motherboard; CPU power supply; two graphics cards, a total of two power supplies (8 + 6pin interface on the graphics card , The power supply is 8pin to 8 + 8pin, just plug it in;); the chassis fan has a power supply (usually connected to the power supply sata power supply) + a mechanical hard disk power supply, a total of two sata power supply lines.
- Power on the machine, pay attention to the power fan is generally washed down, next to the lower ventilation network of the chassis. In this way, the power supply will absorb air to dissipate heat.
- On the base of the CPU cooling fan, different fan different installation methods, you can check the video tutorial.
- Insert sata line, need to insert a few.
- Put on the CPU, just clip it, pay attention to the positive and negative (there is a groove, you can't put it in the opposite direction).
- On the motherboard, don't forget the baffle of the IO port.
- Hard drive with M2 interface (skip if not available), because generally the M2 interface will be hidden under a heat sink on the motherboard. Remember to read the motherboard manual.
- On the memory. Nothing to notice, if you want to group dual channels, remember to choose the slot .
- Put on the CPU cooler, apply silicone grease, apply evenly, just install it. Remember to be optimistic about whether the CPU cooler is ventilated or blown (arrows will be drawn on it). If it is blown, the heat sink is directed at the motherboard io interface area, otherwise the opposite is true.
- Insert the hard drive sata.
- Plug in the chassis panel control cable, audio cable, and various fan power.
- You can plug in all the power except the graphics card.
- So far, you can try it after booting, because the CPU generally has an integrated graphics card (except Xeon series, etc.), and you can now plug in a monitor after power-on, try it after booting. All right.
- Power off, install the video card, and power on.
system
- ubuntu 16.04
- tensorflow
- CUDA
- a.d.
There is nothing to say about this, the mirror copy is copied to the USB flash drive, and UEFI boots the USB flash drive to install the system.
Deep learning environment, just follow the tutorial given to you by tensorflow and nvidia official website.
There are two points to note:
- The graphics card driver, it is best not to follow the online tutorial to install .sh version, it is very easy to go wrong. Update your apt package, then in the system settings, software and update tab, there are additional drivers, select the latest version of the n-card driver, the application will do.
- Pay attention to the version of CUDA that your tensorflow supports. Many times it is not the latest version of CUDA. For example, as of the time of this article, tensorflow1.7 only supports CUDA9.0. So don't install CUDA9.1 by hand. It is recommended to use deb installation (in case you install cuda9.1, deb installation can also save you). Similarly, cudnn should also be downloaded for CUDA9.0. Finally, environment variables are added to .bashrc.
carry out.
