I made my own PC for deep learning. (Windows10, Ubuntu18.04 dual boot)

I made my own PC for the first time in 4 years

** 20201123 Some titles and contents have been changed to make them easier to understand. ** **

This is Qiita's first post, it may be difficult to read, but if it helps something ... My home PC is a core i7-6600k + GTX1060 (6Gb) that I made by myself in 2016, and it has become a little insufficient. I was also interested in switching to the Ampere generation GPU and Zen 3 generation RYZEN, so I decided to make my own PC.

The specifications of the old PC CPU : Corei7-6600k Memory : 16GB(DDR4 2666MHz) MB : ASUS H170 Pro Gaming GPU : GTX1060(6GB) System drive 1: Windows10 Pro NVMe SSD PCIGen3 512GB System drive 2: Ubuntu 18.04 SATA SSD 1TB Data drive: 6TB HDD

The specifications of the new PC CPU : Ryzen9 5950x Memory : 64GB(DDR4 3200MHz) MB : ASUS ROG STRIX B550-F GAMING (WI-FI) GPU: GTX1060 (6GB) (continued to use) System drive 1: Windows10 Pro, NVMe SSD PCIGen4, WesternDegital SN850 1TB System drive 2: Ubuntu 18.04, SATA SSD 1TB (continued to use) Data drive: WesternDegital SN550 2TB

New PC configuration

CPU The CPU of interest this year (2020) will be Ryzen of Zen3. It is said that the performance has been improved by about 19% (IPC comparison *) compared to the previous generation, so there is no reason not to introduce this. In addition, I also do machine learning at home, so I feel that multiple cores are desirable. Initially I was thinking of Ryzen 9 5900x (12C24T), but I changed it to Ryzen 9 5950x (16C32T) due to lack of stock. I bought it at a computer studio in Akihabara about a week after it went on sale.

GPU I want about 12-16GB of GPU memory because it is used for deep learning. The Ampere generation RTX30 series has a dramatically increased number of CUDA cores than the previous generation, and learning is likely to be faster. However, only RTX3090 meets the GPU memory requirements, which is over budget. So I decided to wait for RTX3080Ti (GPU memory 20GB?), Which is rumored to appear in Q1 2021, and decided to use GTX1060 for the time being.

Motherboard

As of November 2020, Zen3 compatible chipsets are either X570 or B550. The X570 is often equipped with a chipset fan to cool the chipset, which is troublesome, troublesome, and noisy. The B550 has low heat generation, is fanless about the chipset, and can use PCIe4, so it's enough if you don't overclock. The manufacturer will be ASUS with a sense of stability. This time I decided on ROG STRIX B550-F GAMING (WI-FI).

memory

I haven't had any trouble with memory in my own experience, but I still want to choose a manufacturer that is stable even if it is expensive. CENTURY MICRO is the safest manufacturer made in Japan. This time, the low-priced version of CENTURY ESSENTIAL DDR4-3200 64GB (16GBx4) released by the company was used. This product has been confirmed to work with DDR4-3600 by XMP, and you can easily overclock the memory. Around this time, it's much easier than it was 20 years ago, and it feels like a world away.

Storage area

Since PCIe4 can be used with much effort, we will introduce the latest PCIe4 SSD. This time, I used Western Digital's SN850 (M2 slot, NVMe, 1TB) as the system drive for Windows 10. I'm using the old SATA SSD for Ubuntu. The drive for storing data is Western Digital's SN550 (M2 slot, NVMe, 2TB). If you decide to store data that you do not use often on the NAS, 2TB of data drive is enough. Well, the PCIe4 SN850 has only the latest SSD, and the speed is ridiculously fast. CrystalDiskMark.PNG

Power supply

I was thinking of a GPU of about RTX3080 in the future, so a certain amount of power capacity is required. I chose the 850W one this time as a Corsair company that I have been using for a long time and have not had any problems. There is a guideline for power capacity that can be calculated on the web (https://www.dospara.co.jp/5info/cts_str_power_calculation_main).

CPU cooler

By the way, simple water cooling is recommended for Ryzen 9, but this time I tried air cooling. I chose Noctua's NH-U12A, which is also famous for those who make it. I actually pumped it up and conducted a stress test with OCCT, but even at MAX, the CPU temperature is around 70 degrees, and at idle it is around 40 degrees (room temperature 20 degrees), so it can be operated without problems. However, upgrading the GPU may be a bit tough. .. .. OCCT.png

Install Windows 10 and Ubuntu 18.04

I didn't screen capture it during installation, so I can't go into too much detail. Windows10 Download the tool for creating installation media (USB) from the following Microsoft site. This will allow you to install the latest Windows 10. https://support.microsoft.com/ja-jp/windows/windows-%E7%94%A8%E3%81%AE%E3%82%A4%E3%83%B3%E3%82%B9%E3%83%88%E3%83%BC%E3%83%AB-%E3%83%A1%E3%83%87%E3%82%A3%E3%82%A2%E3%82%92%E4%BD%9C%E6%88%90%E3%81%99%E3%82%8B-99a58364-8c02-206f-aa6f-40c3b507420d After pumping up the PC, turn on the power and press F8 on the keyboard to select the Boot drive (F8 for ASUS MB). Therefore, if you specify USB as the installation media, the installation will start. With this configuration, the installation will be completed in a few minutes.

To protect SSD

If there is a lot of main memory, hibernation will dramatically increase the amount of writing to the SSD. So don't forget to turn off hibernation. You can run the command prompt with administrator privileges and turn it off only with powercfg / h off.

Install Ubuntu 18.04

This also creates a USB installation USB. First, download the ISO image of Ubuntu 18.04 Japanese version from the following site. https://www.ubuntulinux.jp/products/JA-Localized/download

This time I used Rufus to burn the ISO to USB. You can download it from the following. https://rufus.ie/en_IE.html

Prepare the work so far on a Windows PC, and when restarting, select the Ubuntu installation USB in the boot drive selection with F8. The point from here is important, but I'm sorry I didn't capture it. When choosing the SSD to install to, be careful not to accidentally select the SSD with Windows installed. By the way, NVMe SSD is displayed as / dev / nvme0n1 or / dev / nvme1n1 in ubuntu (note that the displayed name is different for each environment). The installation itself will be completed in a short time. After installation, reboot, enter the UEFI screen, edit the boot order, and bring the SSD with Ubuntu installed to number one. This will cause the Ubuntu bootloader to boot after powering on, allowing you to boot either Ubuntu or Windows.

For the work itself, the following site is very easy to understand. Thank you very much. https://www.pc-koubou.jp/magazine/35542

About the matter that does not recognize INTEL I225-V

This MB has a 2.5GBE INTEL NIC (I225-V), but it is not recognized by Ubuntu 18.04. In my case, it was recognized that I updated the kernel version to the latest 5.4.0. This I225-V is unstable even on windows (NIC is suddenly not recognized), so I think it is better to prepare a USB-LAN conversion cable etc.

in conclusion

I made my own for the first time in 4 years, but I feel that it is getting easier year by year. That's exactly what I made for the first time with 1GHz Athlon 20 years ago, but I miss the fact that I attached Firebird while being scared of the fear of lacking a core. I didn't need a power supply in one shot, and I struggled to learn a lot, but this time it started up so easily. In the future, I would like to update the environment settings for deep learning.

Recommended Posts

I made my own PC for deep learning. (Windows10, Ubuntu18.04 dual boot)
[Paiza] I made my own utility for answering questions [Java]
I updated my own blackjack made with Ruby for my portfolio
Creating a dual boot environment for Ubuntu Server 20.04.1 LTS and Windows 10
Fastest PC setup for deep learning from scratch
I made my own Instagram automatic like tool [Ruby]
Wifi adapter not recognized on Ubuntu with Windows dual boot
Build Ubuntu 18.04.5 with dual boot
Spring Boot for annotation learning