Programmer, ML, part time gamer PC

Hardware Build a PC
1

Hi folks, I want to build a new PC after several years and I don’t know much. Watched some videos and read some blogs. Can you give me some advice? Use case is programming (fast I/O, M.2 SSD maybe newest PCI x4), some entry ML (TensorCores), and some gaming…

First try:

AMD - Ryzen 9 3900X 3.8 GHz 12-Core Processor
Noctua - NH-U12S SE-AM4 CPU Cooler
MSI - MPG X570 GAMING EDGE WIFI ATX AM4 Motherboard
G.Skill - Ripjaws V 32 GB (2 x 16 GB) DDR4-3600 Memory
Samsung - 970 Evo 1 TB M.2-2280 Solid State Drive
MSI - GeForce RTX 2070 8 GB Video Card
NZXT - H500 ATX Mid Tower Case
Cooler Master - MWE Gold 750 W 80+ Gold Certified Fully Modular ATX Power Supply
Noctua - NF-F12 PWM 54.97 CFM 120 mm Fan
AOC - AGON AG273QCG 27.0" 2560x1440 165 Hz Monitor

Second try:

AMD - Ryzen 9 3900X 3.8 GHz 12-Core Processor
be quiet! - Pure Rock 51.7 CFM Sleeve Bearing CPU Cooler
MSI - MPG X570 GAMING PLUS ATX AM4 Motherboard
G.Skill - Ripjaws V 32 GB (2 x 16 GB) DDR4-3600 Memory
G.Skill - Trident Z Royal 32 GB (2 x 16 GB) DDR4-3600 Memory
Patriot - VPN100 1 TB M.2-2280 Solid State Drive
Gigabyte - GeForce RTX 2070 8 GB GAMING 8G Video Card
be quiet! - Silent Base 801 ATX Mid Tower Case
be quiet! - DARK POWER PRO 11 750 W 80+ Platinum Certified Semi-modular ATX Power Supply

2

Only thing I would really change would be your GPU choice. There is literally no reason to go for the RTX 2070 anymore, it’s now EOL and is being replaced by the RTX 2070 super for the same price.

So my recommendation is to go with either the new RTX 2070 Super, or the new Radeon 5700XT from AMD.

Also not sure about the memory clocks. I would wait a couple of weeks and see what the sweet spot for Ryzen 3xxx memory speeds are so you don’t spend additional money for no benefit.

1 Like
3

If you’re going to be using Linux, have an AMD GPU, much easier to deal with drivers. Other than that your choices are great. I’ll be building a Ryzen machine in a year or two as well.

1 Like
4

Update with the memory speed, DDR4-3600 turns out to be the sweet spot:

Boot Sequence Rundown: https://www.youtube.com/watch?v=ZwX7_As9q6s
TechPowerUp article: https://www.techpowerup.com/review/amd-zen-2-memory-performance-scaling-benchmark/2.html

1 Like
5

5700XT

For ML? Not so sure. I have an RX480, which IMO is the most flawless GPU experience on linux but unfortunately the gaming performance on AMD doesn’t justify using the ROCm stack for ML. If you want to use AMD for ML, some things you’ll need to think about:

  • Use Ubuntu 18.04. Running ROCm not in a container can get weird unless you’re on 4.15. You can use a docker image but it’s a pain in the ass to use all the time. As far as I’m aware windows is not an option.

  • Last time I checked there wasn’t a fork of tensorflow-serving using rocm

  • Tensorflow 1.13 works as long as you use the ROCm fork or the docker container, I believe pytorch needs to be built from source. Nothing currently just works out of the box.

  • ROCm is progressing quickly but still lags behind as it is a rather large stack with a lot to do.

  • As of now, navi is not supported.

I recommend the 2070 Super, or a Radeon 7 (if you really need 16GB of HBM, as those caveats still apply).

Also, I have seen some info about NVLINK adapters working in linux to allow for shared VRAM. I don’t know much about this, but if so that is a huge future upgrade path.

2 Likes
6

Thank you. I was not aware of the EOL and the option to be ‘super’ :slight_smile:

7

I’m really not sure if the 2-3% improvement is worth double the RAM price over 3000Mhz Memory for me.