This series is intended to show you how to build a computer with multiple GPU’s used to mine cryptocurrencies such as Ethereum. These systems can also be used for machine learning (ML) and rendering, but we’ll focus our build on mining. We’ve ran these builds through all three of these workloads and they performed well for our uses. This build list is focused on mining and not the cheapest, but with the idea that we can upgrade it with little extra cost to be used for rendering on the RNDR network.
note: Rendering and Machine learning performance will vary depending on render scene complexity/size and your ML model respectively.
Our builds utilize NVIDIA GPU’s because of their multiple uses across many fields such as mining, gaming, machine learning and rendering. There are more products available out there that support NVIDIA, like Octane Render/RNDR Network, but not AMD cards (yet). NVIDIA cards are also easier to do performance tweaks on.
These machines are POWER HUNGRY, so please be aware of what your electricity costs are. For example, an NVIDIA GeForce GTX 1070 can use up to 150 Watts of power, multiply that by 6 and add the rest of your system and your 6 GPU machine can use over 1kW (kilowatt) and will generate LOTS of HEAT.
This series will be broken up into three parts.
Part 1: The Hardware (this post)
Part 3: Setting up the software (coming soon…)
Lets jump right into it…
ASUS Prime Z390-A: This motherboard is reasonably priced ($188 at the time of this writing) and has the ability to install up to 8x GPU’s.
*Added bonus is that it has 2x PCIE x16 and 1x PCIEx4 which gives you better performance if you ever use it for Rendering or Machine learning. You can read more about why the larger PCIE lanes gives better performance in this [[follow up RNDR blog post.]]
For mining you just need the minimum amount of RAM as mining is not RAM heavy (vs GPU RAM). 4GB of RAM would be the minimum we’d recommend. The ASUS Prime Z390-A can hold up to 64GB of ram.
*If you plan on using this for the RNDR network, rendering or machine learning, 16GB is the minimum we would recommend.
Patriot Viper Elite 4GB – (2133 or Greater, 288-pin Desktop RAM) $24.99 at the time of this writing.
If you want to upgrade the system for Rendering/Machine Learning, you can get the Patriot Extreme Performance 32GB PC4-22400 (2800MHZ).
PCIe to USB Riser Version 006c – Most motherboards are not built to fit 6+ GPU’s at a time, but we want to fit as many as possible. This requires us to buy additional hardware to extend the PCIe slots and connect all GPUs safely.
Graphics Cards (GPU)
Power Supply Unit (PSU)
**NOTE: If you buy 6x 1080ti, you’ll need TWO PSUs to provide enough power and connectors. If you’re only running 6x 1070’s, you’ll be able to run them off of one PSU.
Closed or open case? We prefer closed as it’s easier to manage without fear of damaging any component, with an added bonus that they can be installed in standard 19″ server racks. But overall this is a preference and cost factor, we’ll list two cases you can use with success. Both cases will require you to buy Fans to blow the hot air off the GPU’s and keep the system cool.
The biggest thing to remember is to make sure either system is getting enough air flow to keep the GPU’s cool. Rule of thumb is to keep them under 70 Celsius.
Hydra II 8 GPU 6U Server Mining Rig Case, Dual PSU Ready – This case is great as it’s enclosed and easy to mount, move and handle. It also allows better controlled airflow over the GPU’s and can bet installed in a standard data center 19″ rack.
8 GPU Miner Case Aluminum Stackable Open Air Frame – These are only $30 (at the time of this writing) and minimal. Putting together this frame is a bit more involved (think Ikea).
Your cases will need 120mm fans to keep your GPU’s cool. The fans that sit directly in front of the GPU’s will need to have a high static pressure fan rating, while those that are used to blow the air out of the case will need to be high air flow rated. We’ll list two possible options below. Here is a post that goes in great detail about types of fans.
Noctua NF-F12 iPPC 3000 PWM, 4-Pin, Heavy Duty Cooling Fan with 3000RPM (120mm, Black) – High static pressure fan good for pushing lots of air through confined spaces. With a sound rating of 43.5dB, these are loud, but will keep your system cool. Not good for being in an occupied room.
Noctua NF-F12 iPPC 2000 PWM, 4-Pin, Heavy Duty Cooling Fan with 2000RPM (120mm, Black) – Quieter than the 3000 PWM rating at 29.7dB, it’s much more doable if you’ll be working around this machine often.
https://www.gpuminingresources.com/p/psu-cables.html – More details about PSU cables and their power limits.
https://www.biostar.com.tw/app/kr/event/crypto_mining/page4_2.htm – Example of how to connect up a system on a Biostar motherboard (Some examples apply)