Why my GPUs beat your ASICs at crypto mining

There is a constant question for everyone doing cryptocurrency mining – what is the most profitable hardware to mine?  Here we explore that question and try to quantify why today (January 13, 2018) we believe GPU mining is the best for crypto mining – even better than ASICs.

Now we have to quantify “best” as this is very much a moving target.  By best in this case we aim to blend profitability, cost of entry, and risk mitigation in a market that is constantly changing.

A bit of background on some of the (silicon) chips

CPUs GPUs FPGAs and ASICs
CPUs GPUs FPGAs and ASICs, source Microsoft

There are a number of types of chips available.  The performance and other critical characteristics such as power consumption for the chips vary depending on how they’ve been designed.  Without going into too much detail we can compare them as follows:

  • CPUs – Incredibly flexible, and capable of executing a wide variety of singular tasks at a very high rate
  • GPUs – Not as flexible as a CPU and requires a bit more specialized programming to leverage, however modern GPUs contain thousands of cores which are capable of a very high degree of parallelism for repetitive tasks
  • FPGAs – Require specialized programming for each task, but are highly efficient (from a power/performance perspective) and can be flashed with new code as requirements change
  • ASICs – The most simple and efficient (again from a power/performance perspective), they’re also inexpensive to manufacture at scale.  The major disadvantage is that the program is (effectively) burned onto the silicon – it does one thing very well and only one thing.

So what is best for mining?

Initially, when we started crypto mining in earnest, given these characteristics we focused on ASICs.  Indeed if you want to mine the de-facto crypto standard, Bitcoin, there really is not a viable alternative.  That said, if we use the definition of “best” above we find that in the current market with altcoin (non-bitcoin crypto currencies) we can achieve the same level of profitability per cost by GPU mining.  Once more if we factor in the resale capacity of the market for GPUs, it quickly becomes a highly advantageous option.  If the crypto market fails I can resell a GPU whereas resale of ASIC hardware is problematic.

What is the best GPU?

I wrote a separate article on crypto mining with the NVIDIA GeForce 1080 Ti.  At the moment this is the best consumer-grade GPU on the market and as a result it’s in high demand.  If you’d like to get your hands on one of these cards here are links on Amazon to a few of the models that I have:

What do you think?  Think we missed a critical point or disagree?  Please comment, we love to hear your feedback!

Erik

Technology, science, building things and experiencing the world. What more could anyone ask for?

Share this idea:

Leave a Reply

7 Comments on "Why my GPUs beat your ASICs at crypto mining"

avatar
  Subscribe  
newest oldest most voted
Notify of
kok keong
Guest

Mr. Erik, how can I start to learn FGPA, which card to purchase first. Xilinx or Altera? I want to learn to program FPGA for mining purpose but dont know where to start and the know how.

Abe Fouhy
Editor
Hi Kok, Great question! Xilinx vs Altera is kinda like Toyota vs Honda. They are both good, but it is about personal preference and what language you want to program in. I am currently working with both side by side. I would say that Xilinx isn’t quite as proprietary in their license structure with basic documentation and Altera has better documentation with more licenses needed. This said Altera has a more robust (in my opinion) design suite and it easier to learn for the beginner. I would suggest the Terasic DE1-SOC board or the De0-nano board. These are great and… Read more »
Rochester
Guest

oooops. i’d like to see the course of FPGA mining design. 🙂

Rolland
Guest

Hi Abe, Thx for the great overview of this facinating technology. Given that a FPGA is a programmable device, do you think it is possible to load different algorithms into one in a a profit switching scheme (similar to a GPU mining Nicehash)? If so how long to download a new algo and how many times can a typical FPGA be over written?

Abe Fouhy
Editor
Hi Rolland! Thanks so much! I am glad you liked our content. I am truly fascinated by all of the wonderful things the FPGA is capable of, from image recognition, machine learning to mining. The FPGA is capable of mining any algorithm similar to that of a GPU, but one has not proven to be able to switch between algorithms as easy as a GPU yet, because the device is physically changing the hardware layout inside the chipset. I really like your idea of loading bitstreams dynamically from a file system as the algo profitability changes like nicehash. It only… Read more »
Abe Fouhy
Editor

Also — Lets work together on this Rolland! Thanks bud.