You are viewing a single comment's thread from:
RE: Hardware and Project Selection Part 1 - CPU vs GPU
Would it be at all practical to reverse engineer video cards to make multi-GPU boards to handle things like mining, or is that essentially what an ASIC does?
Do you mean having multiple GPUs fixed on the same motherboard? Sure, this is already done in a lot of mining rigs. However, if you had the money to invest another option for GPU mining would be to buy a server/research GPU. These GPUs have incredibly large numbers of cores. For example, the V100 GPU Accelerator from NVIDIA's Tesla range (An FP64 focussed card) has 5120 cores in the single GPU.
ASICs are a bit different, in that they are pieces of hardware designed to only do one job, and do it incredibly well. In terms of mining capabilities, ASICs are far superior to any GPU. However, the problem with such specialized hardware is that it can do pretty much nothing else - BTC ASICs mine BTC, and that is it. There are some rare exceptions, such as chips that mine BTC and LTC, but this is because the chip package effectively incorporates two ASICs, one for each coin. An ASIC could never, for example, render your display.
Ah I see ... I think something like the server gpu is what I was talking about, that way I could still mine pools and get a variety of payouts. I'm guessing the investment for something like that is obscene though and the ROI would be prohibitive?
We could do the math right now if you like!
The V100 GPU Accelerator is FP64 orientated, so lets look at the FP64 project MilkyWay@Home. Vortac is the top miner in that project, and it generates him an income of about 180GRC/day. He does this using 6 Radeon HD7970s, which supply 950 GFLOPS of FP64 processing per card. That means the return per day is about 0.03GRC per GFLOP per day.
As the V100 GPU Accelerator has 5120 GFLOPS of FP64, it would have a predicted yield of 162 GRC/day.
Despite those yields between the 6 HD7970s and the single V100 GPU Accelerator being similar, 6 HD7970s would set you back maybe USD$1000, while a single V100 GPU Accelerator is USD$15000. In terms of power consumption the V100 GPU Accelerator is far superior, but the difference in start-up capital required makes it unfeasible.
To get best ROI, especially in the FP64 GPU applications, older hardware is the way to go. Very specific pieces of older hardware, mind you. Research is required.
My Milkyway numbers are actually way down (at least 30%) because of the summer and the latest heat wave here. I had to severely downclock all my GPUs, plus I have to turn BOINC off during the day otherwise it's simply too hot and I have to run my AC at max power all the time.
There is even something between an GPU - which can theoretically run everything, as can Mincecraft ;) - and ASIC. THis is called FPGA (Field Programmable Gate Array)
As the name says those things can be programmed, but different then the CPUs (hard to explain).
The FPGAs can only do the task they are programmed to do, you cannot run 2 different programs without reprogramming. But you can run any program, so you could first mine BTC and then Ether.
Because of their structure they are likely to be more speedy then CPUs, maybe even a bit faster (or more energy efficient) then GPUs, but you would have to write a programming first, for every single different FPGA. So it is not worth the trouble.
An FPGA might be something useful for the pool I use (nicehash) because I could potentially program one to run Cryptonight and one to run Equihash thus increasing my payout for both.
It's less a matter of reverse engineering video cards but rather porting CPU work to GPU (or FPGA). If you can do this and realize performance gains then you could earn serious GRC.
I'm a bit too new to know what GRC stands for. I'm guessing it relates to H/s?
GRC stands for Gridcoin. It is the abbreviation used on the markets.
You could compare it to how we sometimes write United States Dollars as USD.
Got it. It's a currency I haven't done any research on. I've mostly been looking at ways to improve my payout from nicehash, and maybe get a piece of the Ethereum mining action. Is GRC easier to mine?
Once you are set up, it is exactly the same amount of effort as any other coin. Getting set up requires a little more work though, as Gridcoin pays out based on contributions to scientific research done through the BOINC platform. You can't 'mine' GRC in the traditional sense --> the compute power of the network is actually put to use here.
There are lots of great tutorials on getting set up, both written and videos. If you have a little patience, you could be fully set up mining in a pool in an hour or tow. To mine solo, give it maybe 2 days of intermittent activity.