Out of the Box Performance
Pulling the card out of the box and into my machine, which is still running Windows 7 64-bit, I installed the latest Nvidia GeForce drivers (version 382.33 as of this writing.) I fired up Claymore version 9.3 Ethereum miner and got a mediocre 27 Mh/s which is about what a RX470 can produce for much less cost. Anyway, I was aware of this fact and bought the card specifically for Zcash (Equihash) mining (and a bit of gaming), so I was much more interested in the results under that algorithm. Unfortunately, Claymore does not provide his ZEC miner for Windows, so I used another popular mining software, EWBF’s CUDA Zcash Miner.
EWBF is very nice in that it has a built-in performance meter you can enable (use –pec in the batch file) where you can see a real-time Sols per Watt statistic. As you can see in the screenshot above, I am also using MSI’s Afterburner software to control the clocks. The default settings and performance are shown above and we can see we are getting around a 2.70 Sol/W efficiency rating. The EVGA GTX1080 is running at 1936 Core clock and 4513 Memory clock, with the default voltage displayed at 1031mv. It is also running at a comfortable 67C, which it would be cooler in an open air rig, but it is currently in my enclosed computer case.
You can also see we are getting around 472 Sols/s which while good is not necessarily earth shaking, especially considering two RX580’s for about the same overall price would offer similar results. Where the Nvidia cards supposedly shine however, is they can be overclocked some to offer a higher hash-rate while maintaining an fairly lower power draw.
The first thing I tried was to reduce the Power Limit of the card to 80%. I have read reports of some people running even lower with good results, but 80% seems like a good place to start. You can see in the screenshot above that right away we are getting some good results in that the Sol/s remained about the same at 472 Sol/s, but the power usage dropped from 176 Watts to 143 Watts. At the very top left of the image you can still see the last reading at the default settings before I reduced the PL and took the screenshot for comparison. You can also see the built-in performance utility is now giving us a reading of around 3.30 Sol/W versus the default setting value of 2.70 Sol/W.
Reducing the PL (Power Level) also slightly affected the core clock by reducing it some to 1822 MHz and the voltage we can see has also dropped to 931 mv. This makes sense as with AMD cards the best way to lower power consumption is to reduce the wattage which is usually achieved by lower the core voltage. This becomes more of an art than science at this point, as to maintain the desired PL the core clock will need to adjust downwards to operate stable at the reduced voltages. So the goal becomes to adjust all of these settings in such as way that we can achieve the highest Sol/W reading possible while still maintaining an acceptable overall hash-rate.
So while this is off to a good start, there is still more to be done, namely to get some additional hash-rate out of this card. The next thing I tried was bumping up the memory timings a bit. This should be easy and not contribute much if anything to overall power draw. From other reports, with these cards you should be able to get an extra 100 core and 500 memory clocks without too much of an issue. Since I wanted to see the performance as various points I started off with a simple +50 to core and +200 to memory and left the PL at 80%.
As shown above, we are still fighting the power limit throttle as the core clock with a +50 setting actually reduced even more, although the memory did increase the 200 MHz. All is not lost however, as I did notice the hash rate bump up to around 505 Sol/s and with power usage roughly the same we are now sitting at a 3.5 to 3.6 Sol/W efficiency. Compared to the default settings, we have gained about 30 sols and increased efficiency by almost 1 Sol/Watt.
So with these initial results, I can see it will take some experimentation with different combinations of settings to dial in the optimal values. So to spare you with countless screenshots of settings, I have compiled my research into the following table which I will continue to add to as I zero in on the optimal settings.
|Efficiency||Sols/s||Power Level %||Core Clock MHz||Memory Clock MHz|
The two main things to be watching for is the overall Sol/s which if you have low cost or free electricity (such as a dorm room environment) you might weigh this factor more heavily, and efficiency which compares the Sols for a given wattage, which of course will be of concern to most of us who do pay for electricity.
Setting the PL to 110% and adjusting the core clock to +100 and memory to +600, I managed to get 560 Sol/s, and surprisingly even a slightly higher efficiency of 2.9 Sol/W over the 2.7 with default settings. So while going higher is not as power efficient as lower settings, it is becoming apparent that the default settings are just about the worse ones you can use when it comes to mining.
Initial results indicate that going higher or lower from default will result in a better overall efficiency, with the most gains to be made on the lower end of the spectrum. However, as noted above if you have few rigs or are in a low cost or free electricity situation, the overclock and increasing PL route may be of most interest for you.
Taking things in the lower direction, I was able to maintain above a 500 Sol/s hash-rate while increasing efficiency to around 3.8 Sol/W by using a PL of 75%. The core clock settings tend to be ignored when set to automatic for power adjustments, so by applying a +500 to memory clock at a 75% PL resulted in a core clock reading of 1733 MHz, which still generated a respectable 502 Sol/s.
I am still playing with some of the many possible values in between, but so far it appears going to one or the other of these extremes is giving the most noticeable results, and depending upon your situation you may choose to prefer one route over the other.