There are many factors that affect the performance of mining operations involving GPU-based miners. Central to this group are the GPU's speed, the temperature of the unit itself and the software that is being used by said unit.
For every GPU there a thermal limit exists that may not exceed be exceeded and reaching that limit means the cards will slow down. The essence of having as much cooling is to ensure that the limit is never reached and avoid thermal throttling, which might take away as much as 10% of your overall, average hashrate.
The hash algorithm for your mining software may also act to affect/hinder hashrates as is the factory clock speed. Regarding mining software, it is necessary to test as much as possible and see which particular one renders a good reading for hashrates in 30-60 seconds.
Overclocking can also help reduce power usage and hence saving on electricity bills and even see you achieving cooler temperatures and more life for your machine. However, it would void your warranties with the card manufacturers and may freeze the system to force a system reset. Some manufacturers do provide overclock software that can be used to avoid risking warranty.
Overclocking and GPU performance
GPUs are manufactured to operate at certain speeds which are called the base clock and memory clock speeds, but most have the potential to surpass that speed set by the manufacturer. Overclocking increases the clock cycles per second for the GPU card and optimizing your GPU this way can see you get an increase of 20-40% in hashrate, which could render the GPU involved more productive in cryptocurrency mining.
To carry out overclocking, you need to have a card unlocked for overclocking. You will also be required to have overclocking software such as the MSI Afterburner, which works on any GPU brand and allows you to modify all sorts of settings on the GPU including the core voltage, power limit, temperature limit, core clock, memory clock, and fan speed. These are the most important aspects to overclock. Some mining software also allow you to adjust these settings directly. Other software that you could use for these purposes include: EVGA Precision XOC, FireStorm Utility, Xtreme Engineering, and GPU Tweak II.
Tweaking different settings will have differing effects. Lowering the input voltage of the core GPU or the core voltage can, for instance, lessen power consumption but it is advisable to leave that alone unless you have an issue with temperature or power consumption.
The power limit can be set to between 65% to 85% without reducing performance and doing so will also control the maximum power the GPU is meant to consume or draw from the wall it's connected to. The temperature limit sets the maximum temperature before throttling or shutdown. The core clock can be left at -75 or -100 without affecting performance while the memory can be set by as much as +800 Mhz for some cards and is the most important setting for mining crypto with GPUs. The fan speed is left at auto to let the cards decide and because setting a higher fan speed might reduce the life of the miner and increase the noise and power consumption of it at the same time.
When using software such as Afterburn to overclock speed, you can either overclock speed for one card or for all that are running in your GPU. Since the settings will be applied automatically to all cards on the system, you can change card and overclock one at a time if you want to make changes on each card.
Finding the optimal settings for your GPU crypto miner may not be a straight-forward job; it could involve trial and error or even simply finding out what worked out for other people. So in most cases it is a trial and error but there are many tools to help. You can increase value for each of the setting step-wise, testing each time you vary the values, until the GPU crashes then start to lower the values for the settings slowly and test each time you make a change, until you achieve the optimal or suitable hashrate increase for GPU stability. Some can be unstable at certain hashrates, which means you will need to lower the hashrate for mining cryptocurrency.
Before you kick off tweaking your GPU, you can use UniGine’s Heaven Stress testing software to run a base test and see how the graphics cards first perform with default clock settings. You basically install this software that then run tests in about 10 minutes. Testing will provide crucial data for your GPU at default settings, including temperature, frames, graphics clock, and memory clock and this is the starting point to overclocking. Again, you can save these settings and return to them later if overclocking doesn't go on well as expected.
After stress-testing your miner’s base clock speeds with UniGine’s software and marking the temperature during testing, FPS at the end test, and score at the end of test, you can then install software such as Afterburn and then set new values for unlock voltage control, unlock voltage monitoring, and Force constant voltage and then stress retesting for the new values with the Heaven Benchmark software. You can repeat this several times until you achieve stable hashrates.
With software such as Clocktune, you can overclock multiple GPUs in one rig at once by entering multiple values separated by spaces. This software, therefore, also lets you manage overclocking profiles; for instance, you can apply different overclocking profiles to one or more workers at the same time. With it, you can also assign different or same overclocking profiles to different specific mining algorithms as you wish. For instance, it is possible to define two overclocking profiles in the software then assign one to Equihash mining and then the other for mining Ethash coins for instance. These overclocking profiles can also be edited at any time or duplicated for existing or additional mining algorithms.
If you are mining on a Raspberry Pi, then you can also overclock its CPU by changing the “arm_freq” parameter – Frequency of ARM, overclock the RPi GPU by changing the “core_freq” parameter – Frequency of GPU processor core, overclock the RPi Memory by changing the “sdram_freq” parameter – Frequency of SDRAM, extending extension for the RPi Memory using ZRAM by compressing the memory, overclocking the voltage or the ARM/GPU core voltage, and doing a “Force_turbo” to disable dynamic cpufreq driver. Further details are on this link.
About how far to overclock, it is also hard to work with a friend's GPU settings because each one GPU is different and yours could overclock lower or higher than your friend's. Therefore, besides getting example settings from your friend's you might want to test yours for some further settings.
However, pushing these GPU values too high will either show graphical glitches or the PC may finally crash but it is still possible to push the settings lower after this. Otherwise, BlackScreens, BlueScreens, PC Crashing are completely normal during the process. Other signs such as white dots ("snow") should also mean that you need to reduce the values down.
When working with cards such as Nvidia’s GTX 1060, 1070, 1080, Titan X/XP, or GTX 1080 Ti, you can expect a +50 increase in core clocks which can rise to 80-85% tdp or 120% if concurrently rising electricity costs are not a problem for you. The 1080 Ti can also add a +500 memory overclock as a 1800-1900MHz core clock. Generally, you can expect different overclock settings for different mining algorithms.
Also, when considering what overclocks a miner best -- either a memory or core clock -- evertything depends on the central algorithm involved. Some algorithms such as Ethash are used for memory-hard cryptos and therefore overclocking a memory clock when mining such a crypto algorithm is a workable solution. This works out especially well compared to using Equihash, where you would still need to be overclocking the core clock in addition to the memory clock. However, these two settings, as well as others, can be changed during overclocking.
GPU performance and temperature
First, each of the card has a different TDP (thermal design power) level set by the manufacturer, and this can be changed in order to lower power consumption and reduce heating of the GPU. For GPU temperature, anything less than 75°C is totally fine for any modern GPUs, and this can be attained at a 85% TDP. Meaning a 100% TDP (Thermal design power) on your tweaking software means you may be wasting more power without necessarily getting better mining productivity results.
However, GPUs can now run at as high as 80°C and even 90°C temperature when new but six months down the line, you may have to bear with lower temperatures may be to the extent of 75°C. Plus, higher temperatures will see it wear down faster than when operated at lower temperatures so you may overclock to lesser levels to make your cards last longer.
Running your GPUs at a very high temperature beyond acceptable ranges for a long time may be the easiest way to damage them. Therefore, monitoring the temperature of the GPU can be very good. The truth about GPU operation is one has to balance between performance, temperature, and noise. Most people also reduce power usage to lower temperatures or to reduce power draw and bills. For this purpose, Nvidia cards are capable of handling as low as 60% of power usage while AMD cards can handle 80% power usage.
If you are considering overclocking a GPU, apart from setting its TDP at 85%, which is an ideal level related to power consumption, tweaking other settings such as clock and memory speeds may increase the temperature meaning the GPU will draw more power and increase heating. So again, monitoring the GPU temperature with software can help to detect when things have been overdone (beyond 75°C or say 90°C if they are new) and to take action. Cooling the GPU can also be a great way to ensure you do not reach throttling level. When such target temperatures are exceeded for a given GPU, you might experience lower clockspeeds (throttling) or higher fan speeds or a combination of both. Thus it may be normal to consider extra cooling. And while some GPU cards prioritize quietness of the machine over cooling aspect, it is better to have those that prioritize cooling.
When mining cryptocurrency, acceptable hash rates are produced at a temperature range of 60°C – 69°C and not anything lesser, therefore maintaining the temperature at between 60°C and 75°C is great, but better hash rates are achieved at 70°C – 75°C (90°C if they are new).
Although most GPUs can handle a range of 75°C – 79°C, they will, at this temperature, not be producing any better results than when operating in the 70°C – 75°C range. At 80°C – 89°C, the hardware may start to spoil and power wastage goes on unless the card again is new. Nevertheless, different cards will mine different algorithms best at different temperatures therefore this may be an issue of trial and error to find the best temperature for mining your algorithm.
Dealing with high temperatures in GPU mining
The issue of temperature uncertainty means monitoring of temperatures in mining GPUs is crucial, and it can easily be done using software such as the MSI Afterburner or Tech power up GPU-Z to monitor VRAM and GPU temperatures. Consider also that some cards do not have VRM temp and most of cards do not have VRAM temp sensors which is why you may want to hold it at 60-70C. Core Temp can be used to monitor CPU temperatures and it displays temperature for each individual cores of the CPU.
Cooling can prevent overheating and throttling. Boosting the fan speeds, blowing air directly to the metal and further tuning the voltage/frequency curve or tweaking settings on an SLI or CrossFire setup or on MSI Afterburner or other tweaking tool can see some drop in temperature.
Therefore, when you crank up GPU speed during overclocking, you might also need to crank up fan speed in order to keep the temperatures down although higher RPMs might still mean loud noise for some fans and rigs. With MSI Afterburner, you can deal with hot graphics, whether they are of AMD or Nvidia GPUs types. Plus you can also use the AMD Radeon Settings to bring down the temperatures and speed. This can also be done on EVGA's Precision X1 if working with Nvidia GPUs. You can also do so with different software and utility from other manufacturers. In this case, you will be targeting to tweak fan speed, clockspeed, and voltage.
Fan speed and voltage can be first targets when you want to boost performance, otherwise, you might want to play with all the three when you want to decrease noise. Power usage increases linearly with increase in clockspeed but with square of voltage, hence a 10% reduction in voltage will reduce power consumption more than would a 10% drop in clockspeed. Most GPUs can be turned down by about 0.1V, maybe 0.2V.
Under-volting can better reduce noise and temperatures especially for AMD's Vega cards otherwise for newer Nvidia cards (RTX series) that do not allow under-volting, you can try tweaking voltage/frequency curve. For these, using Afterburner software, press Ctrl+F and flatten out everything past the maximum voltage you want the GPU to use. Make sure to stress test for every change you do and make changes step-wise.
Further enhancement and results, for instance, preventing overheating can be achieved by, in addition to more cooling, and ventilation in case you are using an enclosed PC (you can open one side), and ensuring there is enough space between each of the graphic cards in case you are using a rig. At hot weather, air conditioning the mining room may be a nice idea if you are using rigs...otherwise, you could consider an external fan and open windows to ensure better airflows. Finally, dust, in addition to crashing the GPU, may add in some heating shield.
You might also want to invest in an aftermarket CPU cooler to replace or complement stock fans especially if you are overclocking your CPU, move your GPUs or CPUs closer to ventilation or to more or better ventilated areas. Cable management can also help to avoid blockage of airflow into the cards. For instance, you might want to get a modular PSU in order to have full control of the wires and keep wires from messing with the airflow.
If you are looking for some ways on how to lower GPU temperature beyond these tactics, you might also try replacing the thermal paste after some time of the GPU running, replacing or adding your fans, or better still, buy a GPU water cooling system which are perfect for overclocked GPUs and with good set up can lower temperatures to less than 50°C. Otherwise, if things get worse, you could also try disabling overclocking or underclocking the GPU clock speeds using MSI Afterburner.
Mining software selection and GPU mining
For every GPU, there is a correct hashrate for each algorithm it supports. Different cryptocurrencies use different algorithms and different mining GPU machines for different crypto will therefore have different hash rates when mining different cryptos.
But achieving a higher hashrate is not so much of a preference when looking for a mining software than considering what mining hardware supports that algorithm or even, sometimes, how many algorithms a software supports. For instance you may choose a mining software because it supports multiple mining algorithm allowing you to easily switch the coins to mine at different times. Again the differences in hashrates for different software that support same algorithm may be very minimal for mining that algorithm.
Different mining software support different algorithms and software that supports multiple algorithms can be utilized to mine multiple types of coins. For instance, ccminer does support X11, X13, Quark, Lyra2REv2 and around 60 algorithms in total. In fact including equihash for Zcash mining, Decreed for Decreed coin, cryptonight for XMR, and sha256d for SHA256d (Bitcoin). But that does not make it the best for mining each of these coins.
When mining Monero with ccminer, you can get a hashrate of around 200MH/s but hashrates of around 250H/S with XMR-Stak, another software that supports Cryptonight algorithm. On AMD R9 390X, SRBminer performs better at 874 hashes per secs while Cryptonote performs 855 hashes/s for mining Monero.
Again, CCMiner is more specific and supports only GPU and on Nvidia GPU unlike xmr-stak, which supports CPU and/or GPU mining on CPU, Nvidia GPU, or AMD GPU modes, or any combination of the three. However, for Monero mining, XMR-Stak is known to generate a higher hash rates than other miners. Hence the software chosen will influence the hashrate although to a very small degree.
Last but not least, if you want to know which AMD, GPUs, ASICs to use, MiningBenchmark.net website tracks top ten most profitable AMD GPUs and ten most profitable Nvidia GPUs, 10 most profitable rigs, 10 most profitable ASICS GPUs, 10 most profitable CPUs, latest added 10 GPUs, and latest added 10 CPUs. It includes stats such as their core and memory clocks, mining software they work with, power consumption, speed and algorithm, hash rates/watt and revenues per day and month and earnings per month when electricity is deducted and even some proof uploaded.