Crypto Mining Blog
It Is All About BTC, LTC, ETH, DOGE, KAS mining as well as other alternative crypto currencies
Search
Search Results
Normally when you run a multi-gpu mining rig you are monitoring the operating temperature of the GPU and if it is keeping an acceptable level, then you think that everything is fine and may not pay attention to other components. We have done an interesting experiment with a Gigabyte GeForce GTX 970 WF3OC-based mining rig taking thermal images of the back of one of the cards in the setup while it is mining Ethereum and while it is mining Decred. As you probably know the Dagger-Hashimoto algorithm used by Ethereum is heavier on the memory even though the total power usage is lower while Decred and the Blake-256 implementation it relies on is more power hungry and not so dependent on the memory.
The power usage of the GTX 970 while mining Ethereum is around 150 Watts while mining Decred it goes to about 175 Watts. The GPU temperature of the GTX 970 while mining Ethereum in the test mining rig is around 68 degrees Celsius while mining Decred it goes to about 75 degrees Celsius. This of course is with custom settings for the fans, because normally Nvidia wants to keep the target temperature at 80 degrees Celsius, but that might not be that great for 24/7 mining. But if you look at the back of the GPU with the help of a thermal camera you can find pretty interesting things like the fact that the backside memory chips that do not have cooling like the ones on the front of the card actually get pretty hot. A bit too hot if you ask us especially in the case with Ethereum mining with their temperature reaching about 105 degrees Celsius while the back of the card itself remains at not more than 75-80 degrees. In the case of Decred mining, even though the total power usage is higher, the memory chips remain with over 20 degrees lower temperature and the back of the card is similar in temperature, so the difference of the overall temperature is not that big.
So if you are mining Ethereum, especially if you are getting into it just now, you might want to be sure that the GPUs you are using have better cooling on all of their memory chips, including the ones on the back of the card if there are such. While there might be no issues in short term with such higher operating temperatures of the video memory chips, on the long run temperatures of over 100 degrees Celsius might cause you hardware issues and you would want to avoid these on a mining rig that needs to fun 24/7. Do note that while this experiment was performed on Nvidia-based hardware you can expect to see similar results on AMD-based video cards as well with Ethereum driving the temperatures of the video memory higher than other mining algorithms.
Recently we have picked up the last 20 releases of the ccMiner SPMOD fork for Maxwell and tested them on a Gigabyte GTX 970 WF3 GPU. running at stock frequencies to see what has happened with the hashrate in the X11, Quark, NeoScrypt and LyraREv2 algorithms. There were users that wanted us to repeat the same test with a GTX 750 Ti GPU, so we have picked up an eVGA GeForce GTX 750 Ti video card and repeated the same tests. The first version we tested with ccMiner 1.5.53-git SP is from early July and looking at the results it seems that the most notable improvement is with Quark as well as some slight improvement with X11. Lyra2REv2 support has been added in release ccMiner 1.5.60-git SP, so there are no results prior to that, it seems however that the initial versions and the last one have pretty much the same hashrate with some drop with some in-between versions. The NeoScrypt hashrate hasn’t changed much, there is some slight variance with different versions probably due to the Boost frequency of the GPU variating a bit. Do note that these are results with publicly available and open source code, there could be some higher performance private miners also available and you might be able to squeeze more performance with different intensity settings (other than the default ones) as well as by overclocking the video card.
Last week Nvidia has announced a new mid-range GPU, namely the GTX 950 that is being positioned between the GTX 750 Ti and the GTX 960. Interestingly enough the GTX 950 has 128 CUDA cores more than 750 Ti and a TDP of 90W with a price point that is a bit higher than the 750 Ti, so the big question is if the GTX 950 can be the new affordable and low-power usage alternative for mining suceeding the 750 Ti. We took an MSI GeForce GTX 950 Gaming 2G video card and have tested how it performs for crypto mining in various algorithms using the latest ccMiner fork from SP optimized fro Maxwell GPUs as the GTX 950 is Compute 5.2 card. We have also tested using Ethminer for mining Ethereum’s Ether (ETH) coins as these are pretty hot and profitable to mine at the moment.
The table above does not contain results for the following algorithms that are supported by ccMiner: lyra2, skein, yescrypt, whirl, x14, x17 as when trying to test them we have encountered errors or the miner software crashing. We are not sure if this is directly related to the use of GTX 950 and the default intensity settings not working for some algorithms, or just that there are issues with the specific algorithms in the latest code of ccMiner. The TDP result shows how much power the video card uses for each of the algorithm with 100% TDP being equal to 90W of power usage on a normal GTX 950, but on the MSI factory overclocked Gaming model the 100% TDP is actually equal to 110W. As you can see however very fee of the algorithms are getting close to the maximum TDP of the GTX 950, so the actual power usage could be even lower than 90W on most of the algorithms.