GameFlip

Posts Tagged ‘GTX 980 Ti

ccminer-1-5-54-git-spmod

Time for a new Windows binary release compiled from the latest Git source code of the ccMiner 1.5.54-git SP-MOD fork of the Nvidia GPU miner optimized for the latest Maxwell-based video cards by SP (source). The new updates are mostly focused around the improved djm34 Lyra2RE performance code that was made available, so his work is integrated and further small increase in performance is also achieved with additional optimizations. We are seeing around 150 KHS performance increase with the default settings on the GTX 980 and GTX 980 Ti GPUs over the djm34 release, on slower GPUs the performance increase is lower. Do note that this 54 release may have some issue with local wallet mining as it appears we are having some trouble with that, so if you do experience a problem you should go back to the 53 release that works properly at least for us. The SP-MOD fork of ccMiner is designed for Nvidia Maxwell GPUs such as the already available GTX 750, 750 Ti as well as the newer GTX 960, GTX 970, GTX 980 and GTX 980 Ti and GTX Titan X. The Windows binary release we have made available here is with support for Compute 5.0 and Compute 5.2 GPUs or with other words only for Maxwell-based cards compiled with CUDA 6.5 and VS2013.

To download the latest ccMiner for Maxwell version 1.5.54-git by SP for Windows OS…

ccminer-1-5-53-sp-mod

A quick update in the form of a new Windows binary compiled from the latest Git source code of ccMiner 1.5.53-git SP-MOD fork of the Nvidia GPU miner optimized for the latest Maxwell-based video cards by SP (source). This latest update includes some more fixes and performance improvements in X13 as well as some additional Quark optimizations. The SP-MOD fork of ccMiner is designed for Nvidia Maxwell GPUs such as the already available GTX 750, 750 Ti as well as the newer GTX 960, GTX 970, GTX 980 and GTX 980 Ti and GTX Titan X. The Windows binary release we have made available here is with support for Compute 5.0 and Compute 5.2 GPUs or with other words only for Maxwell-based cards compiled with CUDA 6.5 and VS2013.

To download the latest ccMiner for Maxwell version 1.5.53-git by SP for Windows OS…

amd-radeon-r9-fury-x

The latest high-end GPU from AMD, namely the Radeon R9 Fury X, that has been just recently announced has managed to provide performance high enough to be equal to the competition in the form of the Nvidia GeForce GTX 980 Ti… at least hen we are talking about using video cards based on these GPUs for gaming. We already did some benchmarks to see how fast the GeForce GTX 980 Ti is for mining, so it is time to see if the Fury X will be able to compete with these results like it does for gaming. An interesting advantage that the Fiji GPUs used in the Fury X video cards is that they come with HBM memory and both the video memory and the GPU are water cooled. In theory this means that you get lower temperatures and silent operation and the Fury X does manage to provide that. One more thing that you would normally expect from a water cooled GPU is to have a lot of potential for overclock, but unfortunately this does not hold true for the Fury X or at least not with the first card out on the market at least. We barely managed to squeeze out just about 75 MHz extra from the GPU and there is no option to overclock the video memory for the moment.

amd-radeon-r9-fury-x-quark

So off to do some tests with some of the popular and more profitable algorithms for mining lately, we have used the latest sgminer along with some of the optimized forks available for different algorithms such as the one optimized for Quark and Qubit. Note that there are not yet specially optimized kernels for the Fury X or settings that work best, so we were trying to use ones that we are familiar with already from the 280 and 290 series from AMD. So some of these worked really well, while others not so well and the generally available OpenCL kernels did provide quite disappointing results actually…

AMD Radeon R9 Fury X Results:
– X11 default: 6.778 MHS
– X11 Wolf0 Mod: 8.123 MHS

– X13 default: 5.614 MHS
– X13 Wolf0 Mod: 7.176 MHS

– X15 default: 4.69 MHS
– X15 Wolf0 Mod: 6.335 MHS

– Quark modified: 22.37 MHS
– Qubit modified: 21.15 MHS

– Neoscrypt default: 147 KHS

– Lyra2RE default: 287 KHS
– Lyra2RE Pallas Mod: 450 KHS

As you can see the results are a bit disappointing at this point, everything apart from the Quark and Qubit performance using the modified 280X kernels where the performance is really good and similar to that achieved on the GTX 980 Ti. Other modified/optimized kernels such as the ones for the X algorithms from Wolf0 or the Pallas mod for Lyra2Re do increase the performance a bit, but it it still disappointingly low. In fact on some algorithms you can expect to get worse results with Fury X than with a 280X for example, so we need to find better settings and maybe get optimized kernels especially for Fury X in order for that card to be viable for crypto currency mining. The only thing it is good for at the moment is for mining coins using the Quark and Qubit algorithm with the modified kernels, but then again buying a GTX 980 Ti instead could be the wiser choice for generally better mining performance for any algorithm supported… at least for the moment.


top