GameFlip

Posts Tagged ‘Nvidia

A couple of days ago Nvidia has announced their new Geforce GTX 1070 Ti series of video cards and has opened pre-orders with the shipping and official sales expected to start tomorrow in most regions around the world. Prior to the official release there were a lot of speculations about what the new GTX 1070 Ti will be offering and we already know that the new Ti model is actually not that much different. Nvidia probably did not want to endanger the sales of their GTX 1080 GPUs, so as one might expect the GTX 1070 Ti sits somewhere in between the GTX 1070 and GTX 1080 in terms of performance, though with overclock it will most likely be able to outperform the GTX 1080. There are some speculations that the new Ti will be overclocking pretty well and that is definitely good news for both gamers and miners. Let us take a quick look at the more important specs and how GTX 1070 Ti differs from the 1070 and 1080 GPUs from Nvidia (reference designs and clocks).

Looking at the specs comparison it seems that the new GTX 1070 Ti is almost the same as the GTX 1080 as far as GPU specs go (some extra clock should compensate for the less CUDA cores) with the biggest difference being the memory being used. The GTX 1080 uses the faster in terms of speed GDDR5X video memory and offers more bandwidth, but the memory timings are slower compared to the slower in terms of operating frequency GDDR5 used in the GTX 1070 Ti. While for gamers the memory latencies might not make that much of a difference for users and the faster frequency might be fine, for miners the memory timings are more important. So the GDDR5 memory used on the GTX 1070 Ti is good news for people mining memory intensive altcoins such as the ones using Ethash like Ethereum for example. Do note expect however to get much higher performance on these compared to the regular GTX 1070, in fact the new Ti should most likely get you a bit better performance in more GPU intensive algorithms due to the increased number of CUDA cores.

Of course we’ll need to do some testing when the new GPUs are finally out on the market and see how good their performance compared to the old GTX 1070 and the GTX 1080, but we already have a pretty good idea on what to expect. Do note that although Nvidia is using photos of a reference design GeForce GTX 1070 Ti with a Founders Edition cooler it seems that such a model is going to be only directly sold by Nvidia and not through partners, all other designs are non-reference ones with coolers already used for the GTX 1070 or 1080 models. As a result many people will not be ale to get their hands on the Founders Edition due to the fact that Nvidia does not have good distribution channels of their own and thus the company essentially ignores direct sales in many of the smaller markets around the world. Anyway, we already know that the new GTX 1070 Ti should be a great choice for multi-GPU mining rigs instead of the GTX 1070, although the deciding factor will most likely be the end user price of the new video cards… a price that is also expected to be somewhere in between the GTX 1070 and GTX 1080.

With the switch from the RX 400 to the new RX 500 series (even if there is not much difference in hardware), AMD started having issues with availability. Older GPUs have been pretty much sold out where there were some left and the newer 570/580 models are pretty much out of stock everywhere. If there are some units left they are with an inflated price tag, because the people selling them want to make some extra bucks from the lack of availability and high demand. It is not known how long this lack of video cards will continue, but things were slowly progressing towards this negative outcome the whole month.

We are pretty sure that one of the reasons is mining and the recent peak of the price and profitability of Ethereum that inevitably leads to more and more new miners and mining rigs getting built. It seems however that AMD was really not ready for an increased demand after the release of the RX 500 series, even though RX 400 GPUs were already selling a lot. It may take a couple of weeks for things to get back to normal, unless the craze continues and unless AMD is actually having trouble making enough GPUs at the moment to cover the demand. Maybe the fact that the company is preparing for the release of their newer higher-end Vega solutions has something to do with the problem as well.

Miners are already looking at the alternatives of using Nvidia GPUs that are available plenty on the market, because they are not as good deal for Ethereum mining like the AMD cards in terms of price/performance. If you however take out the Ethash algorithm (Dagger-Hashimoto) out of the equation Nvidia’s Pascal GPUs are actually pretty good in terms of performance/power usage ratio… just not as good in memory intensive algorithms where GPU power is not required as much. Mining aside, gamers already have no other choice but to go for Nvidia if they are currently in the market for a new video card with AMD GPUs being out of stock.

The people that were mining a couple of years back during the Litecoin craze will most likely remember a similar situation where things got out of control for a while with the exchange rate of LTC exploding. This was only temporary as everyone was rushing to get into mining the top altcoin back then and then the price went down and things started to get back to normal. There are no guarantees that the same situation will happen again though, but don’t forget that we may be heading for a POW to POS switch for Ethereum by the end of the year…

Time for another group of tests of the newly released GeForce GTX 1080 Ti Founders Edition video card by Nvidia. This time we have used the latest released of the ccMiner 2.0 RC2 fork by Tpruvot in order to test how all of the supported algorithms perform on the new GPU. Do note that not all of the supported crypto algorithms by this miner may be performing the fastest, so you should use the results as a reference if comparing results from 2.0 RC2 release only. We have managed to successfully test the hashrate on most of the supported algorithms by this miner, though we still did have trouble making some of them work, but that is to be expected with a new GPU release…

We have used ccMiner 2.0 RC2 by tpruvot in benchmark mode with all of the default settings for intensity on all algorithms supported by the miner. The GeForce GTX 1080 Ti GPU was running at the default 100% TDP or with other words a Power Limiter at 250W, though on the Founders Edition GPUs you can increase it by up to 20% all the way to 300W (mind the cooling though). The tests were performed on stock operating frequencies of the video card with the fan set at 100% in order to keep the GPU cool and prevent thermal throttling (dropping of the boost frequency) because of high temperature that might be reached with auto fans.

As you can see in the table with performance results above we have posted the hashrate and the power usage for each algorithm tested (if the test has successfully passed). The power usage values are the ones reported by the video card itself only and are based on its TDP limit, not the actual values measured at the wall (these will be higher). The algorithms pushing the 250W TDP limit might be able to benefit from a power limit increase, though you might want to be careful with that in terms of being able to properly cool down the card. On the other hand forcing the TDP limited down might bring significant power reduction with a little sacrifice in terms of hashrate thus giving you better performance per Watt than the stock settings.

With the BMW algorithm we got an error at the default intensity of 21, and we’ve had to decrease it down to 15 in order to make it work, resulting in very low hashrate. With Cryptonight we got an error at the default intensity of 10.75, but down to 10.25 it worked and you can see the result in the table. With the Jackpot algorithm we also got an error at the default intensity of 20, however lowering it did not seem to help at all and the same thing also applies to the Quark algorithm. With the Wildkeccak algorithm we had trouble making it work properly in benchmark mode due to the specifics of the algorithm needing a scratchpad file.


top