Archive for the ‘Tests and Reviews’ Category

Performance of the AMD Radeon RX 480 for Other Algorithms

4, Jul 2016

amd-radeon-rx-480-nicehash

We have already covered the performance of the new AMD Radeon RX 480 GPUs for mining Ethereum, so it is time to see how the GPU performs with other popular algorithms. We have already mentioned that the RX 480 apparently has some issues running sgminer (at least under Windows), so we were not able to run tests with many of the currently popular algorithms, but it seems that many people missed that part. The good news is that there is now an updated version of the NiceHash Miner that apparently adds support for GTX 1070/1080 and RX 480. So we have downloaded it and ran the built in benchmark and you can see the results above as well as the fact that some algorithms still have issues apparently hence the 0 MHS results, but it is much better than before. We have noticed that the sgminer that is being packaged and used for RX 480 now has some pre-built binaries for Ellesmere (RX 480) included. So even if you are not planning on using the NiceHash Miner on Windows, you still might get the sgminer-5-4-0-general from the bin folder and use it on Radeon RX 480.

amd-radeon-rx-480-vs-gtx-1070-970

Here is a chart comparing the performance of a reference design AMD Radeon RX 480 from ASUS in the other algorithms apart from Ethereum to an Nvidia GeForce GTX 970, namely a Gigabyte WindForce OC model and a reference design Nvidia GeForce GTX 1070 from ASUS. The benchmark did not produce results for NeoScrypt, WhirlpoolX and Blake256r14 as you can see, so it seems that some algorithms may still have issues running on RX 480. Also do note that the RX 480 has been around for just a few days, so there have not been any specific optimizations for the new Polaris architecture that it uses and further performance increases might be possible. The X11 Evo algorithm is not yet supported by the sgminer for of NiceHash, so there are no results as the dedicated miner is having trouble runnign on the RX 480 on Windows (we have not yet tried Linux).

The pleasant surprises are in Blakecoin and the X-based algorithms where the result of the RX 480 beats with a bit what the GTX 970 manages to provide in terms of hashrate. Unfortunately in the others the GTX 970 turns out faster than the RX 480 for the moment and the AMD card can definitely use some improvements in algorithms such as Lyra2REv2 and Quark for example. The GTX 1070 however manages to provide a significantly higher hashrate compared to the RX 480 and with a lower power usage than the AMD card. Again the RX 480 could get some tweaks and fixes and it definitely needs some and will most likely manage to catch up to the GTX 970, but reaching the GTX 1070 is probably too much to expect.

We should not forget that the GTX 1070 is a significantly more expensive GPU than what the AMD Radeon RX 480 sells for, but still we did not expect doubled or almost doubled hashrate provided by the GTX 1070 in most algorithms. It seems that the benchmark results we get for crypto mining and the ones when using the RX 480 for gaming are pretty much the same (apart from memory intensive mining algos such as Ethereum) where the new AMD card manages to be on par or a bit faster than a GTX 970 in some of the cases.

Nvidia GeForce GTX 1070 Overclocking and Mining Hashrate

30, Jun 2016

nvidia-gtx-1070-overclock-settings

Time for some overclocking of the GeForce GTX 1070 Founders Edition and running the tests again to see what hashrate increase can we expect from the GPU with the increased operating frequencies. The Founders Edition cards are somewhat limited in the max power you can get, but the good news is that the GTX 1070 FE still has the same 8-pin PCI-E power like the 1080, even though its default TDP limit has been lowered to 151W and the Power Limiter allows for just 12% increase over the default TDP (169W max TDP). There is already a tool for flashing modified video BIOS files available, so now the only thing we need figured out is how to modify the TDP limits in BIOS and other settings such as frequencies and voltages in order to be able to squeeze some additional extra performance over the stock clock capabilities of the Founders Edition cards and even more from the non-reference designs that are already starting to become available on the market.

We already know that the GTX 1080 and GTX 1070 GPUs are handling quite well overclocking and you can squeeze quite a bit extra performance from them if you are not limited and don’t care than much about the power usage. We are trying the GTX 1070 Founders Edition to see what it can do withing its current limits without touching the core voltage and what we got was: Power Limit + 12%, Core Clock + 210 MHs, Memory Clock + 830 MHs, the max settings that are running stable for 24/7 mining on our test card and the results are below. Do note that these can vary from card to card, so you should experiment until you find what works best for you. Regardless it seems that the GTX 1070 FE cards are doing quite well in terms of overclocking in general, so you should expect an nice extra performance boost from them and even more from the non-reference designs.

nvidia-gtx-1070-overclocked-hashrate

The performance increase we get after overclocking the Nvidia GeForce GTX 1070 Founders Edition card with the clock settings above are pretty much consistent with what the GTX 1080 FE shows. The performance boost in terms of hashrate increase in the various algorithms is about 12-14% higher than at the stock settings and better results could be achieved with increase of the voltages, however with that you will also need to be careful that you are fitting in the TDP limit. It is interesting to note that the GTX 1070 FE does perform better on NeoScrypt (668 KHS stock/771 KHS overclocked) than the 1080 FE, but it it still outperformed by the GTX 980 Ti for example. It seems that the slower GDDR5 video memory used here does perform better with the memory intensive algorithms unlike the faster GDDR5X memory used in the 1080, however the GTX 1070 still needs some fixes for NeoScrypt. As already noted the situation with Pascal GPUs including the GTX 1080 and the GTX 1070 is the same for Ethereum mining under Windows resulting in very low hashrates, so while waiting for a driver fix you might want to go for Linux for Ethereum mining on these cards. All other algorithms we have tested besides the not so great NeoScrypt performance are doing well under Windows 7 and 10, so mining for these you don’t need to rely on Linux, especially if you are no good with it.

Optimizing the Power Usage of AMD Radeon RX 480 for Ethereum Mining

29, Jun 2016

rx-480-def-power-limiter

If you have checked our publication Testing the AMD Radeon RX 480 for Ethereum Mining you should already know that at the moment you are a bit limited in your capabilities of reducing the power usage of the new RX 480 GPUs. The pretty much only way to lower the power usage for mining Ethereum with Radeon RX 480 is to use the Power Target reducing it below the 100% value (150W TDP) and thus forcing the GPU to lower its operating voltage and frequency in order to still fit in the allowed TDP. Using a lower operating frequency and voltage for the GPU with the power limiter set at the right value should lower the power consumption, but not change the performance you get while mining Ethereum. Further lowering it below that optimum spot and you will start seeing a drop in mining performance, so our goal here was to find the sweet spot where the hashrate will not be affected.

We have started with the default settings of the RX 480 for both the GPU (1266 MHz) and the VRAM (8000 MHz) and started lowering the Power Target slider until we have reached the 83% – the value at which the power is getting reduced, but the hashrate remains the same. Going below 83% percent of the 150W TDP started lowering the hashrate below the regular value at 100% power. The result was 149 Watt drawn at the wall, taking in consideration the PSU power efficiency that should be about 130W actual power usage. This is down from 177W at the wall with default settings with pretty much the hashrate unaffected and the temperature also went down from 68 to about 60 degrees Celsius.

rx-480-oc-power-limiter

Repeating the same procedure as above, but with the video memory set at 9000 MHz and the Power Target set a bit higher at 85% we saw a very slight decrease in performance with just about 5W extra power usage at the wall (155W). The performance we got with this setting was about 27.8-27.9 MHS and the operating temperature was down from about 72 degrees to about 63 degrees Celsius, so definitely a good improvement. For the moment the only way to reduce the power usage without a drop in performance is by lowering the TDP limit under the default value of 100% and you can easily reduce the power consumption with something like 30 Watt at the wall (if using 80 Plus Gold PSU). You can experiment by further reducing the power target slider under these recommended values, however with the additional power savings you should also start noticing reduction of the hashrate you are getting. Still if you are looking for optimal efficiency you might try going lower and if you already have Radeon RX 480 and use it for Ethereum mining you are welcome to share your settings in the comments below.


top