Improve performance so it's at least the same as Claymore's miner

[danielocdh]

The performance of ethminer is around 6% lower than Claymore, Claymore uses around 8% more power which seems ok considering the fee

I'd like ethminer to have at least the same or more performance(because of the fee) than Claymore even if that means using the same power as Claymore, it could even be an option (in case there is people that likes the current performance/power usage).

I've been testing claymore and ethminer for weeks now, I'm pretty sure my numbers are accurate

[MariusVanDerWijden]

We would also like to be as fast as Claymore... Under some circumstances we are! If you have an idea how to increase the performance, feel free to submit a PR.

[danielocdh]

@MariusVanDerWijden what circumstances? is that documented somewhere?

[AndreaLanfranchi]

With NVIDIA/CUDA mining we're pretty much the same. Still behind with OpenCL.

[invidtiv]

Sorry @AndreaLanfranchi with NVIDIA / CUDA ethminer is better !!! @danielocdh , overclocking with ethminer is not the same as claymore, you need to push a little more on the gpu clock and watts...

OpenCL :( .... sniff ... Wishing for a good opencl developper to get on board and get some good PR out... In a perfect world OpenCL portion of ethminer would be as good as the NVIDIA/CUDA is...

[StefanOberhumer]

@danielocdh : Comparing miner software is very difficult ! 1.) Don't use hashrate displayed in the miner software - the only thing what counts is hashrate on the pool (maybe some software cheats or does not calculate the hashrate as the other one) 2.) Compare mining softwares: same time (network latency), same pool (same ip-number), same router, same hardware, same temperature(have found my rigs get unstable/loosing hashpower at about 62°C) - That makes comparing the software really difficulty ;-) 3.) Compare it over one week or more ! (stability, reboots,...)

So for this 3 reasons it is really difficult to compare mining softwares ! I can not speak for the OpenCL part as we're using only NVIDIA / CUDA but we switched to ethminer as we decided its better. I also enjoy the opportunity to adapt the miner - the contrbuters of the project are really nice and helpful Come on - get part of it and make ethminer faster ( or as fast as Claymore ;-) ) !

[danielocdh]

@StefanOberhumer I saw consistent results with 2 rigs, different parameters, different pools, same simultaneous pool, switching software, etc. I'm pretty sure my tests were conclusive (at least for my rigs). Claymore always had higher hashrates and power usage (6-8% more). I also use nvidia, do you use any specific parameters/overclock? Thanks

[StefanOberhumer]

@danielocdh - No - I've no specific parameters - but it made for me a difference when I tried 2 rigs or 20+ (whyever ???) Just as info: I've tried a lot of optimizations, in the CUDA kernel part. All current tests resulted in a lower hashrate :-( . I think ethminer is currently on top with NVIDEA/CUDA

[l4Dream]

I can confirm that ethminer have same/better hashrate with nvidia than claymore on both windows and linux.

[AndreaLanfranchi]

1.) Don't use hashrate displayed in the miner software - the only thing what counts is hashrate on the pool (maybe some software cheats or does not calculate the hashrate as the other one)

I believe this statement is gold. Most users keep following absolute instant hashrate as displayed by their miner instead of making long term comparisons on revenues. It's proven that lower hashrates with better stability (no crashes to recover from) increase a lot your revenues and help preserve GPUs.

[ddobreff]

Here is claymore vs ethminer simple 6 hour comparison: http://prntscr.com/k8lwfl - single GPU test 6h SMA - claymore 11.9 - AMD RX560 4GB. http://prntscr.com/k8lwq8 - single GPU test 6 SMA - ethminer latest git - nVidia p106-100 6GB. My personal tests show that ethminer avg and reported is mostly on par, while claymore has small spikes equalizing it from time to time, rest is way below reported.

[SukharevPavel]

I made an improvement for OpenCL mining, which manages to get 3.5% hashrate increase on my RX480 (27.9 vs 27 on default ethminer while claymore has 28.4). But this code changes CLMiner.cpp and kernel file a lot. Also, it breaks default hash counting. I am making a scientific research and want to commit these changes to ethminer to prove my results, but it is rather complicated according to aforementioned causes. Does anyone can to discuss this problem with me?

[ddobreff]

Check latest PR. Proposed faster kernel which exceeds claymore performance by a bit.

[newtoopensuse]

@ddobreff. You mean the pr made by goobur to libethash-cl?

[ddobreff]

Yes. Its already merged. Further optimisations and adjustments are expected to follow.

[newtoopensuse]

Thanks for the info ddo. But which version of ethminer will it end up in? And I see Zawawa’s kernel is going to be used. This must be super exciting.

[SukharevPavel]

I made an improvement proposition, let's discuss it: https://github.com/ethereum-mining/ethminer/issues/1366

[smurfy]

I'm closing this, with the new binary kernels in 0.16 ethminer is as fast or even faster than claymore on OpenCL

[ghost]

It actually makes me feel quite good to learn that my kernel is outperforming Claymore's. Good work. guys!

[AndreaLanfranchi]

@zawawawa Thank YOU !!!