Re: [ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX]

Quote from: bathrobehero on July 28, 2017, 12:50:29 AM

Quote from: 49erminer on July 27, 2017, 08:43:17 PM

Hello everyone,

I'm looking for suggestions on how to increase acceptance rate and effective hash rate for myriad-groestl mining with ccminer.

I have a mining rig with 3 older AMD cards and 2 new GTX 1070s. According to ccminer, the 2 1070s are hashing myriad-groestl at 162.x Mh/s which is 140% of the 115.x hashrate of the 3 AMD cards as reported by sph-sgminer-x11mod running with an updated myriad-groestl kernel.

The pool reports that both miners are mining at the same difficulty in the following screenshot, so I would expect to see the 1070s getting somewhere around 140% of the shares of the AMD cards. I'd also expect to see the effective Hashrate reported by the pool for the cuda worker to be closer to 160, I've never seen it go above 60.
http://imgur.com/pPivYDh

Actually, the 1070s are getting less than half of the accepted shares of the amd cards. Here's a screenshot of both miners, started at the same time, after a 14h:42m run:
http://imgur.com/0i23gg9

I've tried ccminer 2.0 x86, x64, 2.1-tribus, and 811. All exhibit the problem of reporting a high hash rate but finding fewer shares than the slower AMD cards hashing at the same difficulty for the myriad-groestl algorithm.

What sort of tuning might be done to ccminer or the cards to increase the effective hashrate in the pool and/or increase the rate of share acceptance in the miner? Or, is this a problem with the pool?

I'm only seeing this problem with ccminer/myr-gr. When mining equihash with different mining software (EWBF), the 1070s are also hashing around 140% faster than the AMD cards, and the effective hashrate in the pool and the number of accepted shares reflect the higher hash rate.

Hi, try adding -f 2 to your ccminer bat file or check another pool.

Thanks for the response.
Added "diff-factor" : "2" to the config file and gave it a try.

8.11-Klaus-T quickly found 1 share which was rejected with a reason of "reject reason: low difficulty share of 0.23825932042784703" after which it apparently died. That is, it did nothing for the next 12 minutes, with task manager showing 0 CPU activity for the ccminer task.

The 2.1-tribus version ran with "diff-factor" : "2" in the config, and the effective hashrate in the pool was close to the 151.x hashrate shown by the miner. The problem with 2.1-tribus and the suggested option is the almost 50% rejection rate. After 1h 24m, tribus found 839 shares of which only 437 were accepted, the other 402 shares were rejected with the same "low difficulty share" message. During the same time, the amd cards had 548 accepted and 5 rejected shares.

The suggested option fixes the problem of the pool showing a very low effective hashrate for the nvidia worker, and increases the shares ccminer solves to a level that actually exceeds the % difference in hashrates reported by the two miners, but only at the cost of an unacceptably high rate of rejects.

I'll try a different pool tomorrow, and maybe the 2.0 x86 and x64 versions of ccminer if the pool switch doesn't help. Just out of curiosity, which version of ccminer would you recommend for mining myriad-groestl? I like the fact that 8.11-Klaus-T shows a 10Mh/s higher hashrate than the other miners, but that is only useful if the higher hashrate translates into more shares.

I'll post what I find tomorrow after the pool switch.