RX580´s 70 hash "delay".

[jumpnkick]

I couldn´t help to notice that when mining Bittube (CN Saber) with XMR-Stak 2.5.1 (latest compiled for ethOS) I have a kind of delay (in most GPU´s) to get to it´s Max hashrate. I have 8gb Sapphire Nitros, normal and Special Edition, and though the max hashrate for both is different, what I realise is that they mine perfectly for a while, but 70 H/s slower than their max hash, and suddenly and randomly they jump 70 h/s to their best hashrate, in both cases. The leap in the normal GPU´s is from 434 h/s to 504 h/s, and in the Special Ed´s from 455 h/s to 525 h/s. Those numbers are not exact en every case, but almost around that order. What is pretty consistent is the +/- 70 h/s leap in every 580 GPU I have. Also the time/reason (if there is one) it takes to do that leap, seems to be random. A few hours in some cases, a few days in other, and never in other cases. What am certain of, is that when rig is rebooted all SE GPU´s start mining at +/- 455 h/s and normal ones at +/- 434 h/s, every time. I read on ethOS chatroom, that everybody on RX580´s are seeing this "issue", and ethOS support, say it´s probably a miner bug/issue. In any case, thanx a lot in advance.

Basic information

• Type of the CPU: Intel I3 (not mining)
• Type of the GPU: Sapphire Nitro RX 580 8gb, and Special Edition

Compile issues

• Which OS do you use? ethOS

Issue with the execution

• Do you compiled the miner by our own?

Stability issue

• Is the GPU overclocked? Yes
• Is the Main memory of the GPU undervolted? Yes

[psychocrypt]

please look to the 1min and 15min values. The 10sec values can jump a few persent. The reason is the discretisation of the results. If you teduce your intensity by e.g. 2 which means the miner is faster to compute a chunk of nonces the gap will be smaller.

[jumpnkick]

Jejejejej that would have been a basic mistake, Though am not king of mining, or programming, that is not the problem. Actually that happens in my long term results (15 minutes) and if that was the reason, I would be seeing a random variation of results mining faster or slower for short periods,and the variation would be much less than 70 h/s. Attached there's an image of my results on the ethOS panel, and maybe you will get what I mean (sory if I´m not clear enough). The reason isn't either "silicon lottery", cause if that would be the case, I would see consistent results; A certain GPU would mine at a certain max speed, regardless of the time mining, or the time since the last reboot. What I do actually see, is that every GPU I have, mine 70 h/s slower after a reboot, and it takes a random period of time (though most of them take between a few hours to a few days), for them to mine at their max speed. which is always around 70 h/s faster. What´s more, after a reboot some GPU´s never get to make the final 70 h/s leap to their max speed, and this is random too, since maybe on a previous reboot (or later) thay actually where able to do so. Hope that make´s my question clearer. And thanx a lot for the answers, and the great mining soft!

[jumpnkick]

A comment on the picture in the previous post: Take into account ethOS panel just reports the hashrate of only one of the GPU's threads, so the real mining speed doubles the one on the picture. eg: 525 h/s is actually 1050 h/s...

[jumpnkick]

Maybe this make´s it clearer. This is a single rig's hashrate in time graph. There it can be seen clearly what I mean. This is a good case, where every GPU made the 70 h/s leap to it´s best performance. In some cases, days go by, and they never do the leap. And the strange thing is that GPU´s that on a previous reboot easily got to their final speed, on another reboot wont.

[Spudz76]

Acts like power management, or thermal slowdown, watt shortages, or etc. GPUs deciding to work slightly slower on their own do not occur by anything this app does

[jumpnkick]

Spudz, if that was the case, I would see the same behaviour mining Ethash, or any other algo. And that doesn't happen at all, on other algos.

[jumpnkick]

Please nobody take this post as criticism. I love the miner, and asume I might be doing something wrong. Just trying to figure out every possible way.

[Spudz76]

Different algos are different work, can't be compared like that.

That's like saying my car works fine until I drive on hills and then it overheats. Well yeah, different work was provided to it. But you're still "just driving".

[jumpnkick]

Yap, am sure about that. But still it's not the reason. If the case was that the "GPU's are deciding" as you say, at same temp, clock, volt, etc. etc... I would see similar behaviours, regardless of the algo, the miner soft, and everything else. And at the time, I only see this behaviour with xmr-stak at cn saber and cn heavy. I left to try with cn V8. But thanks for the answer anyway spudz.

[psychocrypt]

It could be that this is a temp issue. Some algorithm working more heavy on the gpu main memory. But all I am not sure what it is.

[jumpnkick]

No worries. It works great anyways. I can tell you is not a temp issue cause I haven't seen a single GPU over 65 ° C. Also, being that the case, I would see them lower hash rate when a certain temp is reached, and the only thing I been seeing, I GPU's working like clockwork at a certain hash rate, and seemlesly suddenly and randomly jump 70 h/s up.