Hi all,
Hopefully this is not too off-topic, but in trying to get the latest ethminer working in nvOC (at the speeds expected for my GTX 1060 6gb cards), I had also been posting on their Github page. I am trying to figure out a few things, and would love to hear what you all have to say:
I appreciate the input, as well as the fact that this build even exists, so any help here would be amazing.
Hopefully this is not too off-topic, but in trying to get the latest ethminer working in nvOC (at the speeds expected for my GTX 1060 6gb cards), I had also been posting on their Github page. I am trying to figure out a few things, and would love to hear what you all have to say:
- Someone on that thread suggested that with Linux-based mining, the overclock settings for memory people throw around (which, most things I read are for Windows) need to be doubled. As in, people usually say Samsung memory can get around an 850 memory OC, so they were saying for Ubuntu I would go with 1700? This felt wrong given the 1bash defaults, but I'd love verification
- Is there any way anyone knows, without doing a Windows install, to learn what kind of memory my cards have? I'd rather not install Windows, and I have 26 cards I'd like to verify without changing their order or disconnecting if possible. I have not been successful in finding a Linux utility, but feel I may benefit from indvidual clock tuning
- Has anyone gotten the 24-25 Mh/s the latest ethminer supposedly gets while using in nvOC? If so, what settings are you using?
I appreciate the input, as well as the fact that this build even exists, so any help here would be amazing.
Hi JudoFlash,
- You are totally right for memory in linux, cause in linux u dont OC MemClock technically you OC MemoryTransferRate. MemClock = MemoryTransferRate % 2
- Unfortunately on linux we don't have software that allow you to get memory brand version, easy way to do this is to boot your rig on a single windows card per card
- For this last one i dont have 1060, but i bet with nice OC u can grab this one. Remember to negative your CoreClock when u mine ETH
Don't hesitate to ask more
Thank you kindly. Wow, ok, so I will try ratcheting up the memory overclock some more. i have been underclocking the core by 200, but I appreciate the tip.
Too bad about the utility to read the memory type. Since instability takes a while to crash things sometimes, I bet I'll be tuning things for a LONG time. Especially since I have already had instability issues with my two ASRock H110 BTC+ Pro boards.
I will try some memory tuning and see where that brings me, though I'd love to hear if anyone else running 1060's has had success, and at what settings.
Thanks!
What CPU are you using with your unstable H110 rigs?
Both are running on Celeron G3930's, with 8 GB RAM.
The one that wouldn't stay up for more than a couple of hours has been up for the past 12+ since I disabled TeamViewer. I have had really poor success overclocking the memory more than 600 or so. Any thoughts?
I would recommend using a more powerful cpu with 13x rigs. Although a rig will run on a g3930; it may not be stable. I am now using g4600s on my 13x rigs. I'll let you know if they show instability; so far they have been stable. Also how are you powering your rigs; is there any difference between the stable and unstable one? I had one 13x rig which was unstable; until i identified the problem was do to too much splitting of pcie cables; I modified and now the rig is stable.
I just placed an order for some g4600's. I will see how that works - the rig that had been crashing every couple hours crashed after 24 hours after I turned off teamviewer - hopefully this puts me over the top.
As for power, both are built the same way. Each is running on 2 750w EVGA G3 power supplies. Each SATA/molex power connector is running no more than 2 GPU risers. The molex connectors (and single SATA) hooked to the board each share with the riser for a single GPU (this means 13 GPU's, 2 motherboard molex, and 1 motherboard SATA to connect. The PSU's each have 3 SATA and 1 Perif (molex) connector, hence my using two connections per cable). I did have a 550w PSU in the mix (as a third) to see if that would change things, but it did not.
The only other discernable different is that not all GPU's are the same brand/make (but all are 1060 6GB), and the risers are not all identical (though I have switched them out in troubleshooting).