Author

Topic: XMR-Stak Nvidia (pascal) memory limit? (Read 87 times)

sr. member
Activity: 372
Merit: 250
The road of excess leads to the palace of wisdom
June 08, 2018, 01:11:44 AM
#1
Sorry if I double posted this, it wasn't clear to me if it was posted or not the 1st time I tried....

I know that probably most people using XMR-stak are probably AMD miners, but I am one of the crazy Nvidia miners who uses XMR-stak to mine Loki (cryptonight-heavy) currently. This is mostly because I live in a country with very expensive energy costs, so mining one of the cryptonight variations, or Lyra2z sometimes is one way I try to lower my power consumption. I chose Loki because I have always like Monero and Dash and Loki seems to combine elements of both nicely. I'm mining to hold and hope for the future success of Loki rather than just for immediate profit.

I have a question that I hope another Nvidia miner can help me with:

in the tuning info for XMR-stak, it states that "The memory limit for NVIDIA Pascal GPUs is 16 GiB if the newest CUDA driver is used."

I am using mostly 1070s, and a couple of 1080s with OhGodAnEthLargementPill running to help the utilization of their GDDR5x memory. I get around 650-700h/s from the 1070s and similar with the 1080s, or if I don't run OhGodAnEthLargementPill the 1080s only manage 550-600h/s.

Around 700 seems to be the highest a 1070 can do if I do some Google searching and read about what other miners are posting, but in the config file nvidia.txt for xmr-stak - it seems I am still limited in how much memory I use. I changed the "threads" and "blocks" numbers after reading a post somewhere to T 96 B 15 and I have also tried T 48 B 30 and it makes little difference. Those numbers seem to do better than what XMR-Stak defaulted to, but that is still only 1440 megabytes if I'm not missing something. If I try, for example T 96 B 30 (2880 megabytes) it crashes as soon as it tries to spin up that GPU. Is T 96 B 30 just crazy or am I being memory limited to under 2 gigabytes?

When the XMR-Stak notes say "newest CUDA driver" I guess that means 9.2, so will I need to install the latest CUDA SDK on this rig, or is the newest Nvidia driver enough? This is running on a Windows 8.1 machine so maybe I have to be running Win 10 for the "newest"?

Does anyone know if utilizing more memory on the GPU will even help the hash rate or will it just take longer and therefore produce more stale shares?
Jump to: