FYI, GTX 1080 Ti FE best setting I've found for Windows is --launch=64x56 (the SMX is 28; it's possible to use 72x56 but it's slower and will crash more often).
**xmrMiner** a new high optimized NVIDIA GPU miner
psychocrypt posted this 7 years ago
Dear Community,
there is a new optimized NVIDIA GPU miner xmrMiner available. xmrMiner is a fork of KlausT's ccminer version. ccminer is developed by Christian Buchner's & Christian H.'s and modified by tsiv for Cryptonight mining.
xmrMiner achieves on a NVIDIA Tesla K20 26% more hash/s than ccminer.
Performance Overview
| gpu | launch param | xmrMiner [hash/s] | KlausT ccminer [hash/s] | speedup [%] | clock [MHz] | watt |
|---|---|---|---|---|---|---|
| k80* | 24x39 | 482 | 395 | 22 | 875 | 115 |
| K80* Amazon AWS | 24x39 | 469 | 399 | 25 | 875 | 128 |
| k20 | 24x39 | 397 | 314 | 26 | 758 | 99 |
| P100 | 72x56 | 1640 | 1630** | 0 | 1328 | 92 |
| GTX TITAN X | 16x48 | 633 | 579 | 9 | 1151 | 132 |
* used one of two GPU sockets
Please test xmrMiner and post you feedback on reddit or in this forum. If your GPU is not listed on this benchmark side or you achieve a higher hash rate than currently posted please add your results there.
If you found any bugs please open an issu on github.
download
hashrate psychocrypt
please try --launch=32x60. If you need more interactivity you could also use the option x-bfactor=6 --bsleep=25
You could also try the miner xm-stak-nvidia or latest version In both cases please use always the latest version to get the best performance.
Newbie here... I am running this on my single MSI GTX 1080 Seahawk. Without adding anything to launchParam, I am seeing on average 521 H/s. How do I calculate the launchparams? When launching, it states 'GPU #0: GeForce GTX 1080 (20 SMX), using 80 blocks of 8 threads'. Is that max for this card, or is it generic? It does seem to really bog down my system with this setting. Thanks.
You need to pass --launch=64x40 as parameter to the binary. e.g., xmrMiner --url=... -u ... -p ... --launch=64x40
Your GPU has 20 SMX therefore a good configs could be --launch=64x40, --launch=32x60, --launch=32x80 or --launch=24x40
If you are using windows you need to set the parameter to --bfactor=6 (maybe to 8) and --bsleep=25.
hey psycho, your gpu miner works like a charm, getting 500 h/s from the go with my 1070. i am confused about the performance part of the git. do i type in '--launch=txb somewhere to find the right spec? i'm not sure where to find my multiprocessor count or thread count that my gpu can handle. any help would be great, thank you for your hardwork!
I released a new version 0.2.1 with a better interactivity if the miner is used on the normal dektop pc. For nvidia fermi gpus there is a speedup up to 120% possible.
To use more than 8 GPUs you need to start the miner two times and mask the use able devices. in windows
- start the cmd terminal
- goto the miner directory
- run
set CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 - start the first miner
xmrMiner(do not forget the miner options) - run step 1,2 to start the second miner
- run
set CUDA_VISIBLE_DEVICES=8,9,10,11,12,13,14,15 - start second miner
xmrMiner
In linux use export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
I think you could reach better rates if you use for blocks 2x smx and more threads. Memory should be as near as possible to 1gb or maybe 2gb.
Could you try for the 6smx cards --launch=32x12 or --launch=24x18
Do you perform the tests under linux? If you use windows please wait with the tests for the next release. The next release will allow to run larger launch settings under windows without get killed from windows for blocking the gpu.
thx for this overview. good job
Each socket of the k80 is visible as a independent gpu. If you specified the launch paramter the value is passed to all gpus. The configuration for each gpu is printed during the start of the miner. xmrMiner supports up to 8 gpus which means 4 x k80. If your system contains 8xk80 which means under linux nvidia smi shows 16 gpus than please contact me again and I can explain how you are able to use all 16 visible gpus.
@psychocrypt thanks for this, I had just installed ccminer and after installing yours I see the performance lift as per your table.
I'm running 8x K80 so I used the 24x39, but in your table it says this is best for when only one of two sockets are being used....how do I check what is my config, and what is the optimal parameters if both sockets are being used?
I'm running Windows 2016 server FYI, thanks again!
Tetsted it for several days with standard settings and low cost cards, can try also with Ref Clock and other settings. Playing with max Memory/bxt slowed them down to 50 to 60Hs.
750Ti 256Hs 5SMX/20b/8t/352MB (Core 1462Mhz/ Memory 1450Mhz) x4 PCI Express
950 262Hs 6SMX/24b/8t/438MB (Core 1616Mhz/ Memory 1778Mhz) x16 PCI Express
950 257Hs 6SMX/24b/8t/421MB (Core 1565Mhz/ Memory 1752Mhz) x4 PCI Express
1050Ti 305Hs 6SMX/24b/8t/669MB (Core 1936Mhz/ Memory 2152Mhz) x16 PCI Express