Re: [ANN] lolMiner 0.42, Update: Aug 17th: Better AMD performance for Equi 144/5
MINEX , AMD HD79x0 CARDS
a little hashrate addition, about 300-500 sol/sec
and the same problem with WINDOWS 7: lolminer042 shows high hashrates (about 12.5 as GF970 under EWBF) and lower real hashrates (about 6-8) with negative diff time by time
driver update to win7-64bit-radeon-software-adrenalin-edition-18.8.1-aug17 take no effect
a little hashrate addition, about 300-500 sol/sec
and the same problem with WINDOWS 7: lolminer042 shows high hashrates (about 12.5 as GF970 under EWBF) and lower real hashrates (about 6-8) with negative diff time by time
driver update to win7-64bit-radeon-software-adrenalin-edition-18.8.1-aug17 take no effect
Yes, the MinexCoin code has some flaws currently, the NVidia more then the AMD. But new code is in the making for 0.43
I'm using 0.42 version | ubuntu 16.04 | rocm driver
lol I'm getting even higher hash rate on --setsclk 6 and --memclk 6 core/memory states.
Maybe rocm driver causing the problems.
Is there anyone here who managed to set miner properly on vega rig and rocm driver ?
lol I'm getting even higher hash rate on --setsclk 6 and --memclk 6 core/memory states.
Maybe rocm driver causing the problems.
Is there anyone here who managed to set miner properly on vega rig and rocm driver ?
ROCm and the ordinary driver are definitive different - at least when I use the rocm offline compiler I get very different register counts. I will see in instruction sets if I see something suspicious. That said: next week a new testing GPU (RX 580) arrives here, then I will set up a Linux box with multiple drivers (The amdgpu-pro, free clover drivers and the rocm), so I can better compare them and see how different they behave. That said, next side hits same target:
I am not installed AMDGPU PRO drivers. I have Radeon RX 560 4Gb.
Unfortunately I have no opportunity right now to test on the free drivers. Usually from the used code it should work on them but I do not know the memory allocation rules on this drivers. Clinfo should never the less work with the free AMD drivers as well, so if you could run it and send me the output in a private message that may help me a lot fixing this
Found out today why the free Clover drivers do not work. Will be fixed in 0.43
Confusing about 8 vega's rig.
Linux, 0.42, "DEVICES" : [0,1,2,3,4,5,6,7] or "AUTO"
Trying start with all GPU used. Miner will work only 1st 5 GPU(0,1,2,3,4), speed for GPU 5,6,7 = 0 and power consumption about 20-30w(idle).
Does not matter what GPU used, miner works only with 5.
But! If i run 2 miners with 2 configs "DEVICES" : [0,1,2,3,4] and "DEVICES" : [5,6,7] all devices works fine with full load and good hashing.
For me it not problem run 2 instances, but anyway...
Any ideas?
Linux, 0.42, "DEVICES" : [0,1,2,3,4,5,6,7] or "AUTO"
Trying start with all GPU used. Miner will work only 1st 5 GPU(0,1,2,3,4), speed for GPU 5,6,7 = 0 and power consumption about 20-30w(idle).
Does not matter what GPU used, miner works only with 5.
But! If i run 2 miners with 2 configs "DEVICES" : [0,1,2,3,4] and "DEVICES" : [5,6,7] all devices works fine with full load and good hashing.
For me it not problem run 2 instances, but anyway...
Any ideas?
Hmm - which algorithm causing these problems, 96/5 or 144/5?
One possible reason could be (virtual) memory size in your rig. Both algorithms use a large amount of memory on GPU. That said the AMD Windows driver has the strange habit to use the same amount of RAM as allocated on GPU when creating the memory buffers. So running 8x the 144/5 algorithm requires the sum of your virtual memory + available (free) RAM to be at least 8x3GB = 24 GB. If it is less the miner is not able to allocate memory for all devices. After some start up phase the drivers free the RAM and thus the required amount gets less. Thats why a 2nd instance started a litte later on the remaining devices may work for some Rigs.
Hello! Can we expect more performance in the near future (rx570,144/5,BTCZ)? I guess the + 80% is also true for it. Congratulations to your job!
I have some ideas left what to try. Not such a big jump again, but I would love to see RX 580 doing 22 sol/s now to jump over 25 ... that may be feasible and I will definitely try to make it possible