TheMinero
Newbie
Offline
Activity: 6
Merit: 0
|
|
May 29, 2018, 11:52:20 PM |
|
Didn't know that MUSIC it's multi-algo, but I prefer CN algo because low: heat and wattage I'm using xmr-stak but I it does not have the gpu-thread option like SGMiner, the OP said that increasing that option to 2 gave a little more h/s, appart from the bios mod I have not played with the xmr stak options, I just let it by default and in fact the cards are 1000/1750 since I'm working so not much time in home to check, only with teamviewer BTW what it's your hash in ethash, cn-v7 and cn-heavy?
|
|
|
|
MagicSmoker
|
|
May 30, 2018, 10:30:23 AM |
|
Didn't know that MUSIC it's multi-algo, but I prefer CN algo because low: heat and wattage I'm using xmr-stak but I it does not have the gpu-thread option like SGMiner, the OP said that increasing that option to 2 gave a little more h/s, appart from the bios mod I have not played with the xmr stak options, I just let it by default and in fact the cards are 1000/1750 since I'm working so not much time in home to check, only with teamviewer BTW what it's your hash in ethash, cn-v7 and cn-heavy? No, I meant that I use my 2GB (16CU) RX 560s for both Ethash and CN, but for Ethash I am restricted to MUSIC or UBQ because the DAG files still fit into 2GB. Right now they are mining MUSIC and doing 14.3MH/s per card, which is a little over half what a 4GB RX 570 will do, but at much less than half the price (at least that was the case when I bought them back in Dec/Jan). As for the CN forks, I use xmr-stak 2.4.2 because it performs the best so far and is very stable. It does not perform well "out of the box," so to speak - the config files it generates (amd.txt and cpu.txt) need to be heavily modified. My amd.txt file modified for just 2 cards is shown below. If you have more than 2 cards then increment the index parameter for each additional card but copy everything else (assuming they are identical cards) and put those lines before the ], at the bottom. "gpu_threads_conf" : [ // gpu: Baffin memory:1664 // compute units: 16 { "index" : 0, "intensity" : 416, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2, "comp_mode" : true }, { "index" : 0, "intensity" : 416, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2, "comp_mode" : true }, // gpu: Baffin memory:1664 // compute units: 16 { "index" : 1, "intensity" : 416, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2, "comp_mode" : true }, { "index" : 1, "intensity" : 416, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2, "comp_mode" : true }, ],
I get about 495H/s per card with this congfiguration in CN-v7; I have not tried the other forks because I am not interested in any of the coins that use them.
|
|
|
|
TheMinero
Newbie
Offline
Activity: 6
Merit: 0
|
|
May 30, 2018, 09:54:20 PM |
|
No, I meant that I use my 2GB (16CU) RX 560s for both Ethash and CN, but for Ethash I am restricted to MUSIC or UBQ because the DAG files still fit into 2GB. Right now they are mining MUSIC and doing 14.3MH/s per card, which is a little over half what a 4GB RX 570 will do, but at much less than half the price (at least that was the case when I bought them back in Dec/Jan). As for the CN forks, I use xmr-stak 2.4.2 because it performs the best so far and is very stable. It does not perform well "out of the box," so to speak - the config files it generates (amd.txt and cpu.txt) need to be heavily modified. My amd.txt file modified for just 2 cards is shown below. If you have more than 2 cards then increment the index parameter for each additional card but copy everything else (assuming they are identical cards) and put those lines before the ], at the bottom. "gpu_threads_conf" : [ // gpu: Baffin memory:1664 // compute units: 16 { "index" : 0, "intensity" : 416, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2, "comp_mode" : true }, { "index" : 0, "intensity" : 416, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2, "comp_mode" : true }, // gpu: Baffin memory:1664 // compute units: 16 { "index" : 1, "intensity" : 416, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2, "comp_mode" : true }, { "index" : 1, "intensity" : 416, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2, "comp_mode" : true }, ],
I get about 495H/s per card with this congfiguration in CN-v7; I have not tried the other forks because I am not interested in any of the coins that use them. Cool! Allright I see where is the gpu_threads in xmr-stak, so each line represent a gpu thread, a gpu can have more than one thread I will test it with CN-V7, CN-heavy works differently, I tested using two threads per gpu and it crashed so I need to do some tests here What are your clocks? do you have moded bios? Thanks
|
|
|
|
MagicSmoker
|
|
May 30, 2018, 10:59:05 PM |
|
Cool!
Allright I see where is the gpu_threads in xmr-stak, so each line represent a gpu thread, a gpu can have more than one thread
I will test it with CN-V7, CN-heavy works differently, I tested using two threads per gpu and it crashed so I need to do some tests here
What are your clocks? do you have moded bios?
Thanks
Generally speaking, it is better to drop the intensity value for each thread for heavy rather than go back to 1 thread per GPU (as you've already tried). I use a binary search method to find optimal values in the shortest time. For example, if intensity ranges from 0 - 1024 I will try 512 first and if the card hashes I try 768 next; if it crashes I try 384 next; etc and so on. I am running a modified BIOS on the cards - a "Powersave ROM" from Anorak specific to my card brand/model - and custom voltage profiles. Max core clock is 1200MHz at 1025mV and max mem clock is 1900MHz at 1000mV. These values almost certainly won't help you, but I will say that OverDrivenTool is far better at this sort of thing with AMD cards than MSI AfterBurner, and pretty much the only option if you have 4 or more cards (MSI AB breaks if more than 3 cards are present).
|
|
|
|
TheMinero
Newbie
Offline
Activity: 6
Merit: 0
|
|
May 31, 2018, 09:48:34 PM |
|
Generally speaking, it is better to drop the intensity value for each thread for heavy rather than go back to 1 thread per GPU (as you've already tried). I use a binary search method to find optimal values in the shortest time. For example, if intensity ranges from 0 - 1024 I will try 512 first and if the card hashes I try 768 next; if it crashes I try 384 next; etc and so on.
I am running a modified BIOS on the cards - a "Powersave ROM" from Anorak specific to my card brand/model - and custom voltage profiles. Max core clock is 1200MHz at 1025mV and max mem clock is 1900MHz at 1000mV. These values almost certainly won't help you, but I will say that OverDrivenTool is far better at this sort of thing with AMD cards than MSI AfterBurner, and pretty much the only option if you have 4 or more cards (MSI AB breaks if more than 3 cards are present).
Yes, Heavy it's a lot different than V7, need to do a lot of test to get the proper config The other thing is the driver, now I have 17.4.4, I installed 18.4.1 and one of the three cards was giving half h/s, I quickly uninstalled and installed again 17.4.4 and it's running fine I use MSI AB to set 1000/1750@0.800 cards and does not exceed 60º, I have other rig with 4 rx 560 also with MSI AB and sometimes this rig gives problems, when I have to reboot it, sometimes does not boot, but I really don't know if there is video signal because I only manage it with teamviewer, for now I can't connect a monitor, so I guess this can be MSI AB problem since this rig has 4 cards, well hope so I downloaded OverDrivenTool but this tools need 17.7 driver or newer, I have not tested blockchain driver and I feel that can be a PITA to get a good driver for my setup
|
|
|
|
MagicSmoker
|
|
May 31, 2018, 11:09:37 PM Last edit: June 01, 2018, 12:45:16 PM by MagicSmoker |
|
... The other thing is the driver, now I have 17.4.4, I installed 18.4.1 and one of the three cards was giving half h/s, I quickly uninstalled and installed again 17.4.4 and it's running fine ...
I'm running the latest AMD drivers - 18.something - with no issue. One peculiar thing I've noticed is that I always have to reboot twice to max out the hashrate of all the cards; if not then 1 or more cards invariably get stuck at a lower hashrate for some reason. If you do flash a modded BIOS make sure you get the infamous ATI pixel clock patcher program otherwise Windows will cripple the cards that have been flashed (IIRC, still usable in VGA mode to display a screen, but not at all for mining - you can tell a card has suffered this fate if it has error code 43 in the control panel listing).
|
|
|
|
|