P.S. Phoenix, when we can wait 4.2 version?
A few weeks. We are waiting to be sure that Radeon VII kernels will work properly (the cards are still not here and won't be for another week), as well for some stability tests for the new hardware control features for AMD and Nvidia cards. Too long wait... Maybe you can release 4.1 version with AMD newest drivers support? |
|
|
|
|
Will miner support c29 and c31 for AMD cards? Nicehash?
|
|
|
|
How you set memory voltage to 600mV on RX580? It's minimal 750... I tried to set less than 750mV but it really stay at 750mV.
Mem voltage in the profiling tools is not really mem voltage - it's just a floor for core voltage (it's really for the SOC / mem controller.) So unless you're planning on trying to run your core at 600, there's no use setting mem that low. Simply make sure it's below core voltage, and it's basically unused. If you really want to change your mem voltage (which is about 1.6v) you need to get into the bios. Of course I was mentioned memory controller voltage not memory itself... But I can't set any voltage lower 750 mV on it. If I set lower it stays 750. |
|
|
|
Boy, dumping the memory down to P0 saves A LOT of power.
RX 580, at 1240/1750 does 49.6Mh/s at ~100W per GPU (as per GPU-Z).
Rig has 3x 580, 1x570 and was doing ~530W at wall.
Entered memory P0 state, and lost ~20W per 580 according to GPU-Z. Wall power consumption dropped from 530 to 440W.
Performance dropped with ~0.6MH/s per card.
Best performance/power consmption seems to be around 1200-1250Mhz.
Getting 190.8Mh/s for 440W at wall with 4 cards.
RX 570 - ~45.9MH/s
3x RX 580 - ~49.08MH/s.
Core clocks are all 1240Mhz. Core voltage is ~0.875V-0.893V (2x 0.875, 1x0.887, 1x0.893V).
Raising core voltage to sustain 1280Mhz core raises performance to 50.45MH/s per RX 580, but raises power consumption by ~30W (from 440 to 470). So, 10% power increase for 1% performance increase.
Yeah this algo isn't about memory it seems. I also dumped my Vegas at 500 MHz and RX series at 300 MHz. No big difference in hashrate. My rig have the following cards; 3 x RX 560 4 GB 18 MH/s 4 x RX 570 4 GB 39 MH/s 2 x RX 580 8 GB 43-44 MH/s 2 x Vega 56 85-86 MH/s My total rig wattage is exactly 1000W at 470 MH/s. (Lower power consumption than at CryptoNight V8!). Because I have quite expensive electricity I have quite aggressive undervoltating, 820 mV at core for RX 570/580 and memory all down to 600 mV. The Vega's have like 880 mV at core and 800 mV at memory (can't set lower because restriction in drivers apperantly. At this rates the rig is rock solid stable, havn't crashed one single time since like the fork of Vertcoin. Till now, with my expensive electricity 0.1268$ kWh it have been profitable, like 1.5$-0.3$ (yesterday), but not anymore. So now the fun is over  How you set memory voltage to 600mV on RX580? It's minimal 750... I tried to set less than 750mV but it really stay at 750mV. |
|
|
|
Did you changed -gt parameter? Use auto-tune process.
I have -gt parameter set in config. upd. as it turned out with new version of miner optimal -gt value has changed, it used to be 138, now its 187. So hashrate became the same. That's what I wrote about... P.S. Phoenix, when we can wait 4.2 version? |
|
|
|
There is a performance loss with new version on R9 Fury cards: its only 28M\s now, when it used to be 30M\s, power consumption also dropped by 5-7Wt. Had to roll back to 4.0b.
Did you changed -gt parameter? Use auto-tune process. |
|
|
|
|
What about Cuckaroo29 for nicehash with AMD cards?
|
|
|
|
Sry guys for not responding last days. I had some technical issues with Nicehash and also had to dive deeper into the driver issue on Windows. But also was suffering from flu that made my capacities lower then I had hoped to. Anyways, now its there: lolMiner 0.7 alpha 5You can find it here: https://github.com/Lolliedieb/lolMiner-preview/releases/tag/0.7_a5On the news list: - Fixed a bug with many drivers that made the miner run slower then alpha 2 (especially Windows) - Added Nicehash support - Added a switch to turn on / off TLS (--tls 0 in cmd or “TLS” : 0 in config file to turn if off for BEAM). The default will be coin dependent and is ON for Beam. - Some small mining kernel change. Minimal better performance - Added --help command. Still needs some proper formating, but it works - Some internal bug fixes Have fun mining  Thanks! Nicehash beam normal mines... |
|
|
|
Miner needs support for new AMD 19.2.1 drivers.
4.1c with -clKernels 3 didn't start at all.
We are adding support for 19.2.1 in the 4.2 development version, for now please don't upgrade to 19.2.1. We have also resolved the problems with fan control with the latest AMD drivers (19.x), which will hopefully also solve any fan control issues with the older AMD drivers, which are experienced by some of the PhoenixMiner users. Thanks! Waiting for 4.2... |
|
|
|
|
Miner needs support for new AMD 19.2.1 drivers.
4.1c with -clKernels 3 didn't start at all.
|
|
|
|
|
Will miner have nicehash support?
|
|
|
|
Also, the kernels are designed to handle a downclock of mem all the way down to 300 MHz on Polaris cards and 500 MHz on Vegas, this without hashrate being affected. It probably won't save more than 1-2W per gpu, but always something.
Speed with memory on 300MHz is slower than on full speed... About 1 MH on my RX 580 8Gb YMMV, the 580 in my test rig handles it fine. It's such a tiny power save anyway, just find the spot where you don't lose perf, I've seen people putting it around 1000-1100 MHz, for example. Out of curiosity though, is this your 580 driving monitor(s)? No, to this card not connected anything... With fullspeed memory hashrate is 53MH, with 300MHz memory - 52,2MH. Core clock and voltage the same... |
|
|
|
Also, the kernels are designed to handle a downclock of mem all the way down to 300 MHz on Polaris cards and 500 MHz on Vegas, this without hashrate being affected. It probably won't save more than 1-2W per gpu, but always something.
Speed with memory on 300MHz is slower than on full speed... About 1 MH on my RX 580 8Gb |
|
|
|
After 2h5m no rejected  impressive! Thank you! After filtering for a few random rejected shares due to new jobs there shouldn't really be any rejects though, so we're just where things should be . Also, the kernels are designed to handle a downclock of mem all the way down to 300 MHz on Polaris cards and 500 MHz on Vegas, this without hashrate being affected. It probably won't save more than 1-2W per gpu, but always something. But on my 580 cards memory not 300MHz but on full speed. How I can make downclock of memory? |
|
|
|
clock 1366 mem 1750 undervolt -120
What power consumption? |
|
|
|
RX580 4GB Pulse (Elpida) stock bios strap 1500+ and undervolt -120
54.4 MHs lyra2rev3
Slight performance improvement on 0.3.10 release 54.4 to 54.7 What core clock and voltage you're use? |
|
|
|
After adding --legacy miner start to work on one of two GPU Very interresting, so i may find the bug with no need to reproduce. I suspected the sync (that --legacy disable) to be able to cause such problem. Thanks! Thank you for feedback! I will use legacy mode until you find solution. |
|
|
|
@Unclwish: i mined the same way as you (Nicehash Heavy on GPU 0, one thread) and it worked for me. Sad i couldn't reproduce the bug. What happens when you add parameter --legacy ?
After adding --legacy miner start to work on one of two GPU. But if remove --legacy no 1st or 2nd GPU didn't work separately. After connecting to nicehash attrib.exe close with application error. Bug is not for 1st GPU - it happens on any one of two GPU... Windows 10 x64 latest 19.1.2 drivers. It was on 19.1.1 and on b15,b16 versions too. Start string is: jce_cn_gpu_miner%BITS%.exe -o %POOL%:%PORT% -u %WALLET% -p %PASSWORD% %SSL% %* --nicehash --retrydelay 5 --variation 5 --forever -c config1st580Heavy.txt --no-cpu |
|
|
|
Please, deers, die with your nicehash Please, die without nicehash and shut up, please Author when cuckaroo29 and cuckaroo31 nicehash support would be added? On cuckaroo29 receive error from miner that requsted solver cuckaroo29 not found... |
|
|
|
How to lower intensity for lyra2rev3? I get 52 MH/s on RX 580 8Gb. But power consumption is too high. About 140-150W. It's need intensity option to lower power cons.
Hi! You need to control power draw (and perf) using your clocks, it's just the correct way to do it. Intensity is such a bogus idea for AMD cards for compute algos that doesn't allocate much mem. You're asking for a way to lower the global size of the scheduled jobs to the point where the card runs much less efficiently because of scheduling overhead and execution micro-variance having a larger effect on the overall hashrate. However, that also increases PCIe chatter and host-side load (that is tiny though). Hence, the proper approach is to tweak clocks and voltages until you end up with a power draw profile you're comfortable with. For CN, we provide ways of tweaking "intensity", but that's a very different ballgame. I allready lower clocks and voltages, but on 1 of my RX 580 GPU's is monitor connected and to work it normal I forced to set core voltage higher about 50 mV. Lower than 0.975V leads to BSOD with thread stuck... And on this voltage card consumes about 135-150W on lyra2rev3... connect your monitor to iGPU, or don't connect it at all, and lower the voltage then. I didn't have iGPU (FX8320). And I need connected monitor to work... |
|
|
|
|
Show Posts
