Topic: PhoenixMiner 6.2c: fastest Ethereum/Ethash miner with lowest devfee (Win/Linux) - page 325. (Read 784965 times)

AMD released 19.7.4 driver... In evening I will try them...

EDIT: Thank heaven! In 19.7.4 AMD didn't touched OpenCL. Miner shows 19.7.3...

Thanks again. You helped to rise to 32,5 Mh/sec

I can share my .bat. But what is "lock" values?

PhoenixMiner.exe -pool pool -wal wal -pass x -worker w1 -stales 0 -cclock 1235 -cvddc 856 -mclock 2250 -mvddc 800 -gt 36 -mi 14 -log 0 -amd -tstop 75 -tstart 72 -clKernel 3 -clNew 1 -leaveoc

It's for Sapphire RX 580 8 Gb SE with hynix MJR memory.
-clKernel 3 - is for double kernels. Not for 4Gb cards...

P.S. -gt parameter is individual for each card. It depends on clocks and timings. Not depend on -REF. On -ref 5 and -ref 50 best -gt parameter will be the same. Best way to find optimal -gt value is to use autotune by "z" key during mining. It can last 5-10 minutes to auto find best value. Just be patient. When value will be found you can see it in stats by "s" key and add it in .bat file.

Sorry, but I can't do that. They're not my...

Thanks a lot, friend) Could you share your additional parameters in .bat and lock values? Thanks in advance)

good timings

Just obtained couple of nitro+ with 4gb trying to get the sweet spot with --REF value, looks like 90 is a  good startso we will see...wonder if same goes for samsung memory

P.S. rx570 doesn't like --REF 90

What kind of timings? Can you publish here?

I forgot - I use vbios with flashed ETH optimized timings...

Just tried - set REF to 5 (default) - 32,9MH, REF 90 - 34,2MH
Speed didn't change immediately, just need 10-20 seconds...

Just tried ref 90 on 4gb hynix Sapphire got a small bump to 32,6

guess 4gb cards cant go more than that 

I get the same speed on any drivers allready about 2 years.
On phoenix I have a bit more speed than Claymore. If I use double kernels (2x more vram needed) I got additional 200-300 kH.
Phoenix allways faster Claymore with the same core and memory clocks even on usual kernels. If you get lower speed on Phoenix, you forget to find optimal -gt parameter values for your cards.

With use of AMD Memory Tweak Tool it's possible to get additional 500-1000 kH of speed.
One of my RX 580 8Gb cards - Sapphire with hynix MJR memory gives on Claymore 33,5-33,7MH/s with -REF 90 on AMD Memory Tweak tool. And 34,2 MH/s on Phoenix with double kernels. About 34MH/s with usual optimized kernels.

... and what hashrate do you have on 19.7.3? On bc drivers for my RX480 8Gb I've got 31,5 Mh/sec. With Claymore - 32,5 Mh, but more power from the wall.

I use latest 19.7.3... But I have RX 580 8Gb. For 4Gb cards better driver that uses less memory...
If you have Windows 10 1903 use at least 19.5.1 driver. If not try 19.4.3.

Thank you very much. So what best driver for rx 570?

Phoenix wrotes: "This works only on Windows and with relatively recent drivers (from the last 6-8 months)"
Your 18.5.1 drivers is too old - allready more than year past...

Dear PhoenixMiner,
I can not use -mt option. How to fix it?
https://s5.upanh.pro/2019/07/28/AF72208E-8E87-4A6A-95E4-BE9A690FE585.png

@PhoenixMiner,

Perfect, now works on all my cards.

Just one thing, the extra file (PhoenixMiner_NVRTC_Windows.zip), can you put it in the same download location for those that don't know where to find it.

Thanks.

  The "bug" is that you did not change the wallet address in epools.txt to your own. It was already discussed several times in the last few weeks. In the new version, the example epools.txt file is named epools_example.txt to avoid similar problems in the future. If you want to use epools.txt, just rename epools_example.txt to epools.txt and put your own wallet address there.


  No, not at all, it was just a warning that trying -bench 380 or above on AMD card will fail even it has more than 4 GB of VRAM. We will get there after about 16 months, so either AMD will release fixed drivers by then that allow allocation of more than 4 GB, or we will implement a workaround, using more than one buffer.

  We are planning to prepare a mini-guide but here is the short version:
  
    1. The overclocking and fan control options now work with Nvidia cards under Windows. Since Nvidia cards actually accept only an offset from the base clocks, you should specify e.g. -cclock -300 if you want to decrease the core clocks by 300 MHz, and -mclock +400 if you want to overclock the memory by 400 MHz. Note that you still can specify absolute values (e.g. -cclock 1400 or -mclock 4200) but we will get the default base clocks as reported by the card and then apply the difference as clock offset anyway.

    2. Note that in most cases Nvidia cards work in P2 state instead of P0 state when mining. This isn't a problem except that the default memory clocks are usually lower in P2 state (for example 3800 Mhz instead of 4000 Mhz in P0 state). This means that you will need to apply higher memory offset to achieve the desired memory clocks. In the previous example you will need to apply -mclock +600 to get to memory clocks 4400 Mhz because the base clock is only 3800 in P2 state. In order to see the actual core and memory clocks, you should use the new -hstats 2 option which will provide core and memory frequencies, voltages, and P-state for each card (if reported by the driver).

    3. Nvidia cards doesn't allow meaningful changes to the core or memory voltages. So -cvddc and -mvddc will have no effect on Nvidia cards.

    4. The best strategy of setting up the Nvidia cards is to avoid lowering the core clock via -cclock but instead use the -powlim option to lower the power limit of the card. Start with -powlim -50 to set the power limit to 50% and see if the hashrate is lower. The reason for this is when core clocks are lowered, most Nvidia cards doesn't lower the core voltage, so the card continues to use the same power as before.

    5. As for the fan settings, they work similarly on Nvidia and AMD cards. You should set the desired temperature with -tt (e.g. -tt 65 to set 65C as desired card temperature). PhoenixMiner will control the fan speed, trying to keep the cards temperature at 65C. You can also set the card max temperature with -tmax. At this temperature the card will start to lower its core clocks (i.e. to throttle) to avoid overheating. By default these temperatures are quite high (usually over 80C) so it is a good idea to set it to something about 5C higher than the -tt temperature (e.g. -tt 65 -tmax 70). If the fans can cool the cards enough so it won't reach 70C, it will mine with the same speed. You can also use the -fanmin and -fanmax to set the range of allowed fan speeds in percent. For example a good values are -fanmin 20 -fanmax 60, which will keep the fans no lower than 20% and no higher than 60%. If the fans hit 60% but are still unable to keep the card temperature under -tmax, the card will start to throttle to keep the temperature at about 70C, which will lower the hashrate but will avoid overheating the card and will avoid excessive wear of the fans.

    6. Note that all options (-cclock, -mclock, -cvddc, -mvddc, -tt, -tmax, -fanmin, -fanmax, -powlim) can be applied either on all cards at once or for each card individually. For example you have five cards and want to specify different -fanmax for the first two, use something like this: -fanmax 55,55,70,70,70 which will set the max fan to 55% for the first two cards and 70% for the other three cards.

    7. The power consumption reported by the cards is not always accurate (sometimes it is off by a big amount), so you can calibrate it for more accurate results with the -pidle and -ppf options. PhoenixMiner calculates the total power as follows: P = Pidle + Pgpu * (Ppf / 100). PIdle is the idle power in Watts which is specified by -pidle option, and -ppf is the correction that is applied to the GPU reported power to get the real power.

      For example, lets say that without specifying -pidle and -ppf, PhoenixMiner reports that all GPUs use 540 W. You get a power meter and use it to see the actual power consumption from the wall socket. Let's say it is 650 W. Then press p in the miner console and wait a few minutes until all GPUs stop mining and power levels stop decreasing. Then again note the reported GPU power from PhoenixMiner (let's say it is 50W, and the power meter says 100W from the wall). Then calculate the difference between full power and idle according the power meter and divide it by the same difference as reported by PhoenixMiner. In our case this is (650 - 100) / (540 - 50) = 1.12 This means that your GPUs draw 12% more power from the wall than reported (this includes both inaccuracy in GPu power reporting and the efficiency of the PSUs). In this case you need to use -ppf 112. In our case the appropriate -pidle value will be 100 - (50 * 1.12) = 44 W. So you use the options -ppf 112 -pidle 44. Now PhoenixMiner will report idle total power consumption of 100W and under load 649W, which is much closer to reality and can be used as reasonably accurate figure without having a power meter connected to each rig.

  As you may have different cards, you can also specify -ppf for each card individually but then the calibration will be longer, as you must calculate ppf for each card separately.

  8. If you have calibrated the power reporting with proper -pidle and -ppf values as explained above, you can also also specify the cost of your electricity with the -prate option (in USD per kWh). If we use the previous example and your electricity cost is 0.11 USD per kWh, with the options -ppf 112 -pidle 44 -prate 0.11, and with reported total power usage 649 W, PhoenixMiner will show you that you are paying approx 1.71 USD per day for electricity.

  9. The -mt option is only for AMD GPUs and is intended as quick and dirty replacement of VBIOS modding. -mt 0 (the default) uses the VBIOS memory timings (which will be quite slow if the card is Polaris based and isn't modded), -mt 1 and -mt 2 are faster. Note that changing the -mt value while mining (by reloading config.txt with the c key from miner console) often leads to crashes. NOTE that properly modded VBIOS or carefully selected memory timings with AMD Memory Tweaker will always be better than using the -mt option.

Thanx!

The release candidate of 4.5 is ready. You can download PhoenixMiner 4.5c from here:

(MEGA links are no longer active)

If you want to check the integrity of the downloaded file, please use the following hashes (you need the last file PhoenixMiner_NVRTC_Windows.zip only if you want to mine BCI with Nvdia cards under Windows):

Here are the new features for this release since 4.5b:

• Added support for the latest AMD Windows drivers 19.7.3
• Fixed issue with ProgPOW BCI mining on Nvidia RTX20x0 and GTX16x0 cards

The changes in 4.5b since the previous official release (4.2c):

• Changes in Nvidia kernels for more stable work and better utilization of the GPU
• Implemented ProgPOW BCI mining for Nvidia cards too. Under Windows you must download and unpack the file PhoenixMiner_NVRTC_Windows.zip in the same folder as PhoenixMiner.exe in order to mine BCI with Nvidia cards.
• Added support for all new AMD Windows drivers up to the latest 19.7.2
• Added support for all new AMD Linux drivers up to the latest 19.30-838629
• Added hardware control and monitoring (clocks, voltages, fans, etc.) for Nvidia cards under Windows. Note that you should specify the clocks for Nvidia cards relative to the defaults (e.g. -mclock +400 instead of -mclock 4400)
• Added hardware control and monitoring (clocks, voltages, fans, etc.) for Radeon VII cards (Windows only)
• Added advanced hardware monitoring (specify -hstats 2 to activate it). It will show you the GPU core and memory clocks, voltages, and P-states (if supported)
• Added GPU power consumption if the driver reports it. You can also specify the idle power (-pidle), PSU/GPU efficiency (-ppf) and electricity cost (-prate) and have the miner calculate the daily electricity costs for you. If you are using third-party mining software which doesn't expect the power consumption figures in the miner's log, you can disable it by specifying -hstats 0
• Added new parameter -mt to set the memory timings on AMD cards. -mt 0 is the default, which uses the VBIOS memory timings. -mt 1, and -mt 2 use predefined memory timings that are independent from the VBIOS timings. This option is useful for mining with RX580/570/560/480/470/460 cards without modding the VBIOS. This works only on Windows and with relatively recent drivers (from the last 6-8 months)
• The bench mode now works for dual mining and ProgPOW. The algorithm is determined by the -coin and -dcoin parameters. E.g. -bench 10 -coin bci will bench the BCI ProgPOW, and -bench 10 -dcoin blake2s will bench the dual mining
• Increased the maximum supported DAG epoch to 450. Note that AMD drivers don't allow buffers larger than 4 GB so the biggest possible DAG epoch with current AMD drivers is 378
• Added the ability to read the "FriendlyName" from Windows registry and use it (if available) instead of the generic card name from the driver
• Fixed the devfee pools for some of the alternative coins like PIRL and others where many pools either disappeared or are using old pre-fork versions of the clients and are mining on the wrong blockchain
• Other small improvements and fixes.