PhoenixMiner 6.2c: fastest Ethereum/Ethash miner with lowest devfee (Win/Linux)

[laik2]

About the AMD Linux drivers 19.50-967956:

These drivers fix the hardware control issues from the previous Linux drivers and support RX5500. However, they reintroduce the Navi TLB problem and the hashrates are about 10-14 MH/s, so you are better off with the previous Linux drivers for Navi cards (or use Windows).

They seem to forgot adding this in 19.50, here is how to fix the problem on Ubuntu or other debian:

Open a nano/vi/joe sessions and paste this:

Code:

diff -rupN a/amd/amdgpu/gfxhub_v2_0.c b/amd/amdgpu/gfxhub_v2_0.c
--- a/amd/amdgpu/gfxhub_v2_0.c  2019-12-12 06:08:03.000000000 +0000
+++ b/amd/amdgpu/gfxhub_v2_0.c  2020-01-11 09:49:24.517473976 +0000
@@ -155,6 +155,15 @@ static void gfxhub_v2_0_init_cache_regs(
        WREG32_SOC15(GC, 0, mmGCVM_L2_CNTL2, tmp);

        tmp = mmGCVM_L2_CNTL3_DEFAULT;
+       if (adev->gmc.translate_further) {
+               tmp = REG_SET_FIELD(tmp, GCVM_L2_CNTL3, BANK_SELECT, 12);
+               tmp = REG_SET_FIELD(tmp, GCVM_L2_CNTL3,
+                                   L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
+       } else {
+               tmp = REG_SET_FIELD(tmp, GCVM_L2_CNTL3, BANK_SELECT, 9);
+               tmp = REG_SET_FIELD(tmp, GCVM_L2_CNTL3,
+                                   L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
+       }
        WREG32_SOC15(GC, 0, mmGCVM_L2_CNTL3, tmp);

        tmp = mmGCVM_L2_CNTL4_DEFAULT;
diff -rupN a/amd/amdgpu/mmhub_v2_0.c b/amd/amdgpu/mmhub_v2_0.c
--- a/amd/amdgpu/mmhub_v2_0.c   2019-12-12 06:08:03.000000000 +0000
+++ b/amd/amdgpu/mmhub_v2_0.c   2020-01-11 09:50:06.586975312 +0000
@@ -142,6 +142,15 @@ static void mmhub_v2_0_init_cache_regs(s
        WREG32_SOC15(MMHUB, 0, mmMMVM_L2_CNTL2, tmp);

        tmp = mmMMVM_L2_CNTL3_DEFAULT;
+       if (adev->gmc.translate_further) {
+               tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3, BANK_SELECT, 12);
+               tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3,
+                                   L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
+       } else {
+               tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3, BANK_SELECT, 9);
+               tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3,
+                                   L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
+       }
        WREG32_SOC15(MMHUB, 0, mmMMVM_L2_CNTL3, tmp);

        tmp = mmMMVM_L2_CNTL4_DEFAULT;

Save it and untar amdgpu-pro-19.50-967956-ubuntu-18.04.tar.xz

Code:

cd amdgpu-pro-19.50-967956-ubuntu-18.04 
dpkg-deb -R amdgpu-dkms_19.50-967956_all.deb amdgpu-dkms_19.50-967956_all
cd amdgpu-dkms_19.50-967956_all/usr/src/amdgpu-19.50-967956
patch -p1 --dry-run < /path-to-your-created-patch-file # If it doesn't fail, remove --dry-run and rerun it.
/* You should see something like that:
checking file amd/amdgpu/gfxhub_v2_0.c
Hunk #1 succeeded at 155 (offset 4 lines).
checking file amd/amdgpu/mmhub_v2_0.c
Hunk #1 succeeded at 142 (offset 5 lines).
*/
cd ../../../../
dpkg-deb -b amdgpu-dkms_19.50-967956_all amdgpu-dkms_19.50-967956_all.deb
And rerun installer!

Enjoy mining with newest driver

[Batmaxpt]

what mining option you have with your gtx 1070ti that you see you get betther results ?

and anyone can help me, i have one rig with 13 gtx 1070ti with afterburner to operate the OC, when the rig start all the fisrt all goes well but after some hours it has some drops hash in some gpus what could be, also notice the pc get a lot more slow, i have 2 psu with 2000w each and also changed the gpu's so the problem is not the gpu, not the psu(i think) and not the risers what could be ?

[VasilyS]

It looks like you have to small size of swap memory. Try to increase it.

[platinum4]

It looks like you have to small size of swap memory. Try to increase it.

64GB page file size, any other suggestions?

[Batmaxpt]

I found myself, using more than 6 Nvidia cards on rig, with or without OC, trying virtual memory between 20 and 90GB (now I keep at 85000MB and works perfect), the rig fails random, one or other GPU is giving error and lock miner and rig too. I used 8xGTX1070 on that rig with Biostar B250 BTC+ motherboard (7xPCI-E 1X and 1xPCI-E 16X), last bios update, 1 stick of 4GB ram, Pentium G4400 GPU (more than needed).

On that times (couple months ago) I found the sollution to "split" rig into more miners (multiple folders with .bat file) so no more than 4-5 cards on miner sequence (.bat file).

Since then, I got no more problems, but I sold all GTX1070 at very good price and replaced with Radeon VII and Radeon 5700XT, doing the same "splitting" cards into multiple miners, because rig with more than 6 cards gives random crashing GPU (same early problem), and lock the miner and the rig too (no self restart).


What I found in time:

Best match power consumption/hashrate for Radeon VII cards is driver 19.7.4 and PhoenixMiner 14.7C, I highly recommend that combination for Radeon VII owners.

Also Radeon VII works great with 19.2.3 drivers and Phoenix Miner 4.8C, even if miner cannot recognize drivers (Unknown open CL Drivers).

Trying with other combinations drivers/miner versions, Radeon VII cannot sustain maximum boost freq for GPU (I use +5% power limit, 1900MHz for GPU and 1100MHz for HBM 2 memory at all cards), and even card is under 65 celcius degrees, GPU boost not increase above 1815MHz ocasionally, most time staying between 173x MHZ and 176x MHz.

Using combinations of driver 19.7.4 and PhoenixMiner 14.7C or 19.2.3 drivers and Phoenix Miner 4.8C (even if miner 14.8C cannot recognize drivers), GPU boost is most time above 186xMHz, making above 97.xxMH/sec with same settings  (+5% power limit, 1900MHz for GPU and 1100MHz for HBM 2 memory at all cards).

Check this printscreen to better understand it (click once more on image, for his largest size).






https://postimg.cc/XBnQ796V

97mh/s what card is doing that

[platinum4]

I think he had a RadeonVII I thought those could max around that.

Only issue is this has happened very recently either since 4.8c/4.9c or 441.87 since it appears to have worked with 441.66/4.7c

[mineshop.eu]

https://postimg.cc/XBnQ796V

97mh/s what card is doing that

Must be Radeon Vll

[WhiteSpot]

More efficient are Radeon 5700XT with under 100Watts consumption for around 51MH, but more cards needed.

I manage to get 300mhs for 900w only..could you show me how to reduce to under 100w each card? Talking about 5700XT ofc

[Trehjulingsleffe]

I found myself, using more than 6 Nvidia cards on rig, with or without OC, trying virtual memory between 20 and 90GB (now I keep at 85000MB and works perfect), the rig fails random, one or other GPU is giving error and lock miner and rig too. I used 8xGTX1070 on that rig with Biostar B250 BTC+ motherboard (7xPCI-E 1X and 1xPCI-E 16X), last bios update, 1 stick of 4GB ram, Pentium G4400 GPU (more than needed).

On that times (couple months ago) I found the sollution to "split" rig into more miners (multiple folders with .bat file) so no more than 4-5 cards on miner sequence (.bat file).

Since then, I got no more problems, but I sold all GTX1070 at very good price and replaced with Radeon VII and Radeon 5700XT, doing the same "splitting" cards into multiple miners, because rig with more than 6 cards gives random crashing GPU (same early problem), and lock the miner and the rig too (no self restart).


What I found in time:

Best match power consumption/hashrate for Radeon VII cards is driver 19.7.4 and PhoenixMiner 14.7C, I highly recommend that combination for Radeon VII owners.

Also Radeon VII works great with 19.2.3 drivers and Phoenix Miner 4.8C, even if miner cannot recognize drivers (Unknown open CL Drivers).

Trying with other combinations drivers/miner versions, Radeon VII cannot sustain maximum boost freq for GPU (I use +5% power limit, 1900MHz for GPU and 1100MHz for HBM 2 memory at all cards), and even card is under 65 celcius degrees, GPU boost not increase above 1815MHz ocasionally, most time staying between 173x MHZ and 176x MHz.

Using combinations of driver 19.7.4 and PhoenixMiner 14.7C or 19.2.3 drivers and Phoenix Miner 4.8C (even if miner 14.8C cannot recognize drivers), GPU boost is most time above 186xMHz, making above 97.xxMH/sec with same settings  (+5% power limit, 1900MHz for GPU and 1100MHz for HBM 2 memory at all cards).

Check this printscreen to better understand it (click once more on image, for his largest size).






https://postimg.cc/XBnQ796V

I have one of these cards. Radeon Vii. I'm doing 93,5MH/s at 220watt with 1830Mhz core (965mV) and 1030Mhz mem (945mV).
I've tried my best to tweak stability, power consumtion and MH/s. I want to try your settings but i cant find what voltage you are using. As far as i've learnt is it really important on, at least, my card. It is stable at these voltages, but even higher volt makes the hashrate lower at certain Mhz.

Please provide your voltage settings

Best regards
//Trehjulingsleffe

[lite12]

About the AMD Linux drivers 19.50-967956:

These drivers fix the hardware control issues from the previous Linux drivers and support RX5500. However, they reintroduce the Navi TLB problem and the hashrates are about 10-14 MH/s, so you are better off with the previous Linux drivers for Navi cards (or use Windows).

They seem to forgot adding this in 19.50, here is how to fix the problem on Ubuntu or other debian:

Open a nano/vi/joe sessions and paste this:

Code:

diff --git a/amd/amdgpu/gfxhub_v2_0.c b/amd/amdgpu/gfxhub_v2_0.c
index b7de60a15623..a5b9e8266a9e 100644
--- a/amd/amdgpu/gfxhub_v2_0.c
+++ b/amd/amdgpu/gfxhub_v2_0.c
@@ -151,6 +151,15 @@ static void gfxhub_v2_0_init_cache_regs(struct amdgpu_device *adev)
        WREG32_SOC15(GC, 0, mmGCVM_L2_CNTL2, tmp);

        tmp = mmGCVM_L2_CNTL3_DEFAULT;
+       if (adev->gmc.translate_further) {
+               tmp = REG_SET_FIELD(tmp, GCVM_L2_CNTL3, BANK_SELECT, 12);
+               tmp = REG_SET_FIELD(tmp, GCVM_L2_CNTL3,
+                                   L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
+       } else {
+               tmp = REG_SET_FIELD(tmp, GCVM_L2_CNTL3, BANK_SELECT, 9);
+               tmp = REG_SET_FIELD(tmp, GCVM_L2_CNTL3,
+                                   L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
+       }
        WREG32_SOC15(GC, 0, mmGCVM_L2_CNTL3, tmp);

        tmp = mmGCVM_L2_CNTL4_DEFAULT;
diff --git a/amd/amdgpu/mmhub_v2_0.c b/amd/amdgpu/mmhub_v2_0.c
index 1a20c3a0f5a3..d9cb98f390e3 100644
--- a/amd/amdgpu/mmhub_v2_0.c
+++ b/amd/amdgpu/mmhub_v2_0.c
@@ -137,6 +137,15 @@ static void mmhub_v2_0_init_cache_regs(struct amdgpu_device *adev)
        WREG32_SOC15(MMHUB, 0, mmMMVM_L2_CNTL2, tmp);

        tmp = mmMMVM_L2_CNTL3_DEFAULT;
+       if (adev->gmc.translate_further) {
+               tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3, BANK_SELECT, 12);
+               tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3,
+                                   L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
+       } else {
+               tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3, BANK_SELECT, 9);
+               tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3,
+                                   L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
+       }
        WREG32_SOC15(MMHUB, 0, mmMMVM_L2_CNTL3, tmp);

        tmp = mmMMVM_L2_CNTL4_DEFAULT;

Save it and untar amdgpu-pro-19.50-967956-ubuntu-18.04.tar.xz

Code:

cd amdgpu-pro-19.50-967956-ubuntu-18.04 
dpkg-deb -R amdgpu-dkms_19.50-967956_all.deb amdgpu-dkms_19.50-967956_all
cd amdgpu-dkms_19.50-967956_all/usr/src/amdgpu-19.50-967956
patch -p1 --dry-run < /path-to-your-created-patch-file # If it doesn't fail, remove --dry-run and rerun it.
/* You should see something like that:
checking file amd/amdgpu/gfxhub_v2_0.c
Hunk #1 succeeded at 155 (offset 4 lines).
checking file amd/amdgpu/mmhub_v2_0.c
Hunk #1 succeeded at 142 (offset 5 lines).
*/
cd ../../../../
dpkg-deb -b amdgpu-dkms_19.50-967956_all amdgpu-dkms_19.50-967956_all.deb
And rerun installer!

Enjoy mining with newest driver

applying the patch give me:

Code:

checking file amd/amdgpu/gfxhub_v2_0.c
Hunk #1 FAILED at 151.
1 out of 1 hunk FAILED
checking file amd/amdgpu/mmhub_v2_0.c
Hunk #1 FAILED at 137.
1 out of 1 hunk FAILED

while the md5sum on the file are:

Code:

dec90c4bb18b1bb59a70280737f9631f  amd/amdgpu/gfxhub_v2_0.c
c489344413c00d38a9afa488ed20cac9  amd/amdgpu/mmhub_v2_0.c
e0e2bd654befbca20676a15517d59de8  /root/patch.file

[laik2]

applying the patch give me:

Code:

checking file amd/amdgpu/gfxhub_v2_0.c
Hunk #1 FAILED at 151.
1 out of 1 hunk FAILED
checking file amd/amdgpu/mmhub_v2_0.c
Hunk #1 FAILED at 137.
1 out of 1 hunk FAILED

while the md5sum on the file are:

Code:

dec90c4bb18b1bb59a70280737f9631f  amd/amdgpu/gfxhub_v2_0.c
c489344413c00d38a9afa488ed20cac9  amd/amdgpu/mmhub_v2_0.c
e0e2bd654befbca20676a15517d59de8  /root/patch.file

Did a clean patch without using patchwork posted for mainstream drivers, retest with edited code in my previous post.

[black1ight]

Hi,

The average share difficulty for dev mining is 5 Mhs but regular share difficulty is 20 Ghs . Why such a huge difference ?

[PhoenixMiner]

About the AMD Linux drivers 19.50-967956:

These drivers fix the hardware control issues from the previous Linux drivers and support RX5500. However, they reintroduce the Navi TLB problem and the hashrates are about 10-14 MH/s, so you are better off with the previous Linux drivers for Navi cards (or use Windows).

They seem to forgot adding this in 19.50, here is how to fix the problem on Ubuntu or other debian:

...

Enjoy mining with newest driver

   Nice find! Thank you very much for this information!

   We have put together the patched files from the driver and here are the slightly easier installation instructions for the patched driver:

1. If you have installed the unpatched driver or any other AMD driver, uninstall it first:

Code:

amdgpu-pro-uninstall

  and then reboot the rig.

2. Download the driver from AMD web site (it will be named amdgpu-pro-19.50-967956-ubuntu-18.04.tar.xz)

3. Download the patch from here: https://github.com/PhoenixMinerDevTeam/files/blob/main/amdgpu-pro-19.50-967956-patch.tar.gz?raw=true (the file name is amdgpu-pro-19.50-967956-patch.tar.gz)

4. To verify the downloaded patch file, you can use the following checksums:

Code:

   SHA-1: 1f8ca89e98f9cb625ae9ad6194a012a642f4b39d
 SHA-256: c969e326971acbaef9a3e20d0aa2459f70ac5b6cd5cbde79614289c99051a9aa
 SHA-512: e96652ffac48375b97b756bed02296180ef92a07f9b407db66bf003c58d805e80f15e9c9a2525e2706b0de7e4391e732ae6800adf90c9730fb25546581a1a6a5

5. Place both files in the same folder and first extract the files of the driver with the following command:

Code:

  tar -Jxvf amdgpu-pro-19.50-967956-ubuntu-18.04.tar.xz

 
6. Then extract the patched files, which will overwrite some files in the extracted driver:

Code:

  tar -zxvf amdgpu-pro-19.50-967956-patch.tar.gz

 
7. Finally, go to the extracted driver folder and start the installation with the following commands:

Code:

  cd amdgpu-pro-19.50-967956-ubuntu-18.04
  ./amdgpu-pro-install -y --opencl=pal,legacy --headless

Hi,

The average share difficulty for dev mining is 5 Mhs but regular share difficulty is 20 Ghs . Why such a huge difference ?

  The share difficulty is determined by the pool. When the pool used for devfee is different than your pool, the difficulty can also be different.

[mineshop.eu]

Cards are Radeon VII, those makes around 88-90MH when stock and more than 95-96MH with OC, could be pushed around 100MH, but I think safe and stable are at 1100MHz Mem and 1900MHz GPU, for 95 to 97MH for 260-290Watts consumption.


https://postimg.cc/Hjkmpkbt


More efficient are Radeon 5700XT with under 100Watts consumption for around 51MH, but more cards needed.

I guess people are looking software watt usage and not looking for actual power usage from the wall.
I have about the same 320mh/s on 850 watt

[vydelek]

It is possible to set MINIG SW remote restart of RIG as it should be SRBMINER.

Thank you for your help

[melpheos]

OK I did some more investigation.
For some reason, ETC speed is half the speed of ETH speed with the same batch file.
This doesn't make sense in my view. Am I missing something

[tg88]

OK I did some more investigation.
For some reason, ETC speed is half the speed of ETH speed with the same batch file.
This doesn't make sense in my view. Am I missing something

This is due to the difference in DAG size, ETC has the largest DAG and some gpus architectures have degraded performance for this reason.

[papampi]

Any way to prevent  DevFee Dag switch when mining on nicehash?
Tried "-coin eth" but looks like its not working

[laik2]

- NVIDIA Pascal expected slowdown compared to AMD Vega -

BLOCKHIGHT: 1000000( 1 million ) - DAG size: 1.28GB
GTX 1070 8GB: 30.75MH
Radeon RX Vega64 8GB: 49.21MH
Radeon RX Vega56 8GB: 43.12MH

BLOCKHIGHT: 2000000( 2 million ) - DAG size: 2.33GB
GTX 1070 8GB: 30.72MH
Radeon RX Vega64 8GB: 49.11MH
Radeon RX Vega56 8GB: 43.01MH

BLOCKHIGHT: 9000000( 9 million ) - DAG size: 3.4GB
GTX 1070 8GB: 30.70MH
Radeon RX Vega64 8GB: 49.07MH
Radeon RX Vega56 8GB: 42.97MH

BLOCKHIGHT: 10000000( 10 million) - DAG size: 3.66GB
GTX 1070 8GB: 29.41MH
Radeon RX Vega64 8GB: 49.07MH
Radeon RX Vega56 8GB: 42.97MH

BLOCKHIGHT: 15000000( 15 million ) - DAG size: 4.3GB
GTX 1070 8GB: 23.78MH
Radeon RX Vega64 8GB: 49.05MH
Radeon RX Vega56 8GB: 42.94MH

BLOCKHIGHT: 20000000( 20 million ) - DAG size: 6.2GB
GTX 1070 8GB: 21.47MH
Radeon RX Vega64 8GB: 49.02MH
Radeon RX Vega56 8GB: 42.94MH

For everyone who keeps asking why Pascal(NVIDIA GTX 10x0)[Not tested on Turing] is getting slower on each new DAG epoch.

[0xcosmos]

OK I did some more investigation.
For some reason, ETC speed is half the speed of ETH speed with the same batch file.
This doesn't make sense in my view. Am I missing something

which gpu are you using?
the problem only persist with nvidia gpus not with amd ones
my rx 580 8gb hynix bios modded does 30.1 mh in eth as well as etc