Mining ethereum with NVIDIA 9xx cards in Windows 10 with latest drivers

[benether]

Hi everyone,

It's my first post on this forum and I want to share my experience in how to efficiently mine ethereum with Windows 10, Nvidia 9xx GPU (maxwell) and uptodate drivers.
I'm spend lot of time from various resources to have a decent hashrate, and i though it would be usefull for other to have a complete tutorial.

My setup is very simple : a gaming computer with one NVIDIA KFA2 GTX 970 (Black OC edition).
My current hashrate is around 21 MHs. (and started at 3 MHs, without any optimization).
I'm using Claymore miner and ethermine pool, but you can choose other if you want to.

Here some downloads links you will need if you want to follow this tutorial :

• Latest nvidia driver  - 384.94  as i'm writing : http://www.nvidia.com/download/driverResults.aspx/120905/en-us
• Claymore 9.8 Miner : https://bitcointalk.org/index.php?topic=1433925.0
• MSI afterburner : http[Suspicious link removed]/vga/MSIAfterburnerSetup.zip

Install and configure nvidia driver

Installing last driver should be easy, but some fine tuning is necessary.
Once driver is installed, go to "Nvidia Configuration Panel" (right click menu on your desktop).
You should see the following screen (mine is in french, but i will translate for you what needed to be tuned)

https://thumb.ibb.co/n5W3h5/NVIDIA_Control_Panel_2.jpg

Go to "3D Settings" => "Manage 3D" setting and :

• Make sure "CUDA - GPUs" is set to "All"
• Set "DSR - Factors" to "2x"
• (and most important) Set "Optimise for Compute Performance" to "Yes"

Now you have a nvidia drivers optimized for mining

Install Claymore

Installing Claymore is straightforward (just unzip where you want to)
Create a new file in the main directory (mining_ether.bat) with the something like :

Code:

setx GPU_FORCE_64BIT_PTR 0
setx GPU_MAX_HEAP_SIZE 100
setx GPU_USE_SYNC_OBJECTS 1
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_SINGLE_ALLOC_PERCENT 100
EthDcrMiner64.exe -epool eu2.ethermine.org:14444 -ewal 0xdc7eFDbBE4aFD15c3385d1156b371333550FdDcF -epsw x -eworker benether.2 -mode 1 -allpools 1

• setx parameters are for GPU configuration
• -epool eu2.ethermine.org:14444 : for using ethermine pool (you can mine in other one)
• -ewal 0xdc7eFDbBE4aFD15c3385d1156b371333550FdDcF : your wallet address. MAKE SURE TO USE YOURS
• -epsw x : pool password if you have one (ethermine don't have)
• -eworker benether.2 : an arbitrary name for your worker
• -mode 1 : only mine ether (and no dual mining - devfee is 1% for single ether mining - 2% for dual mining)
• -allpools 1 : needed for devfee mining

GPU/Memory Frequency (pstate) and overclocking

Start mining before following this part (for having load on your GPU)

Nvidia cards have a special mechanisms for auto ajusting GPU/Memory frequencies based on actual GPU workload.
It's called pstate (Performance State)
P0 is the maximum pstate, that is used when gaming for example.

When mining ether, nvidia drivers dont detect this workload as an intensive one, and then use the P2 pstate.
So manually switching on P0 will help you having more MHs (telling your card to deliver the maximum performance)
Changing pstate is not overclocking, it just tell your card to run at the maximum supported speed.

You can change pstate using nvidia-smi.exe (installed with nvidia drivers in C:\Program Files\NVIDIA Corporation\NVSMI).

I will illustrate this with my GTX 970, but you should have different value depending on your nvidia model.

You need to be "administrator" for changing pstate.
Just launch "cmd" as administrator (type cmd in cortana bar, and right click for launching it as an admin)

First command is for checking your actual pstate

Code:

nvidia-smi.exe -q -d PERFORMANCE

You must look at "Performance State". It should be P2.

You should have something like this (my card is on P0 on this screenshot)

https://preview.ibb.co/dZXHs5/cmd_pstate.jpg

Second command is to get clock frequencies that you card support (without overclocking) :

Code:

nvidia-smi.exe -q -d SUPPORTED_CLOCKS|more

Just note the two first clocks number (Memory and Graphics). With my card it's 3505 Mhz/1506 Mhz

https://preview.ibb.co/jbE8zk/cmd_supported.jpg

Then, you can switch to P0 state using the following command :

Code:

nvidia-smi.exe -ac 3505,1506

If you now check with the 'nvidia-smi.exe -q -d PERFORMANCE' you will see that your card is in P0 state.

But… (and here I don't really understand what's happening, just doing observation and tuning )
Launch MSI afterburner.

You wil see that your card isn't at the GPU frequency you manually set with nvidia-smi.
In my case I have 1404/3505.
If you manually slide Core Clock slider you can set GPU to the desired frequency (for me +102)
At this point you will use the maximum frequency (and hashrate) that your card support without overclocking.
If you continue sliding the Core Clock (for example +123) then, the frequency won't go further, but you will undervolt your card (mine is 1212mv as factory setting, setting +123 Mhz will switch to 1187 mv).
Undervolt can be cool as your card will consume less electricity, but can be instable as running high frequencies need power.

But… (still don't really understand what's happening).
It you save in MSI some settings into profiles (for example +131Mhz GPU / + 50 Mhz Memory), and applying it, then you will overclock your card. Voltage will be set to factory setting (1212 mv in my case) but GPU and memory will then be overclocked.
This setting is getting my card to 1535 / 3557 (using 100% power).
That's my usual setting for mining, achieving 21 MHs with no OC errors on Claymore.

https://preview.ibb.co/dMexs5/mining.jpg


I hope this tutorial will help you getting some precious MHs
If you feel generous, fell free to send me some ether on my wallet : 0xdc7eFDbBE4aFD15c3385d1156b371333550FdDcF

If something is not clear or working, let me a question on this topic, I will try to help you if i can.
If you have an explanation for smi/afterburner behavior i would be pleased to understand what's happening.
If you have other tips for optimize nvidia mining, I would be pleased to used them too.

Bonus : Mine Ethereum Classic

Mining ETC is almost as profitable as mining ETH.
It's very simple to switch on ETC mining.
You just need an ETC wallet/pool. Ethermine provide ETC pool also (eu1-etc.ethermine.org:14444).

You just have to change mining.bat command. Mine is :

Code:

setx GPU_FORCE_64BIT_PTR 0
setx GPU_MAX_HEAP_SIZE 100
setx GPU_USE_SYNC_OBJECTS 1
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_SINGLE_ALLOC_PERCENT 100
EthDcrMiner64.exe -epool eu1-etc.ethermine.org:14444 -ewal 0xe3b6b8a0fb6f783006594198f674ccbd45471906 -epsw x -eworker benether.1 -mode 1 -allpools 1

One advantage is that you will generate ETC more quicky than ETH and have more frequent payout with ethermine without any fee. (there no fee is you ask to be paid each 1 ETC -or 1 ETH- you mine). But mining 1 ETC is quicker than mining 1 ETH

Benoit

[1714794204]

1714794204

[Munhenhos]

I don't have any 970. But this is outstanding work

[Temporel]

Hi,

FYI:

you wrote Set "DSR - Factors" to "2x" but it is set to 4x in your picture and there is no "Optimiser pour les performances de calcul" for 1080Ti cards unless Im missing something.

Great guide anyway,

Thank you !

[szafa]

Someone looking off rgb in gtx 1080ti download tweak.

[benether]

Indeed, i put the wrong screenshot - I just update the post.
I've see better result with 2x.

10xx (Pascal) do not have "Optimize for compute" option. It's only for Maxwell arch. (but seems to be activated by defaut for pascal arch). See http://nvidia.custhelp.com/app/answers/detail/a_id/4370/~/what-does-the-setting-optimize-for-compute-performance-do%3F

Only thing needed for 10xx to have de decent hasrate, it's to have at least windows anniversary update (or creators update). (WDDM 2.1)

[TigTex]

Instead of the all nvidia-smi stuff you do there, I do a different aproach. This is valid on ALL nvidia gpus!

Using the latest nvidia profile inspector (you can grab it here https://ci.appveyor.com/project/Orbmu2k/nvidiaprofileinspector/build/artifacts ) you search for the option "CUDA - Force P2 State" that defaults to ON and change it to OFF. Apply and reboot.
Now you will always run cuda apps on P0 state without messing with nvidia-smi!

Example:

[stellan0r]

great first post!

But I do not get the idea why the DSR setting should increase your hashrate.. I will test it anyway tomorrow with my 10xx rig.

[Temporel]

great first post!

But I do not get the idea why the DSR setting should increase your hashrate.. I will test it anyway tomorrow with my 10xx rig.

it doesnt on my 1080 and 1080Ti

[benether]

@TigTex : thanks for this tips. I will try it. Setting P0 manually is boring


@stellan0r : Thanks !
DSR will not increase the hashrate, but i had errors with claymore if not set it to 2x (CUDA error - cannot allocate big buffer for DAG)

[stellan0r]

@stellan0r : Thanks !
DSR will not increase the hashrate, but i had errors with claymore if not set it to 2x (CUDA error - cannot allocate big buffer for DAG)

actually I found a slight Hashrate increase with Verge Coin (X17 in ccminer) with my gtx1060 rig, approx. 0.1 mh/gpu more.. but it may also be just the position of the moon or the direction of wind..

[benether]

Did you set it to 2x or 4x ?

[stellan0r]

2x

Strange thing today, I went back from mining SIGT for a couple of days and when I tried to start eth or zcash mining the miners kept crashing or were showing low hashrates, some cards didn't mine at all. Rebooting did not help, resetting OC in Afterburner did not help. I am now mining XVG again (which is working fine in ccminer) and will roll back all changes I made the last days after work. My main rig did update windows this morning though, maybe I need to check what was updated. Both my main rig and my secondary rig had a firewall permission request (private or public network access) from claymore miner this morning, which is uncommon, too..

[AzzAz]

2x

Strange thing today, I went back from mining SIGT for a couple of days and when I tried to start eth or zcash mining the miners kept crashing or were showing low hashrates, some cards didn't mine at all. Rebooting did not help, resetting OC in Afterburner did not help. I am now mining XVG again (which is working fine in ccminer) and will roll back all changes I made the last days after work. My main rig did update windows this morning though, maybe I need to check what was updated. Both my main rig and my secondary rig had a firewall permission request (private or public network access) from claymore miner this morning, which is uncommon, too..

Saw this too. Some rigs I managed to mine eth and zec, used Winpatrol to erase SIGT bat from start files ( "manage start files" ), few restarts...

[stellan0r]

2x

Strange thing today, I went back from mining SIGT for a couple of days and when I tried to start eth or zcash mining the miners kept crashing or were showing low hashrates, some cards didn't mine at all. Rebooting did not help, resetting OC in Afterburner did not help. I am now mining XVG again (which is working fine in ccminer) and will roll back all changes I made the last days after work. My main rig did update windows this morning though, maybe I need to check what was updated. Both my main rig and my secondary rig had a firewall permission request (private or public network access) from claymore miner this morning, which is uncommon, too..

Saw this too. Some rigs I managed to mine eth and zec, used Winpatrol to erase SIGT bat from start files ( "manage start files" ), few restarts...

my problem kind of developed into a major f*ck up, I have problems with all miners (ccminer on X17 and Skunk), ethminer, ewbf zcash miner and claymore ( 9.7 + 9.8 ) since  this morning and the little -whatever- automatic windows update.
After dozens of reboots, unplugging risers, removing the placeholder 1050ti and so on i can barely run some of the miners if i do not or if only slightly oc my gtx 1060s. Each "standard" oc setting (fitting the needs of the algos i used) I used up to yesterday produces a certain crash of the miner.
I always encounter some kind of memory error, e.g. "Cuda error in func 'quark_blake512_cpu_setBlock_80_sp' at line 577 : unknown error" (ccminer x17); "cannot initialize nvml. temperature monitor will not work" (ewbf zcash miner, now working on very low OC); "Error CUDA mining: an illegal memory access was encountered" (ccminer);  "CUDA error - cannot allocate big buffer for DAG. Check readme.txt for possible solutions." (Claymore, changed DAG value -eres to 1)...

All cards are Gigabyte 1060s with 6GB RAM. Rolled back nvidia drivers, updated again, removed the windows security update from this morning. Nothing really works. Increased DAG can't be the single issue, as all my altcoin algos are affected.
Anyone any ideas?

[benether]

Last windows update raise my hashrate by 0.1 :p
But behavior may be related to GPU (GTX 970 for me)

Did you try to reinstall last nvidia drivers ?
Maybe you can try also beta one (http://www.nvidia.com/download/driverResults.aspx/121776/en-us)

[AzzAz]

2x

Strange thing today, I went back from mining SIGT for a couple of days and when I tried to start eth or zcash mining the miners kept crashing or were showing low hashrates, some cards didn't mine at all. Rebooting did not help, resetting OC in Afterburner did not help. I am now mining XVG again (which is working fine in ccminer) and will roll back all changes I made the last days after work. My main rig did update windows this morning though, maybe I need to check what was updated. Both my main rig and my secondary rig had a firewall permission request (private or public network access) from claymore miner this morning, which is uncommon, too..

Saw this too. Some rigs I managed to mine eth and zec, used Winpatrol to erase SIGT bat from start files ( "manage start files" ), few restarts...

my problem kind of developed into a major f*ck up, I have problems with all miners (ccminer on X17 and Skunk), ethminer, ewbf zcash miner and claymore ( 9.7 + 9.8 ) since  this morning and the little -whatever- automatic windows update.
After dozens of reboots, unplugging risers, removing the placeholder 1050ti and so on i can barely run some of the miners if i do not or if only slightly oc my gtx 1060s. Each "standard" oc setting (fitting the needs of the algos i used) I used up to yesterday produces a certain crash of the miner.
I always encounter some kind of memory error, e.g. "Cuda error in func 'quark_blake512_cpu_setBlock_80_sp' at line 577 : unknown error" (ccminer x17); "cannot initialize nvml. temperature monitor will not work" (ewbf zcash miner, now working on very low OC); "Error CUDA mining: an illegal memory access was encountered" (ccminer);  "CUDA error - cannot allocate big buffer for DAG. Check readme.txt for possible solutions." (Claymore, changed DAG value -eres to 1)...

All cards are Gigabyte 1060s with 6GB RAM. Rolled back nvidia drivers, updated again, removed the windows security update from this morning. Nothing really works. Increased DAG can't be the single issue, as all my altcoin algos are affected.
Anyone any ideas?

Well now you can only make rig(s) again. Clean all, reinstall windows. And remember, once you get it working-dont move cards/risers. This is remaining software problem.

[stellan0r]

Did you try to reinstall last nvidia drivers ?
Maybe you can try also beta one (http://www.nvidia.com/download/driverResults.aspx/121776/en-us)

I reinstalled, but have to go the way of complete deinstallation of the drivers (and CUDA and everything related) and reinstalling.
I will try the beta this weekend when I have more time but will most likely stick to an older version of the NVIDIA drivers.

2x

Strange thing today, I went back from mining SIGT for a couple of days and when I tried to start eth or zcash mining the miners kept crashing or were showing low hashrates, some cards didn't mine at all. Rebooting did not help, resetting OC in Afterburner did not help. I am now mining XVG again (which is working fine in ccminer) and will roll back all changes I made the last days after work. My main rig did update windows this morning though, maybe I need to check what was updated. Both my main rig and my secondary rig had a firewall permission request (private or public network access) from claymore miner this morning, which is uncommon, too..

Saw this too. Some rigs I managed to mine eth and zec, used Winpatrol to erase SIGT bat from start files ( "manage start files" ), few restarts...

my problem kind of developed into a major f*ck up, I have problems with all miners (ccminer on X17 and Skunk), ethminer, ewbf zcash miner and claymore ( 9.7 + 9.8 ) since  this morning and the little -whatever- automatic windows update.
After dozens of reboots, unplugging risers, removing the placeholder 1050ti and so on i can barely run some of the miners if i do not or if only slightly oc my gtx 1060s. Each "standard" oc setting (fitting the needs of the algos i used) I used up to yesterday produces a certain crash of the miner.
I always encounter some kind of memory error, e.g. "Cuda error in func 'quark_blake512_cpu_setBlock_80_sp' at line 577 : unknown error" (ccminer x17); "cannot initialize nvml. temperature monitor will not work" (ewbf zcash miner, now working on very low OC); "Error CUDA mining: an illegal memory access was encountered" (ccminer);  "CUDA error - cannot allocate big buffer for DAG. Check readme.txt for possible solutions." (Claymore, changed DAG value -eres to 1)...

All cards are Gigabyte 1060s with 6GB RAM. Rolled back nvidia drivers, updated again, removed the windows security update from this morning. Nothing really works. Increased DAG can't be the single issue, as all my altcoin algos are affected.
Anyone any ideas?

Well now you can only make rig(s) again. Clean all, reinstall windows. And remember, once you get it working-dont move cards/risers. This is remaining software problem.

If nothing helps I will do that, I am expecting two more (replacement) 1060s (one for the 1050ti, one for a single-fan 1060 which has been too noisy) tomorrow and see how it goes. If everything works (but full OC is not possible, system unstable or one of the miners not working at all (like ethminer/claymore) i will rebuild the rig (disconnecting all gpus, format c: from usb stick after backup of my batch files and reinstalling windows 10 with anniversary update).
Problem still is, even if everything is working as expected, you never know if some windows update you could not prevent messes everything up as I don't know if I can fully disable all windows updates.

[umine]

Hi everyone,
.....

Benoit

I've ASUS Strix GTX970 and got about 20 Mh/s on Ethash and 275 Sol/s on Equihash (Win10 x64Pro) little bit downloaded via Maxwell Bios editor so power consumption about 140W

[koya]

Benether, since you are already using claymore miner, did you also try dual mining? And did you play with -dcri values for balancing eth/dual coin

Ofc in case that overclocking core on 970 gives adequate ethash boost, that means that ethash algo uses all gpu power, in that case dual mining will not be of much use

[benether]

Benether, since you are already using claymore miner, did you also try dual mining? And did you play with -dcri values for balancing eth/dual coin

Ofc in case that overclocking core on 970 gives adequate ethash boost, that means that ethash algo uses all gpu power, in that case dual mining will not be of much use

Yes, I currently mine ETC+SIA with claymore.
You can live test "dcri" effect without stop/start mining using "+"/"-" keys ( Cf claymore readme : You can change the intensity in runtime with "+" and "-" keys and check current statistics with "s" key.)

For me, the default value (30) is fine, but i havent tested it a lot.

I've see a very little hashrate drop when dualmining vs simple mining.
I've also noticed that their is a strict 10x ratio between ETH hashrate and Sia ones (if you have 19 MH/s in ETH, you will have 190 MH/s for SIA) and i dont think OC or not change anything (i supposed sia mining is just using what is not used from eth mining, OC will just raise both hashrates)

Let me know if you find some interessing settings for dcri/OC.