HD6970 not mining after Monero hard fork

[pintybor]

Until now I have been using xmr-stack v2.5.0 with HD6970 and it was working OK. After Monero hard fork on 18.10.2018 hashrate dropped to 0 and it stopped mining. I upgraded to version 2.5.1 but the issue is still there, hashrate is 0 and GPU is not mining at all.

I'm using AMD Catalyst 14.4 and Windows 10 1809.

[psychocrypt]

please try to delete the opencl cache. The path is shown if you start the miner. I tried two days ago a system with different hd6xxx and all works fine. which currency do you selected? Maybe the pool blocked you?

[psychocrypt]

try also an other pool.

[pintybor]

thnx for quick reply. I already tried clearing opencl cache, but it didn't help. I selected currency "monero". It all worked fine until about 18.10.2018 12:00 UTC but after that it stopped. I will try another pool and see if that works.

[pintybor]

I tried another pool but the issue remains. I get logged into pool OK but hashrate is 0, GPU load is 0...

[Masikai]

i am having the same issue with my HD 6870 1gb gpu's. used to get ~100 hs

running catalyst 15.7.1 with XMR Stak 2.5.1, deleted all my opencl cache files.

still no joy.

any help would be greatly appreciated.

[xmronadaily]

Im running HD 6990s on WIn 10 64, driver is 14.4, and they're running okay, all valid shares. So try use DDU and clean everything, then reinstall drivers, maybe that fixes it.

[pintybor]

I have tried reinstalling the drivers and cleaning using AMD Cleanup Utility and also DDU. I tried 14.4 dirver, 15.7.1 and 15.12 driver. The issue remains no matter what. Everything seems OK, code is compiled, pool is logged in, but hashrate is always 0 and GPU load is also 0.

I really don't know what else to try, so any help would be greatly appreciated.

[Wacholek]

What are your AMD settings? OC ? I had the same problem with HD6990.

[xmronadaily]

I remembered something now, this used to happen to me whenever I had HD 6990 as a display card, but when I switched (I put a spare hd 7850 in the main x16 slot as display card) the problem with 6990 went away, it might be that it doesn't like outputing display at the same time it's mining, so try that if you have additonal cards... Other than that, there's no reason why 14.4 shouldn't work.

[pintybor]

I have also tried that to have another card be for the display but the problem remains.. I have tried xmrig-amd miner and the problem is exactly the same. Code is compiled pool logged in but hashrate is 0. As I was able to mine OK before with cryptonight_v7 but now after hard fork to cryptonight_v8 hashrate is 0 no matter what. Could it be, that this new protocol is somehow not compatible with HD6970? Is there anything else I could try? Some settings for the thread or something?

[Wacholek]

I have 4 HD6990 and they are working fine on on V8. But on Windows 7. 14.4 is not for Windows 10. It has no full access to the cards. At least I did not. I could not OC. I had problems with the stability at 2.5.0 PR but now is rock stable at very different setting than on V7. The memory clock does make a difference and it have to be on 1250Mhz. Core 930Mhz +20% and I get 240h/s per core.

The best settings I found:

"intensity" : 768, "worksize" : 32, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true

[Simaex]

@Wacholek 14.4 driver works great with Win10, no problem overclocking. I have used Claymore 9.7 miner to set proper clocks :-), then stop it and use xmr-stak for monero mining. Do you try worksizes from 12 to 21? It use to be optimal for CNv7 but I have not try CNv8 yet. Please share you results cause 240 H\s do not make sence for me powerwize :-(

[pintybor]

I also don't think that this is Windows version or driver issue as I have tried every combination. Simply HD6970 with Cryptonight_v8 is not working... If anyone finds a solution for this issue I would really appreciate it...

[Wacholek]

HD6970 is compatible with CNv8. HD6990 is a 2xHD6970 and is working. Slow but it is working. This is an old GPU and is not good for V8. For now 240h/s is the best what I can get. @Simaex Interesting idee. I have found that HD6990 on Win10 was as fast without OC as Win7 with OC. I have to give it one more try. I have tried 8, 16, 32 worksize. I will make some test today.

[Wacholek]

[mijaque]

Hi,

I have the same problem with my 2 hd6970 and 2 hd6950. Before the upgrade in V8, I had between 350 and 380 h / s by cards. But since the upgrade, 140h / s by cards ... Config:

• PC windows 7 64b
• Driver 15.3
• Config AMD : { "index" : 0, "intensity" : 875, "worksize 16" : "affine_to_cpu" : false, "strided_index" : 1, "mem_chunk" : 2,"comp_mode" : true

I tried to change the settings and no changes. I am re-installing my drivers and doing other tests

[psychocrypt]

140h/s is normal for this card since v8 is active.

[Wacholek]

No, it is not normal. "intensity" : 768, "worksize" : 32, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true Driver only 14.4. OC 930/1250 and you will have the same results as I have.

By the way, what is the "unroll" value? For my Vega the best set is "1".

[psychocrypt]

Why should it not be normal. You can not compare v7 hash rate with v8. All old gpus have noe a bad hash rate. Unroll controlls how many times the compile can transform the loop in explicit unfodet comand flow. This allow better optimizations. But if it is ro large you need to much register and it will slowdown your gpu.

[Wacholek]

I just wanted to point out that my results are far better than 140h/s. I can only compare v8 results. So 140h/s to my 240h/s. Thanks for info.

[mijaque]

With your parameter i hit the 220h/s per card 😀. What is the soft you use to OC ? Thks

[Wacholek]

Msi Afterburner.

[mijaque]

I tried and when i activate the officiel overclock settings, the system won't boot anymore (stuck in Windows opening logo) I had to start in safe mode and restore my system to the last point. Any idee ?

[Simaex]

@mijaque a lot of people have problems using AB for OC with obsolete GPUs. I already mentioned workaround. Claymore 9.7 miner can set OC. Then you stop it and start proper xmr-stak. OC will stay until reboot. So I wrote simple batch file that performed the sequence in case of accidental reboot. You may need to start from DDU after AB problem even if you restore from last point.

[mijaque]

@Simaex Thank you ,i'll give it a try tonight

Is there a way to decrease the watt consumption on the V8 ? cause i understand that we lost some hash power in the update but for the same watt voltage. Thanks

[Wacholek]

@Simaex Claymore 9.7 OC does not work form me. I does not change settings on 3 out of 8 my GPU's. So decided to edit vBios to have clocks and voltages what I wont. Unfortunately I cannot edit fan speed in vBios. I will try some more.

[Simaex]

@Wacholek are you still using Win7 for more then 5 GPU? I have a lot of problems with Win7 for more then 5 GPU. For HD6990 you really do not need to change fan speed, automatic settings work great until you need to reapply termal grease.

[Wacholek]

@Simaex Yes still Win7. But I am planing to install Win10 and see if Claymore 9.7 will work ok. I like to keep my GPU temps way below 70'C. This is not possibile on auto fan. Like my Vegas max 41'C @264W.

[pintybor]

Unfortunately I still cannot get my HD6970 to mine at all... All I have been able to do is, if I set mem_chunk setting to 5 or above then card begins to mine and also GPU activity is full load. But unfortunately using this setting all I get are invalid results as shown below. Hashrate looks ok but all invalid results. If this helps anyone solve the problem using some other settings that would be great. I'm running at 920/1375 which is stock for my card and P+20%. My settings are: "intensity" : 256, "worksize" : 32, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 8, "unroll" : 8, "comp_mode" : true

[psychocrypt]

I got again access to a system with a gpu which also creates invalid shares. Bad news: I have not found the issue.

Please try to use a newer driver. I think the new code trigger some compiler bugs in the driver.

[mijaque]

@pintybor what driver you use ?

[pintybor]

Right now I'm using 14.4. I also tried 15.7.1, 15.12 and driver which windows 10 installs by default. With all drivers result is the same.

[mijaque]

Have you tried the settings of wacholek : intensity" : 768, "worksize" : 32, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true It's Works for me

[Wacholek]

@pintybor Do not use other driver than 14.4. Other will not detect full available memory of the card. Have you selected Cryptonight_V8 algo in config file? Which pool are you using? I have had similar problem with 2.5.0. All of my cards where generating Invalid shares after some time (15min). On 2.5.1 problem disappeared. But now I problem with stoping hashing. I have 8 and after a day of mining one or two or one and a half stop hashing. But cards did not turn off and the driver did not restart. When I turn off XM-STAK and fire it up once again without reboot hashing fine for another day. Reboot and driver reinstall does not help. I will switch to Win10 and see fit that helps.

[mijaque]

@Wacholek I have the same problem as you. Randomly, on of my cards stop hashing and show "na". I have to stop the worker of reboot the rig

[Simaex]

@mijaque it is not uncommon to use watchdog for a mining rig. Surely it's better to have stable rig but one forced reboot daily is no problem IMO

[Wacholek]

@mijaque Have you tried not to reboot the rig? Just close XMR-STAK and start again? For me it is working without reboot. @Simaex Do you have some kind of a watchdog which can monitor XMR-STAK hashrate? Program? Or a USB reset stick?

[Wacholek]

Unfortunately I still cannot get my HD6970 to mine at all... All I have been able to do is, if I set mem_chunk setting to 5 or above then card begins to mine and also GPU activity is full load. But unfortunately using this setting all I get are invalid results as shown below. Hashrate looks ok but all invalid results. If this helps anyone solve the problem using some other settings that would be great. I'm running at 920/1375 which is stock for my card and P+20%. My settings are: "intensity" : 256, "worksize" : 32, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 8, "unroll" : 8, "comp_mode" : true

First of all to low intensity. If You cannot push it higher it means that Your driver does not detect full memory of the cards.

Do this: Change the amd.txt config file name to amdback.txt and run XMR-STAK 2.5.1. That will generate new amd.txt automatically. Post the content of new amd.txt.

And by the way this is not a good hashrate at those clocks (low intensity).

[mijaque]

@Wacholek Yes and sometimes it's no enouth. I use Minerstat to manage my rigs and for now it's way more stable than juste use XMR-STAK only. i'll see if your upgrade ti win10 resolve the problem or wait for an other version of the miner

[Simaex]

@Simaex Do you have some kind of a watchdog which can monitor XMR-STAK hashrate? Program? Or a USB reset stick?

I am using HIVE OS program watchdog, it is universal for all HIVE-supported miners. Unfortunately HIVE do not support Antilles and Cayman GPU so it is irrelevant to your issue. There is Hive for Windows but I do not test it up to the moment. Many people use to praise SimpleRigResetter as best hardware-based watchdog. If you can justify it's price for your case ... A lot of simple chinese USB watchdogs available for small farms but it is not easy to recommend any one without big chance of bad feedback later ;-)

[Wacholek]

I have to check Minerstat. Since I have only 2 rigs its going to be free for me.

Now I know where I have problem with one GPU. Temperatur control. Minerstat reads the GPU temp. Every time when there is a temp read this GPU stops hashing. I have had similar problem before and I have read about that before. I have to look for the solution.

[pintybor]

@Wacholek It is no problem, I can set intensity also up to 768 no problem using 14.4 driver, with newer drivers this is max possible intensity. But in any case setting it at higher value does not change the problem with no hashrate or invalid results. My auto generated amd.txt is:

"gpu_threads_conf" : [ // gpu: Cayman memory:1663 // compute units: 24 { "index" : 0, "intensity" : 768, "worksize" : 8, "affine_to_cpu" : false, "strided_index" : 2, "mem_chunk" : 2, "unroll" : 8, "comp_mode" : true },

],

/*

• Platform index. This will be 0 unless you have different OpenCL platform - eg. AMD and Intel. */ "platform_index" : 0,

[Wacholek]

@pintybor One mor think. Do you use any kind of program to read temp of the gpu/OC?

[pintybor]

No usually not. When I want to read temps and fan, I use AMD Overdrive inside catalyst software.

[Simaex]

@pintybor Try to install without AMD Overdrive. I have some problems with it but can't remember exactly how I resolved them. It may be the reason you GPU do not work.

[pintybor]

@Simaex I have now used DDU to clean all AMD related stuff and then only installed the driver manually using device manager from 14.4 package. There is no other GPU monitoring app now installed. Unfortunately the result is the same, Hashrate is 0 and GPU load is 0%. :(

[Simaex]

@pintybor Try low intencity like 100. If the GPU start hashing try to find maximum intencity that do hash. If it do not work try to use Linux with 15.3 driver as @Spudz76 suggest.

[pintybor]

@Simaex Even setting intensity as low as 16 does not solve the issue. :( I will try first using Win7 (I'm on Win 10 now) and if that doesn't work I can also try Linux to see if that helps but I have little hope... To which suggestion form @Spudz76 are you referring too??

[Spudz76]

Linux is fickle, you have to install Ubuntu 14.04 and a specific old kernel so it works with fglrx 15.3 and it ends up working but almost not worth it. I haven't had that rig powered up for 9 months. GPUs too old to bother.

[Spudz76]

However, Linux doesn't require dummy plugs. If you don't have your GPUs connected to outputs, they will not show up for OpenCL in Windows. Apparently back then, "headless compute" wasn't even a thing.