Invalid Shares with AMD 19.2.2

[MohitYadavGitHub]

Same settings for xmr-stak gives invalid shares with Vega 64 on 19.2.2 and works completely fine with 18.5.2 . Cant really say its a xmr-stak issue, but maybe useful for people.

Also got a vega vii which has invalid shares with 19.2.2 (about 80% valid). I guess with vega vii dont have the option to go back to 18.5.2

[psychocrypt]

please run xmr-stak -V and post the version here

[MohitYadavGitHub]

[MohitYadavGitHub]

sorry had closed by mistake.

[psychocrypt]

mhh other user reported that 19.X and 2.8.3 is working together. if not you must downgrade the driver :-(

MohitYadavGitHub notifications@github.com schrieb am Sa., 23. Feb. 2019, 12:36:

sorry had closed by mistake.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/fireice-uk/xmr-stak/issues/2257#issuecomment-466641032, or mute the thread https://github.com/notifications/unsubscribe-auth/AYsxttrV__omCevyyG-zB_lvcf4LTl92ks5vQSe4gaJpZM4bLZel .

[MohitYadavGitHub]

downgrade works yep for vega 64 (18.5.2 is absolutely fine). Not sure its an option for vega vii since its just launched. maybe will give it a try in the next days. Will post the results here.

[MohitYadavGitHub]

just tried 19.2.3 . same issue. Didnt get time to try 18.x.x

[Spudz76]

Generally once a mining-specific function stops working, future drivers never fix it again. Client code must follow whatever AMD Software Departments latest idea is, without them really telling anyone.

After 19.2.1 they changed out the PCI driver maybe that broke things. Nearly impossible to figure out which driver is the oldest one with Vega20 chip support.

AMD is also horrible at any sort of historical documentation, it's like earlier software was an embarrassment to be forgotten (probably, because it is haha)

[heavyarms2112]

v8 was working fine on Vegas and Adrenalin 19.2.x and above drivers with xmr-stak 2.8.2 and #2241 fix. It broke again with 2.8.3+ version. @psychocrypt

[Spudz76]

Try the 2.10.0 release please

[heavyarms2112]

@Spudz76 i mentioned 2.8.3+ version. so it's broken for above versions. I already tried 2.10

[MohitYadavGitHub]

couldnt make the old drivers work with vega vii... With the latest 19.3.1 drivers ,still get a lot of invalid shares

ofcourse with 2.10.0

[MohitYadavGitHub]

One Side question , Does any one know how to use more than 8 gb of memory on Vega VIi while mining?

[psychocrypt]

which pool do you use? The strange thing is that other user report that all works fine. Could you please test tomorrow after the fork if maybe cryptonight_r is solving the issue.

[MohitYadavGitHub]

Supportxmr . Will try and let you know

[heavyarms2112]

@psychocrypt i haven't had success mining with 2.8.3+ version for any of the Vegas. Personally I know at least 5 folks who haven't had success. Which users are reporting it's working for them? Many switched back to 18.6.1 drivers or so. Vega VII works only on Adrenalin and above so that option is not available.

@MohitYadavGitHub FYR 2.8.2 with #2241 fix works fine for v8 on VII

[psychocrypt]

Chris in the Ryo channel reported numbers for vii and vega and tested my fixes.

[MohitYadavGitHub]

@MohitYadavGitHub FYR 2.8.2 with #2241 fix works fine for v8 on VII

Thanks :) . But i guess its not going to be useful after tomorrows fork.

[heavyarms2112]

Chris in the Ryo channel reported numbers for vii and vega and tested my fixes.

one user? imho it's evident that it's producing invalids from multiple reports here.

[Spudz76]

You can't possibly use all that memory, but to use more of it choose the largest scratchpad coins you can (CN-Heavy based, 4MB per block)

You will always run out of processor and memory bandwidth well before you fill more than ~3GB with cryptonights

If you need to use memory then go Ethereum, it's designed to waste memory instead of processor and width.

[psychocrypt]

Yes one user. It is not that I not thrust you that it is not working the problem is that I have not access to a windoes system and with chris I had a fast channel to communicate and we did many iterations until wie solved the issue.

Could you try if you get valid shares with xmrig-amd. If so we can start to spott differences. Are you able to compile the code by our own?

[MohitYadavGitHub]

All accepted ! so problem is unfortunately with xmr-stak :(

I am able to compile myself , if u want me to try a commit /branch

[psychocrypt]

ok thanks for the test I will provide a few test changes to you soon. Do you have also the issue with cryptonight_r?

[heavyarms2112]

@psychocrypt yes i can compile stak. I have tried xmrig and it works fine.

[psychocrypt]

Hi could you please test https://github.com/psychocrypt/xmr-stak/archive/windowsBug1.zip if it solves the invalid shares.

[heavyarms2112]

@psychocrypt will test and report.

initial tests on cn_rwz work fine. initial tests on cn_r work fine too. Looks like the issue is fixed. Thanks.

[MohitYadavGitHub]

Yes confirm that the issue seems to be fixed. Great !

[psychocrypt]

I will send you a second test because I removed something which should not be removed for linux. So I need you for a second test

[psychocrypt]

please test https://github.com/psychocrypt/xmr-stak/archive/fix-vegaWindowsAMDInvalid.zip for a last test

[heavyarms2112]

tested the above pvt build. works fine.

[WhiskeyBlocks]

I'm having the same issue with my Vega 56's. Any idea when the fix will be added to the full release?

[psychocrypt]

I think in 2 to 3 days. I like also to fix a few other issues. Maybe I will provide a beta version before.