Maybe you will sign driver at least once? If it will be revoked - then let it be...
I still think it's not a good idea to waste kernel certificate. Oh Claymore i allive again  Good, very good. Keep up your good job Thanks good lord, Claymore, I have rx480 8 gb samsung, using latest 14th ver and reaching only 28.8mh/s, any help to see those 33mh/s?
I use "-strap 1@2250" or even "-strap 3@2250" for these cards, read Readme for more details. |
|
|
|
2 Claymore
Please, how to uninstall your driver and switch back windows from test to normal mode?
And about "paid straps from Eliovp" - I think everyone can examine them with Eliovp's tweak utility while your miner is running. Or are there some hidden gems?
1. Use "-driver uninstall" option with admin rights and reboot. Check Readme for details. 2. Registers contain only part of original strap, some values are modified. |
|
|
|
|
I uploaded a small update to fix issue with fan in latest AMD drivers. I did not change the version number.
Please redownload v14.0 and check the log file, it must contain "b707" at the beginning which means latest build number.
|
|
|
|
And another one question: you made memory straps by yourself? Or you take it from somewhere? Can you share this straps for us, so we just flash it on our bioses. Those who can't use driver?
P.S. Some time ago zawawa (he maked GGS miner) can apply memory straps in his miner without any drivers on the fly... So it's exist a way to do it so...
Eliovp sold me his straps for my miner, I'm not going to disclose them, sorry. Zawawa's app is very similar to Eliovp's amd memory tweak app, i.e. it just changes memory registers, it does not apply entire straps. As soon as you try to copy entire strap to these registers, you will get BSOD. Except my miner, currently there is no software (at least public) that can apply entire straps on the fly. If I have already flashed the vbios of my 580 and 570 cards, do I need to restore the stock bios before attempting the custom -straps in the command line?
You can use your current bios. it seems -tt won't work on 19.4.3?
You are correct, AMD added another API update to manage fan. Again... |
|
|
|
How much hashrates for ETH were you able to pull out of 580 8gb cards with these new straps?
Not huge numbers, something about 33MH on RX480 8G Samsung with stock BIOS and voltage. So, without disabling secure boot no chances to run with straps, right? It's not good...
Do you really need secure boot on your rigs? I could sign the driver, then some people will use my miner as malware, as always, then some AV will mark it as virus, then the certificate would be revoked... No way. Any plans to implement this feature for linux version ? (as far as its possible)
I'm not going to do the this trick in Linux, sorry. |
|
|
|
until memory clock will go idle (in a few seconds), then enable GPU again in the miner, new clock will be applied after that.
It will be the same as restart miner? I remember that at startup miner try ti find best dcri value, so trick with enabling/disabling GPU with new memclock may be not best way, because optimal DCRI for this card with new memclock may change? Or optimal DCRI may change only if CORE clock have changed? It depends... Anyway you can use "+" and "-" keys to change dcri to see if they are good for new clock. Or just set fixed -dcri, find highest stable clock and then find best -dcri value for this clock. |
|
|
|
Secure boot didn't allow to load unsigned drivers with UEFI bios.
I tried to run miner with admin rights but I got error:
In log:
23:39:57:796 4e14 Driver is not installed! Trying to install it...
23:39:57:854 4e14 Failed, you must run miner with admin rights.
On screen system error message appears before "Failed..."
"Value protected by secure boot policy and can't be changed"
Oh, you are correct, I completely forgot how it must work. If you spend so much time, how is the fastest way to found the best strap/mem clock?
Is there way to found max mem clock without restarting whole miner? As I remember, when I overclock memory, hashrate go up only after restartminer... so restart miner every time is slow... maybe you found beter way for fast applyong new memclock and see new hashrate without restarting miner?
I spent much time in attempts to implement this trick, not in overclocking. Anyway, use WattMan or AfterBurner for overclocking. Put hdmi dummy plug into card and it will apply new clocks immediately. Or set new clock and disable GPU in the miner ("0" key for first GPU, "1" for second GPU, etc) and wait until memory clock will go idle (in a few seconds), then enable GPU again in the miner, new clock will be applied after that. |
|
|
|
Sorry for asking , whats the dev fee on this beta version ?
Please read Readme. |
|
|
|
I think it's a good idea to quote Readme about "-strap" option. In general, you use "-strap 1" value and then try to overclock memory to see the highest stable clock and hashrate. Then you can try "-strap 2" and do the same. After these tests just select the best strap index and clock. At the same clock "-strap 4" will show better hashrate than "-strap 1" but it will not support very high memory clocks, so final hashrate after oveclocking can be lower than "-strap 1" with very high memory clock. -strap applies specified memory timings (strap). This option is available for Windows only and requires AMD drivers 18.x or newer (most tests were performed on 19.4.3).
Currently only AMD Polaris cards are supported, support for other cards will be added later.
Miner has built-in straps database, all straps are separated by memory (4GB or 8GB) and memory type (Samsung, Elpida, Hynix, Micron).
Straps are sorted by intensity, i.e. "-strap 1" supports higher memory clock than "-strap 2", etc. For the best hashrate you must also set high memory clock, so "-strap 1" is a good start point for tests.
You can specify just strap index, for example "-strap 1" will apply first strap from database for all Polaris GPUs based on GPU memory size and memory type, miner will show full strap name detected.
Or you can specify strap directly in format "POL8S1": "POL" means Polaris, "8" means 8GB, "S" means Samsung memory, "1" means index.
Zero index means default strap from VBIOS, i.e. no strap is applied.
You can also use "@" character after strap to specify memory clock, it works like "-mclock" but overrides it, for example, "-strap POL4E2@1900".
You can also specify values for every card, for example "-strap 1@2100,POL4H3,0".
If strap is applied, miner will return old strap and memory clock when miner is closed.
The best approach to find best strap is to set "-strap 1,0" (it sets strap #1 for first card and no straps for the rest of GPUs) and then raise memory clock to see what clocks and hashrate you can reach.
Then so the same for strap #2 etc.
You can also specify raw strap string (96 characters). Note that single option value means that this strap is applied for all GPUs, use "0" to apply strap on single GPU,
for example "-strap 0,1@2200,0" applies strap #1 and memory clock 2200MHz for second GPU only.
NOTE: if specified strap fails, Windows is crashed. After reboot default timings are restored and you can try some different settings. |
|
|
|
|
Forgot to mention: this version uses straps that were provided by Eliovp, they must be very good.
|
|
|
|
Thanks!
But if secure boot enabled driver wouldn't be installed, right? Only disabling SB can help?
Without your driver straps can't be applied?
P.S. Why 14th version? There is no 13th...
Secure boot is not related to the driver. Miner will set "Test Mode" in Windows to use unsigned driver. I don't like "13" number so I skipped it It is an improvement, but unfortunately I'm still experiencing issue with Chinese directory names.
Sorry, I did not have time for it, though I checked this issue and confirmed it. The reason is that the miner does not use unicode, it will take some time to convert all paths and filenames to unicode strings. Claymore, I know I may be asking a lot here, but just in case it's not too much additional effort... The Radeon Pro Duo is technically Polaris. It's got Samsung Mem (16GB on each GPU); two WX7100s/RX 580s.
Was hoping your straps would work without having to try bios flashing the master and slave gpu bioses, but the miner instaquits it I run as admin with -strap 1. If I don't run as admin, miner runs fine, but with no straps. Believe it's related to errors reading the boot configuration data.
Driver is latest Pro Driver, 19.Q1.2. Have used the AMDMemTweak CLI from Eliovp, and it works, but would like to try the straps you are using in your exe. Others messing with AMD workstation gpus may get some good here too! Thanks!
Check the log file for details. Or create "config.txt" file with options and right-click EthDcrMiner64.exe and select "run as admin", it must work. Or start "cmd.exe" as admin and start miner from it to see what happens, miner cannot close without any messages. Any hashing performance increase vs latest 12 on linux?
Nothing changed for Linux version so v12.0 is the latest for Linux. |
|
|
|
Will we see an nvidia version for 'straps'? the Eth-pill exists but if there's soft-mod for AMD cards surely there's something for Nvidia?
Yes, I will add Nvidia support as well. This magic with straps took me a lot of time and efforts, so I will implement it step by step. |
|
|
|
|
v14.0 Beta:
- Windows only: added ability to apply optimized memory timings (straps) on-the-fly, without flashing VBIOS, up to 20% speedup compared to stock BIOS. Currently only Polaris cards are supported, other cards will be added later.
Drivers 18.x or newer are required. Best straps for Ethereum are included. Check "-strap" and "-driver" options for details.
- Windows only: added temperature/fan/clocks management for Radeon VII.
- a few minor bug fixes and improvements.
|
|
|
|
|
Finally. I will release new version within 24 hours.
|
|
|
|
I have same issue with AB, restarting miner twice always helps. That would be easy fix for 1st beta.
It seems you are right, it helps. I will try to find some workaround better than restarting. |
|
|
|
|
New version is almost ready, but during testing I found some strange issue: sometimes if a card has no display connected I cannot change its memory clock. I tried both WattMan and AB to change memory clock, both show that new memory clock is applied, but no changes in hashrate until I use hdmi dummy plug. This is serious issue because most straps are good only with memory overclock. I use 19.4.3 drivers and see this issue on most Polaris cards. But sometimes it works for some reason...
So new version is postponed until I find some workaround (or confirm that there is no way to fix it), it's not a good idea to use hdmi plug for every card in rig.
|
|
|
|
Claymore, another one question: can you make double memory usage to increase a bit eth speed? Like Phoenix maded?
In my tests double memory usage shows zero speedup. |
|
|
|
|
I don't like flashing bios so I solved this problem finally.
v12.1 is almost ready, I'm going to release it in 3-4 days.
It cannot perform runtime tuning of memory timing registers, it will be added a bit later, in another update.
But it can apply entire straps on-the-fly (Windows only). Currently for Polaris cards only, may be later I will add support for other cards too. So you can use stock vbios and don't flash your cards to try new straps.
Instead, you just put new strap string to config file and the miner applies it immediately in runtime.
If specified strap is not working it will not cause any problems, after reboot you will have stock strap again because bios is not changed.
This way you can keep stock bios and also test new straps safe and easy.
|
|
|
|
What you did (and Claymore is going to), is a direct consequence of eliovp releasing his open-source tool. You could have been giving him credit at least before putting it (in whatever form) into your paid miner.
Just to clarify: Yes I'm going to use some parts of Eliovp's code/work but I have his permission for that. |
|
|
|
Got in the problems while adding new features? No hurry but... No worries, it just takes more time than I expected, so I'm still working on it. |
|
|
|
|
Show Posts
