Bitcoin Forum

Hello,

I didn't want to highjack Claymore's thread to discuss something off topic.

Since miners are kind of forced to use AMD's drivers ver 15.12 & 16.3.2 for best performance or sometimes even for the miner to work at all
Is there a way to install a newer driver then copy/overwrite the needed old OpenCL stuff from 15.12 ?

Example, my triple Radeon Pro Duo setup will never ever install all cards with old drivers.
Newer drivers installs fine but with either crashing the miner, hashing @ 0, or best case with 16.12.1, hash at half speed.
Also facing the exact issue with 4x Sapphire's Fury Nitro


Any ideas ?

yea i have an idea, why dont you try it  you lazy fuck...

You can do that if the version difference between the old and new driver is not too different. You just need to try.

@AlphaSun I tried multiple things, none worked.

I will test this : http://cryptomining-blog.com/tag/amd-opencl (http://cryptomining-blog.com/tag/amd-opencl) but my guess it won't work as i don't see any opencl dll's in the miner (Claymore's) directory itself.

• Installed 16.12.1, all cards detected and Claymore's miner hash at half speed.
• Extracted the files from the 16.3.2 exe.
• Replaced amd_opencl32.dll,  amd_opencl64.dll, amdocl.dll & amdocl64.dll in system32 drivestore with the ones from that old driver.
• Claymore's miner now shows error : no opencl gpu's found

Is this even possible ? or am i just trying something that does not make sense to begin with.

Maybe you need to use DDU to uninstall and then reinstall. Or you need to have a clean OS.

have you tried to lower the intensity

@cracker666, lowering the intensity/ disabling asm will end up with half the hashing speed.


Windows 10 aside for now. I am testing the setup under linux (EthOS and Simplemining.net).
Hashing at full speed, OS sees all 3 Pro Duo's (total 6 GPU) but Claymore sees only 5.  ???




37 36 39 44 36 39 ℃
70 0 70 0 70 0 %

 GPU #0: Fiji
 GPU #1: Fiji
 GPU #2: Fiji
 GPU #3: Fiji
 GPU #4: Fiji
 ZEC - Total Speed: 2222.592 H/s, Total Shares: 820, Rejected: 1, Time: 00:12
 ZEC: GPU0 461.767 H/s, GPU1 439.077 H/s, GPU2 436.302 H/s, GPU3 448.697 H/s, GPU4 436.749 H/s
 GPU0 t=41C fan=70%, GPU1 t=40C fan=0%, GPU2 t=46C fan=70%, GPU3 t=50C fan=0%, GPU4 t=43C fan=70%, GPU5 t=47C fan=0%
 ZEC: 03/18/17-21:55:25 - SHARE FOUND - (GPU 0)
 ZEC: Share accepted (941 ms)!
 ZEC: Share accepted (1090 ms)!
 ZEC: 03/18/17-21:55:27 - SHARE FOUND - (GPU 0)
 ZEC: 03/18/17-21:55:27 - SHARE FOUND - (GPU 0)

Hi cracker666,
I've been following your problem with the 3 x Pro Duo with interest as I have 4 x Pro Duo's on the way!
I was hoping to put them in a single rig!

I have a few questions / suggestions:

1) Are you still using the same Z170A XPOWER GAMING TITANIUM EDITION motherboard?

2) Are you using the latest version of EthOS (1.2.0 released 3 days ago)?
I note that this includes an update to latest consumer card AMDGPU 16.60 linux drivers which don't appear to support Pro Duo.
AMD site shows only workstation linux drivers as supporting Pro Duo http://support.amd.com/en-us/download/workstation/previous/detail?os=Linux+x86_64&rev=15.302.2001 (http://support.amd.com/en-us/download/workstation/previous/detail?os=Linux+x86_64&rev=15.302.2001)

3) Have you tried the latest version of Pro Duo Windows Drivers (17.3.2 released 4 days ago)

Good luck!
I'm hoping there is a solution.

@Collycoin

Hi there,

To save you time and to make you at least start where i stopped, this is what i tried so far :

MSI Z170 Xpower
MSI Z170 Mpower
MSI Z170 Gaming 7
MSI Z170 Gaming 5
MSI Z170 Gaming 3
ASUS X99E-WS
ASUS Z9 PE-D8 WS
Single EVGA 1600w, Dual EVGA 1600w

It's clearly a driver issue, at least in my case.

The only drivers which allow me to install all 3 cards either via the application setup or via manually via device manager are

Crimson ReLive Edition 16.12.1 499 MB
http://support.amd.com/en-us/download/desktop/previous/detail?os=Windows%2010%20-%2064&rev=16.12.1 (http://support.amd.com/en-us/download/desktop/previous/detail?os=Windows%2010%20-%2064&rev=16.12.1)

Radeon Pro Software Crimson ReLive Edition 16.12.1 630 MB
http://support.amd.com/en-us/download/workstation?os=Windows%2010%20-%2064#pro-driver (http://support.amd.com/en-us/download/workstation?os=Windows%2010%20-%2064#pro-driver)

The issue is, this driver will crash claymore at default settings, if you disable asm it will hash at half speed.
Tried optiminer, works with the default settings at 300ish/sol too.
Nicehash benchmark using Eth sgminer gives full speed though @ 28h/s


In EthOS and Simplemining.net, its a bit better. 5 GPUs worked at full speed, sadly i am not a Linux guy so i have no idea how to tinker with it or update/swap drivers.

I didnt try 17.3.2 for windows yet, will test it today




 

Pro Duo's work great.  There are a few nuances that need to be changed.  First is they will throttle down at a very low temp unless the Bios is changed.  Second is they are splitting the PCIe rail so you need to change those numbers in the Bios as well to get maximum hashrate or it will throttle for that also.  Thirdly, the workstation drivers etc work fine but the Pro Duo's do best in linux.   I have mine running at about 980 mining ZEC and I can easily get them stable to 1050+.  

Of course I also mod the bios for the above reasons and also the memory straps and a -62.5mV to -75mV offset at 1050+ core and a much larger offset if in the 950 core range

BTW- I had all kinds of troubles with 16.32 on a WIndows machine so I just dumped the windows  :o

Last point- to run 3-4 Pro Duos with stock ROMs you need plenty of PSU

oops, i almost forgot.  Remember there are 2 dies per Pro Duo and you need to factor that in when placing them on a mobo because of the PCIe lanes.  Also on linux it works best with a PCIe delay of about 96ms (i think that is what i have set)

Hi Crytominer,
Well its definitely not a motherboard issue -that's some serious motherboard testing  :)

I was just re-reading your Claymore extract above.
It IS seeing all 6 GPUs in the temperature & fan % readings, just not in the Hashing reading.
So maybe it is a Claymore issue - you should try raising it with him.
In any case with the Pro Duos now available for $800 they are a good Zcash mining option which more people will be using and Claymore should support.

I also suggest you try the official AMD linux drivers for Pro Duo (FirePro which I linked in my first post).

Did you have all 6 GPUs working on Nicehash benchmark using Eth sgminer? (should be doing better than 28 though - would expect 31+ 62 total) See https://forum.ethereum.org/discussion/6707/radeon-pro-duos (https://forum.ethereum.org/discussion/6707/radeon-pro-duos)

If not you could have a bad GPU?

Thanks kilo17.

When you say Pro Duos are best in linux do you use the FirePro workstation drivers (Ver 15.302.2001) or the AMDGPU (Ver 16.60) drivers?

I'm looking at 2400 watt of Platinum 12v PSU for the 4 x Pro Duo cards and ~500 watt ATX supply for the motherboard.

What miner are you using to get 980-1050 sol/s on ZEC?

Can you explain further re card placement on mobo.
I'm planning to use a MSI X99A SLI PLUS Motherboard with all 4 cards in the 4 PCIe 3.0 x16 slots. 3 will be mounted directly and the 4th via a PCIe x 16 riser ribbon.

I'm glad to hear that this can all be made to work and that you're getting excellent ZEC sol/s.

 ZEC - Total Speed: 2677.347 H/s, Total Shares: 184, Rejected: 2, Time: 00:02
 ZEC: GPU0 455.432 H/s, GPU1 459.048 H/s, GPU2 435.766 H/s, GPU3 439.702 H/s, GPU4 445.887 H/s, GPU5 441.513 H/s


The good news, IT WORKED !
The bad news, no fan control.
simplemining.net OS stopped showing temps and the fans were all at idle, almost BBQd the cards. Too tired to investigate further for today.

How it worked ?
Think about the silliest thing ever, now think about something even sillier !

The suspense is killing me!

The solution was......wait for it......  plugging the monitor to the built-in intel gpu  :o
This made it work under smOS, probably will work for EthOS and any other Linux distro,

Windows is still a no go.

Coming back to this thread. Cryptohminer what's the latest?

I am trying to decide whether to build a 3 card rig with these cards. I can buy them at $585 refurbished from a local supplier. Or I can get Nitro+ 470 4G for $155 each. If you had these choices and you could do it all again, what would you do?

Great feedback. Thanks kilo17.

Are those BIOS roms available somewhere? Or proprietary?

As I posted above, I have the chance to pick up 3 cards for $585 each from a local shop. Debating if I should go for it or not... Appreciate any advice.

Ended up doing 2 cards per rig. Go Nvidia if you want less headache.

I've also been unable to get more than 2xPro Duos working properly in Windows 10.

Problem seems to be be in BIOS which can't allocate enough resources with "Above 4Gb decoding" disabled.
Enabling this option seems to cause all sorts of problems with the Pro Duos, including making BIOS inaccessible thereafter (requiring CMOS reset).

Without it, best I have managed is 5x Fiji with 3 Pro-Duos' installed (wasting half a card) or 4 x Fiji + a RX470 (which can't use same drivers so no=go)
I once got 6 cards recognised briefly by disabling the USB ports but then couldn't use a keyboard/mouse. :-)

I've tried with both X99 and H270 mobos. Hoping for better luck with Z170 Gaming Pro Carbon but I hear that Fijis are a problem there too.

16.10.1 (I think that's the WHQL cert version of the 16.10 range) works reasonably well in my testing of it.

 ALL of the "relive" versions I have ever tried have been lower performance bloated JUNK.

It's not about PCI lanes, the same X99 WS motherboard which runs 7x 1070 or 7x 1080 Ti's will not run 3 Duo's unless you use the newer AMD drivers which reduce the mining speed by half.

 Have you tried any of the newer versions OTHER THAN 16.12 series, which had KNOWN ISSUES on all cards in a mining setup with poor performance?

 There's a reason I specified the 16.10.x series (and the WHQL version in particular).

I have not had success with the new blockchain driver. Just wondering if that was just me

 Use 16.9.2, it matches the performance of 15.12 very closely even on old GCN cards while supporting up to the RX 4xx series.

 The new "blockchain" driver is a disaster and a waste of time in my limited testing of it.
 It worked on my RX 470 cards but NOT well, and didn't even INSTALL on anything older.


 After doing some machine building this week, I was reminded that 16.10.1 was NOT the WHQL version I was thinking of, does not appear to support APUs at all - though it DID work OK on older discrete GPUs when I was testing driver versions.