Pages:
Author

Topic: Crimson Relive Win 10 Drivers Now Support 12 AMD GPU's with Compute Workloads (Read 7904 times)

brand new
Activity: 0
Merit: 0
Hello there.
I'm going crazy.
8 amd, 5 nVidia I was doing smooth Mining. (Asus b250)
But yesterday I wanted to do mining with 12 AMD cards. I did not do what I did. It is reset immediately after Windows is turned on. I have installed windows10 many times but nothing has changed. So which AMD drivers should I consume?
newbie
Activity: 4
Merit: 14
Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\0000]
"KMD_EnableInternalLargePage"=dword:00000002

And for 0001, 0002, etc.

I created a simple tool for made this easier  Cool
https://bitcointalksearch.org/topic/tool-amd-compute-switcher-for-switch-automatically-to-compute-mode-2815803

If you have updates for it please inform me  Wink
newbie
Activity: 1
Merit: 0
Confirmed 13 gpu on whql-win10-64bit-radeon-software-crimson-relive-17.11.1-nov10
Need turn off some cards first (in device manager) to enable compute mode on all cards because theres bug. Once enabled on all, turn on all cards and it stay.
https://ibb.co/jkYpmb


Could you give more info, im trying for days to make 12 gpu work but i just cant...What version of windows you use 1703,1709? What motherboard you use? What version of claymore?
newbie
Activity: 25
Merit: 0
y but AB still dont work, i was getin 29.5mhs with rx570 nitro+, when i hook up 12 gpus, and do everything, everything and still ab dont work, and cards hash at 25.5mhs
sr. member
Activity: 2142
Merit: 353
Xtreme Monster
Confirmed 13 gpu on whql-win10-64bit-radeon-software-crimson-relive-17.11.1-nov10
Need turn off some cards first (in device manager) to enable compute mode on all cards because theres bug. Once enabled on all, turn on all cards and it stay.



Mind sharing the computer specs?
newbie
Activity: 42
Merit: 0
Confirmed 13 gpu on whql-win10-64bit-radeon-software-crimson-relive-17.11.1-nov10
Need turn off some cards first (in device manager) to enable compute mode on all cards because theres bug. Once enabled on all, turn on all cards and it stay.
https://ibb.co/jkYpmb
sr. member
Activity: 2142
Merit: 353
Xtreme Monster
EDIT: figured it out KMD_EnableInternalLargePage - now i'm getting 29 from all 12 cards on windows

I guess this is the dag fix setting.
jr. member
Activity: 155
Merit: 4
is there any way to set cards to compute mode using command line or registry or somehing? It takes 10 minutes for each card in amd settings.

for now stay away from this driver. switching to compute mode is unbelievably annoying! made reg files to do this faster - somehow works. msi afterburner beta21 does not work at all. trixx 6.5.0 undervolting resets driver. watman is crap. hashrates are kinda same.  power consumption went up. tools for undervolting are less.
How?
I have 12 cards setup and if i'm unable to set compute mode in 30 minutes i'm, returning all 12 cards as there is no OC in linux that works, drivers where 12 cards work for windows does not support OC in claymore and needs compute mode to be set up, which is super annoying in amd settings.
Drivers with no need of setting compute mode supports 8 cards max.

EDIT: figured it out KMD_EnableInternalLargePage - now i'm getting 29 from all 12 cards on windows
member
Activity: 248
Merit: 26
I'm using the 17.10.4 drivers, Hashrate is the same as blockchain drivers. If you don't mine Ethereum, you don't need to switch from graphics to compute.
full member
Activity: 173
Merit: 100
Anyone find the COMPUTE button in Windows 7? Can't find it anywhere

Nope. Can someone find this button on newest drivers on Windows 7?
newbie
Activity: 1
Merit: 0
I need your help guys  Huh

I have 16x RX580 (6xAsus,4xMsi,6xSapphire) for 8+8 rigs

I tried blockchain and 17.10.2 drivers on win10(ver1703) but i cant add 6th or more gpu. 5 gpus working but when I add 6th, 2 gpus stop and dont work on claymore. I really want your helps I dont solve this problem.

My second problem is with my Asus Dual RX580 gpus. They are really unprofitable and unstable. I tried everything but I got just pain  Cry I got just 26.5MH with original bios which is most stable. The other 30+MH with less electricity.

I will be really grateful if you guys help  Kiss Thanks from now
jr. member
Activity: 98
Merit: 1
for now stay away from this driver. switching to compute mode is unbelievably annoying! made reg files to do this faster - somehow works. msi afterburner beta21 does not work at all. trixx 6.5.0 undervolting resets driver. watman is crap. hashrates are kinda same.  power consumption went up. tools for undervolting are less.
member
Activity: 119
Merit: 10
Quoting my self from another thread, so I can share my own experience with you guys:

Unfortunately I am having big problems with the latest drivers.

Even though I installed them and they are working great on my PC which has 1x RX 580 8GB Nitro+ (Hynix), my rig was a different story.

It took many, many minutes for the drivers to get installed. When they did, AMD Radeon Settings would change half of my cards to "compute" with no problem but for the other 4 I would have to restart my rig in order for the setting to apply.

Upon running Claymore miner, the rig became unresponsive while setting DAG file. What I mean by saying "unresponsive" is that it the cursor would remain blinking without moving on the next usual commands -creating DAG file-. In the log file the last entry was "start building OpenCL program for GPU 7". At this point, if and when I performed a single click on the desktop, the whole rig froze.

I have performed DDU and gone back to BlockChain Beta Drivers (with which my rig was performing well) and it was back to normal. To be sure I did nothing wrong I tried:

  • Going back and forth between BlockChain and latest Drivers 3 times
  • Earlier Claymore miner versions
  • Setting global settings to Graphics (instead of compute)

The symptoms remained. I do not have time right now to clean install Windows 10 and try again, I hope I can do so in the weekend. In the meantime, if anyone else is experiencing the same problems, please let us know.


My rig:

ASUZ Prime Z270-P
8x RX 580 8GB Nitro+ (Hynix)
Windows 10
Radeon™ Software Crimson ReLive Edition 17.10.2
Claymore Dual Miner v.10.1
Modded BIOS

I decided not to perform a clean install and go back to BlockChain Beta drivers
hero member
Activity: 1036
Merit: 606
Since the blockchain drivers don't support the Creators fall update, there is no point in updating to v1709. The blockchain drivers worked for me on v1709, but it made the Windows UI very unstable. If you want to stick with the blockchain drivers, it's best to stay on v1703 and defer the Windows feature updates in the Group policy editor until a stable version of the new Crimson Relive compute toggle drivers is out.

gpedit ==> Computer Configuration ==> Administrative Templates ==> Windows Components ==> Windows Update ==> Defer Windows Updates ==> Select when Feature Updates are received

There you can defer Windows version updates for up to one year.

While in the Group Policy Editor you can also keep Windows Update from updating the the drivers.

gpedit ==> Computer Configuration ==> Administrative Templates ==> Windows Components ==> Windows Update and set 'Do not include drivers with Windows Update' to enabled.

There is also a new Crimson Relive 17.10.3 driver out, but according to the release notes only game fixes were added.

https://support.amd.com/en-us/kb-articles/Pages/Radeon-Software-Crimson-ReLive-Edition-17.10.3-Release-Notes.aspx

The release notes also mention the new drivers drivers have a known issue with 12 GPU's in compute mode.

Quote
A random system hang may be experienced after extended periods of use on system configurations using 12 GPU's for compute workloads.
newbie
Activity: 13
Merit: 0
Allright.. these are my results from yesterdays extensive testing

First of all:
AsRock H110, 8x PowerColor RX470 4GB (samsung), Claymore 10

Windows 10 Pro 1703, AMD 17.10.2 with Workload = compute

We had the rig running stable for weeks/month with BlockchainDriver(23aug).. 228MH's on ETH with this crappy cards
We did DDU, installed new driver 17.10.2(oct23) (full installation with GUI) / took like 15 minutes ! restart + PixelPatched + restart
Manually setting Workload = Compute for all the cards .. grrrr!!! after setting all 8, you have to restart the computer !
After start: OverdriveNTool 0.2.0 for aplying OC/UV profiles to the cards..
result is very same like the BlockchainDriver(23aug) .. 228MH's .. stable for 24hours

So my insight on this.. for those running 8 card rig with BlockchainDriver(23aug) there is no point of switching to the 17.10.2 .. there is no improvement, just the pain (aplying Compute ON) to even get the same result

-----
COMPLETE DDU 17.0.7.7

Adding 4 more PowerColor RX470 .. sooo .. 12 Cards in total (http://prntscr.com/h31ld0)
Windows booted up quickly
Installed full version of 17.10.2 took like 40 minutes !!! Insanly long! restart + patch
All cards has been detected, I saw 12 in Device Manager (http://prntscr.com/h31kuu)
manualy turned on compute workload (took another 10minutes) + restart
Computer freezes.. restart, freezes, freezes..
After a long while it booted without any changes at all
Saw all 12 cards, modifified Overdrive batch file to OC/UV all my 12 cards + applied this profile
Ran Claymore 10.1 with -gser 2 option
All 12 cards has been recognized by claymore (http://prntscr.com/h31l1y)
It started mining while still creating DAG for another cards.
I saw that some of the card started mining 28.5 Mh's (that is normal value for my cards)
Than.. Cannot create a DAG file for GPU9 and GPU3.. Claymore restarted.. since than, I was not able to start it over again
After another computer restart I was not able to boot into Windows anymore!!! Windows died!
Disconnected those 4 additional cards and here we go.. windows booted up again..
Ran DDU + restart + back to the BlockchainDriver(23aug) + patch + restart
Cards started to mine ony 14MH's no matter what.. spend like an hour playing with it.. there is some problem with applying cvddc/mclock after DDU uninstalling 17.10.2 and reverting back to Aug23 drivers
Windows installation is obviously corrupted (or the new DDU 17.0.7.7 is leaving some traces behind after uninstalling the 17.10.2(oct23) driver

Had to restore Windows installation image from backup!

So, anyone that don't want to waste the time, stay away from 17.10.2 while running 8+ cards (especially 12!)

To that reg fix..
IojkinKot has been faster, I do agree with his findings.. there are more changes but only that one seems to be relevant to me
If anyone can tripplecheck, here is exported regedit key before/after applying COMPUTE mode. Do the file comparison and please let us know the result
Applying this regedit can save 40min of installation time + another 10 minutes of manually applying compute mode for each GPU
This might actually get very important.. if AMD will stay with such a option of turning Compute mode ON in the future!!!

https://drive.google.com/open?id=0B5E2YhEbdRIxUXZpejNaajNnc2s

And such a nice Friday it was  Embarrassed (6 hours wasted!)

PS: I'm NOT doing this test over again with 1709 Fall creators update !!
PS2: For those who want to say thank you, you've saved my a lot of time and pain, look down bellow the line Smiley
hero member
Activity: 1036
Merit: 606
New version of Afterburner v4.4.0 beta 21 is out. It increases the core undervolt limit to -200 mV for Fiji, Polaris and Vega. It's working on Windows 10 Pro v1703 with the blockchain drivers. Can anyone using the new Crimson Relive v17.10.2 drivers with the compute toggle on the Windows 10 v1709 Fall Creators update verify whether overclocking is enabled? Thanks.

https://forums.guru3d.com/threads/rtss-6-7-0-beta-1.412822/page-46#post-5486250

Quote
Changes list includes:

· Minimum voltage offset has been extended to -200mV for AMD Fiji and Ellesmere GPU families
· Improved drag and drop implementation for the list of active hardware monitoring graphs. Now it is possible to drag and drop items below the bottom edge of the list to move the graphs to the end of the list
· Improved multiple selection functionality in monitoring properties:
· When multiple graphs are selected you may press + inside group name edit field to apply group name changes synchronically to all selected graphs
· Now “Override graph name” and “Override group name” checkboxes are also affected by multiple selection functionality, so you may hold when toggling those options to toggle it synchronically for all graphs or hold when toggling those options to toggle it synchronically for all selected graphs
· Now group names support macro definitions (the same as prolog/epilog in OSD layout editor), so for example you may override group name for GPU temperature and set it to %GPU% to utomatically display your GPU name as a group name in OSD
· Added %GPU1%..%GPU8% and %FullGPU1%..%FullGPU8% macro names support for mGPU systems
· Now alarm zone is visualized on monitoring graph instead of displaying just the threshold. This simplifies visual identification of min/max thresholds
· Now you can define closed alarm range if specified maximum alarm threshold is less than minimum
legendary
Activity: 3738
Merit: 1708
Anyone find the COMPUTE button in Windows 7? Can't find it anywhere
newbie
Activity: 63
Merit: 0

So you can OC memory with Overdriventool? Someone says that it's not working. Is GPU-Z showing those memory clocks? HWinfo64 not always show the right info, but I only had problem with showing voltages there.
Are you using 0.2.0? There is newer version 0.2.1.

Yes, I had version 0.2.0 installed and running with blockchain drivers (23aug).
Just did the DDU, installed new driver (complete suite due to needed workload (compute) change using GUI) .. patched, restarted, changed Workload to compute for all GPU, retarted again ..
OC/UV profile from OverdriveNTool are started automatically (via batch file) when the Windows starts

As I said previously, no performance drop, all working just fine with 8 cards..
Going to play with it tonigh and will try to make 12 or 13 cards works on AsRock H110

PS: Also, will try to setup compute workload using regedit.. I've been suggested a solution so no AMD GUI will be needed if that works

Hi, can you share you reg file for compute mode ON ?

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\0000]
"KMD_EnableInternalLargePage"=dword:00000002

And for 0001, 0002, etc.
newbie
Activity: 32
Merit: 0

So you can OC memory with Overdriventool? Someone says that it's not working. Is GPU-Z showing those memory clocks? HWinfo64 not always show the right info, but I only had problem with showing voltages there.
Are you using 0.2.0? There is newer version 0.2.1.

Yes, I had version 0.2.0 installed and running with blockchain drivers (23aug).
Just did the DDU, installed new driver (complete suite due to needed workload (compute) change using GUI) .. patched, restarted, changed Workload to compute for all GPU, retarted again ..
OC/UV profile from OverdriveNTool are started automatically (via batch file) when the Windows starts

As I said previously, no performance drop, all working just fine with 8 cards..
Going to play with it tonigh and will try to make 12 or 13 cards works on AsRock H110

PS: Also, will try to setup compute workload using regedit.. I've been suggested a solution so no AMD GUI will be needed if that works

Hi, can you share you reg file for compute mode ON ?
full member
Activity: 1148
Merit: 132

So you can OC memory with Overdriventool? Someone says that it's not working. Is GPU-Z showing those memory clocks? HWinfo64 not always show the right info, but I only had problem with showing voltages there.
Are you using 0.2.0? There is newer version 0.2.1.

Yes, I had version 0.2.0 installed and running with blockchain drivers (23aug).
Just did the DDU, installed new driver (complete suite due to needed workload (compute) change using GUI) .. patched, restarted, changed Workload to compute for all GPU, retarted again ..
OC/UV profile from OverdriveNTool are started automatically (via batch file) when the Windows starts

As I said previously, no performance drop, all working just fine with 8 cards..
Going to play with it tonigh and will try to make 12 or 13 cards works on AsRock H110

PS: Also, will try to setup compute workload using regedit.. I've been suggested a solution so no AMD GUI will be needed if that works

i did the same as you however i cant change mem clocks after the install, by any means , wattool, overdrive tool or ab
any ideas? this is on two 10 gpu rigs all a mix of 480, 570 and 580
Pages:
Jump to: