Have pair of RX 560 in a Dell XPS-9100. One is in a 16x slot the other in a 1x with a riser. All was working fine on windows 7. I even installed the latest 19.1.1 Adrenaln for win7, dated Jan 10 2019.

Decided windows 10 had better security so I upgraded using that still freebie 10Upgrade program. As soon as the final boot occurred I spotted 4 RX-560s in the device manager instead of 2 and also found a boatload of error'ed milkyway tasks in boinc. Clearly this is a windows driver problem but am I the only one that has run into this?

I had the same problem on a pair of RX570 and gave up and put in Linux. Had same problem with Vega-64 on a system for one of my kids. Ended up removing boinc as I was just using boinc to burn in his computer.

I have had a lot of systems with HD7950, HD7850, S9x00 and windows 10 and never saw a problem like this.
right now, I am using revo uninstaller to get rid of all amd stuff and will try that jan20 win10- version of the 19.1.1 amd driver.

[EDIT] Made some progress: that REVO uninstaller took out the extra pair of RX560 that showed up in the device manager BUT BIONC STILL SEES 4 OF THEM!

Tried clean install of BOINC using revo uninstaller. Still have 4 RX560s. However, I checked another system that had HD7950 also a pair of RX560 in risers and that was working fine. It did NOT have that Jan 20 driver, the 19.1.1 which seems to have a problem. I need to find the 9/22/2017 one whatever that was.

Think I solved this problem, not sure exactly how I got into this mess or if the problem will stay fixed.

I went back to 2016 and got an old boinc and an old amd driver. Used revo uninstaller to get rid of all boinc and amd software.

Same problem. Had 2x as many RX560 as I should have. I remembered that the RX560 was released in mid 2017 but I had just put in 2016 drivers. This was suspicious.

I uninstalled amd software again, this time I removed the ethernet cable before rebooting. There was no possibility of windows 10 nor amd updating. Sure enough, I am back to a single amd RX560. No longer have that phantom.

I am guessing that during the boot up windows goes off to find a new driver and puts it in and that is why I got stuck with that extra gpu board.

Now I got to update boinc to the latest and make sure, somehow, that I don't get a 2nd phantom gpu when I go back online.

Would like to know if anyone else has seen this problem.

thanks for looking, hope I didn't bore anyone.

[edit] Note to myself: Be sure system if offline during first boot after adding additional riser video cards as microsoft goes off to find new drivers instead of using the exiting ones that work fine.

I went back to 2016 and....

Time machine? TARDIS? Special K Brand?

:)

Hello,
I have the same problem with 8970M. Already posted to Einstein@Home
https://einsteinathome.org/content/single-8970m-now-appears-and-runs-bit-faster-gpu0-and-gpu1#comment-170092
Thanks MMONNIN for redirecting to this thread.
@BeamBiker - how does it compare speedwise? Two WU serial on a single GPU vs. both of them in parallel on the two copies?

Here are my numbers
Task ID Workunit ID Sent Time Reported or Deadline Status Run time CPU time Granted Credit
LATeah1053L_172.0_0_0.0_36917685_0 395337611 11 Mar 2019 22:58:48 UTC 12 Mar 2019 0:08:32 UTC Completed and validated 2,047 135 3,465 Gamma-ray pulsar binary search #1 on GPUs v1.18 () windows_x86_64
LATeah1053L_172.0_0_0.0_36919316_0 395337612 11 Mar 2019 22:58:48 UTC 12 Mar 2019 0:07:28 UTC Completed and validated 2,046 134 3,465 Gamma-ray pulsar binary search #1 on GPUs v1.18 () windows_x86_64
LATeah1053L_172.0_0_0.0_36920947_0 395337613 11 Mar 2019 22:58:48 UTC 12 Mar 2019 1:58:10 UTC Completed and validated 1,125 110 3,465 Gamma-ray pulsar binary search #1 on GPUs v1.18 () windows_x86_64
LATeah1053L_172.0_0_0.0_36922578_0 395337614 11 Mar 2019 22:58:48 UTC 12 Mar 2019 1:35:37 UTC Completed and validated 1,130 104 3,465 Gamma-ray pulsar binary search #1 on GPUs v1.18 () windows_x86_64

So to get 2 WU I need 2050 seconds when run in parallel and 1125+1130 = 2250 s, when in serial on a one GPU. So there seems to be 200s (10%) advantage when running on a "ghost" copy in parallel with the main one.

Pawel

The Einstein project has a supported, official, route (via project preferences) to enable parallel processing of two or more tasks on any GPU. Two tasks per card is often a sweet spot, but it does vary by card. Search for posts on this subject by archae86.

Parallel processing is not necessarily advantageous at every project. If it works for you at Einstein, it would be wiser to use the supported route, rather than try to emulate the 'phantom GPU' (multiple drivers) route being discussed in this thread.