19.7.3 released now. I scanned through the release notes and the known issues and did not see any openCL issues. Not sure if it will fix anything but it is worth a shot.
19.7.3 released now. I scanned through the release notes and the known issues and did not see any openCL issues. Not sure if it will fix anything but it is worth a shot.
I am on the fence for a 5700 until I see some support for compute so your help testing is appreciated by myself and many others.
Thanks!
Thanks for this ace_quaker. I had spotted and download 19.7.3 but lacked the time to immediately test, having now done so as an 'over the top' upgrade to 19.7.2 and sebsequently as a clean install nothing has changed from the cruching perspective... As before, all tasks complete successfully but are ultimately deemed invalid when compared to the wingies result.
Lets hope next weeks driver addresses the problem! Until then please stay on the fence!!
Running a 5700XT in Linux. Drivers are a mess at the moment, but I did get it up and running.
I grabbed one bunch of tasks to see how it would run.
So far the results are the same as most other OpenCL workloads. Only running FGRP tasks on this host.
Thanks for your input TicToc, I was beginning to think I was the only one to have taken the 5700 plunge.
You have 1 task that has validated from the 82 you had chrunched at time of posting, that's 1 more validated task than I have recieved after doing several hundred!! What is interseting about your validation is that your wingman for that task also has a 5700XT but his is running on Windows OS. (I need to look at his other tasks to ascertain his validation rate).
Now I know of your (albeit single) validation and that there are other 5700XT owners out there I will fire my machine back up and hope to be paired with another owner of this GPU type to see what happens :-)
After persevering and crunching a lot of tasks I finally got a single valid on the 1st August. That task was interestingly validated against another host with a 5700 XT, in fact the same host TicToc had success with previously (although his result may have slipped off the server by now)...
My host is now using driver 19.7.5 and there is no no change from my earlier reports so I shut the machine in question down again a couple of days ago.
I'm now thinking I will wait for another driver iteration to see how that goes before nudging Bernd to take a look hopefully before my valid task disappears!
The question is: are Navi results completely wrong or they are just a bit different but still correct?
In my case for ~1 month and 400-600 tasks validation passed only 4 times when 2 of 3 GPU were 5700 ( https://einsteinathome.org/host/12764660/tasks/4/40 ), then Navi gave the same(wrong?) result and "win" by majority rule, because Navi GPU results differ from any(?) other GPU results.
If 5700 results were wrong we could miss potentially important information like new star etc (false negative). So einsteinathome team should recalculate all Navi data for all this 2(3) x Navi-cases.
Updated to the latest AMD Linux driver (19.30-855429). I am now running the 5700XT on the host rather than an Ubuntu VM, but other than much quicker run-times, the outcome appears to be the same. Tasks run to completion, but then are marked as invalid. It looks like the only changes were to move the Navi specific driver code into the unified driver stack and update the install scripts to support RHEL/CentOS/SUSE.
Current task status:
Valid: 0
Invalid: 6
Pending: 55
I'll let the rest of my queue complete, but it looks like there is nothing new in the driver, other than it being a unified driver for all recent AMD GPUs.
System also has a Radeon VII and a pair of Vega 64's, and they are crunching away on this driver without issue.
System also has a Radeon VII and a pair of Vega 64's, and they are crunching away on this driver without issue.
That's unfortunate about the invalids. Assuming that will get worked out eventually, how do task completion times compare among the 5700XT, RadeonVI and Vega64?
Ideas are not fixed, nor should they be; we live in model-dependent reality.
It seems that many OpenCL
)
It seems that many OpenCL applications are affected by this issue, here a report from Folding@Home : https://community.amd.com/thread/241222
Anandtech reported that most
)
Anandtech reported that most OpenCL compute applications failed during their review.
19.7.3 released now. I
)
19.7.3 released now. I scanned through the release notes and the known issues and did not see any openCL issues. Not sure if it will fix anything but it is worth a shot.
https://www.amd.com/en/support/kb/release-notes/rn-rad-win-19-7-3
I am on the fence for a 5700 until I see some support for compute so your help testing is appreciated by myself and many others.
Thanks!
ace_quaker wrote:19.7.3
)
Thanks for this ace_quaker. I had spotted and download 19.7.3 but lacked the time to immediately test, having now done so as an 'over the top' upgrade to 19.7.2 and sebsequently as a clean install nothing has changed from the cruching perspective... As before, all tasks complete successfully but are ultimately deemed invalid when compared to the wingies result.
Lets hope next weeks driver addresses the problem! Until then please stay on the fence!!
Gav.
Running a 5700XT in Linux.
)
Running a 5700XT in Linux. Drivers are a mess at the moment, but I did get it up and running.
I grabbed one bunch of tasks to see how it would run.
So far the results are the same as most other OpenCL workloads. Only running FGRP tasks on this host.
https://einsteinathome.org/host/12785597/tasks/0/0
Valid: 1
Invalid: 9
Pending: 72
Thanks for your input TicToc,
)
Thanks for your input TicToc, I was beginning to think I was the only one to have taken the 5700 plunge.
You have 1 task that has validated from the 82 you had chrunched at time of posting, that's 1 more validated task than I have recieved after doing several hundred!! What is interseting about your validation is that your wingman for that task also has a 5700XT but his is running on Windows OS. (I need to look at his other tasks to ascertain his validation rate).
Now I know of your (albeit single) validation and that there are other 5700XT owners out there I will fire my machine back up and hope to be paired with another owner of this GPU type to see what happens :-)
After persevering and
)
After persevering and crunching a lot of tasks I finally got a single valid on the 1st August. That task was interestingly validated against another host with a 5700 XT, in fact the same host TicToc had success with previously (although his result may have slipped off the server by now)...
My host is now using driver 19.7.5 and there is no no change from my earlier reports so I shut the machine in question down again a couple of days ago.
I'm now thinking I will wait for another driver iteration to see how that goes before nudging Bernd to take a look hopefully before my valid task disappears!
The question is: are Navi
)
The question is: are Navi results completely wrong or they are just a bit different but still correct?
In my case for ~1 month and 400-600 tasks validation passed only 4 times when 2 of 3 GPU were 5700 ( https://einsteinathome.org/host/12764660/tasks/4/40 ), then Navi gave the same(wrong?) result and "win" by majority rule, because Navi GPU results differ from any(?) other GPU results.
If 5700 results were wrong we could miss potentially important information like new star etc (false negative). So einsteinathome team should recalculate all Navi data for all this 2(3) x Navi-cases.
Updated to the latest AMD
)
Updated to the latest AMD Linux driver (19.30-855429). I am now running the 5700XT on the host rather than an Ubuntu VM, but other than much quicker run-times, the outcome appears to be the same. Tasks run to completion, but then are marked as invalid. It looks like the only changes were to move the Navi specific driver code into the unified driver stack and update the install scripts to support RHEL/CentOS/SUSE.
Tasks running on the 5700XT are the tasks sent on August 18 https://einsteinathome.org/host/12597202/tasks/5/0?sort=desc&order=Sent
Current task status:
Valid: 0
Invalid: 6
Pending: 55
I'll let the rest of my queue complete, but it looks like there is nothing new in the driver, other than it being a unified driver for all recent AMD GPUs.
System also has a Radeon VII and a pair of Vega 64's, and they are crunching away on this driver without issue.
tictoc wrote:System also has
)
That's unfortunate about the invalids. Assuming that will get worked out eventually, how do task completion times compare among the 5700XT, RadeonVI and Vega64?
Ideas are not fixed, nor should they be; we live in model-dependent reality.