Radeon drivers 20.1.1 should fix the issue seen on S@H. Going to try now here...
No it won't until the beta application is moved to Main. Eric had to update the ati applications to remove a OpenCL kernel compiler flag that was causing issues.
So please do join Seti Beta and get the new 8.24 application and test it on tasks at Beta. The sooner it is proved out there, the sooner the app can be moved to Main.
I already have 21 tasks validated here and no invalid yet
With previous drivers I had no single valid one, so it looks good to me.
Host: https://einsteinathome.org/host/12798703
I already have 21 tasks validated here and no invalid yet
With previous drivers I had no single valid one, so it looks good to me.
Host: https://einsteinathome.org/host/12798703
By the time I looked just now, you had reached 29 validations.
However there was one inconclusive:
Task 912372190
If the quorum resolves not in the Navi's favor, that would be a 3% invalid rate but the sample is much too small for a good estimate. It is clear that it is massively better than before.
Your elapsed times cluster near 350 seconds. Are you currently running 1X? Any overclock or other oddities? Any indications as to power consumption?
If this holds up it is very good news. I'm sad that I just installed two RX 570s in my main machine to replace the Radeon VII I gave up on.
OK, good to hear that the new drivers appear to be working here at Einstein without recompiling the application. I have a question for you, the stderr.txt here does not report the internal driver version number. Can you say what it is?
It would be helpful to know what driver version the cards at Seti are attempting to use. For example this is what we suspect to be a 5700XT running the newest driver.
57 valid by now, 0 invalid yet.
Yes, I'm running 1X. Power reported by the card was ~160 W, measured difference at wall socket between GPU idle/load ~180 W. No OC, nothing changed except increased fan speed in order to keep it < 70 C.
Keith, where can I find that internal driver version? I don't have direct access to that machine during weekend, so can check it on Monday only.
Thanks for the heads up Mumak, this new driver is looking good so far. I currently have 28 validated Gamma-ray tasks, 97 pendings and crucially no invalids or error tasks. All tasks thus far have been run x1.
@Keith
As far as I can make out 3004.8 is the internal driver number, taken from Boinc startup messages.
Thanks @Gavin, I don't have any AMD cards and am only able to inspect AMD users task stderr.txt outputs. Thanks for confirming the internal driver version number reported at BOINC startup by gpudetect.
Still monitoring the cards at Seti Beta to see if the new application fixed the issue there.
Glad to see the new driver has fixed the issue here at Einstein without needing a new app. Looks like the app developers here never implemented that iffy compile parameter to speed the app up.
Radeon drivers 20.1.1 should
)
Radeon drivers 20.1.1 should fix the issue seen on S@H. Going to try now here...
-----
First task validated! Wingman
)
First task validated! Wingman was Radeon R9 380, thanks Gavin ;-)
-----
Mumak wrote:First task
)
I'm glad I was of assistance without knowing it!
I will upgrade my Navi drivers and see if I can reproduce your success with Gamma Ray work :-)
Gav.
Mumak wrote:Radeon drivers
)
No it won't until the beta application is moved to Main. Eric had to update the ati applications to remove a OpenCL kernel compiler flag that was causing issues.
So please do join Seti Beta and get the new 8.24 application and test it on tasks at Beta. The sooner it is proved out there, the sooner the app can be moved to Main.
I already have 21 tasks
)
I already have 21 tasks validated here and no invalid yet
With previous drivers I had no single valid one, so it looks good to me.
Host: https://einsteinathome.org/host/12798703
-----
Mumak wrote:I already have 21
)
By the time I looked just now, you had reached 29 validations.
However there was one inconclusive:
Task 912372190
If the quorum resolves not in the Navi's favor, that would be a 3% invalid rate but the sample is much too small for a good estimate. It is clear that it is massively better than before.
Your elapsed times cluster near 350 seconds. Are you currently running 1X? Any overclock or other oddities? Any indications as to power consumption?
If this holds up it is very good news. I'm sad that I just installed two RX 570s in my main machine to replace the Radeon VII I gave up on.
Thank you for sharing your results with us.
OK, good to hear that the new
)
OK, good to hear that the new drivers appear to be working here at Einstein without recompiling the application. I have a question for you, the stderr.txt here does not report the internal driver version number. Can you say what it is?
It would be helpful to know what driver version the cards at Seti are attempting to use. For example this is what we suspect to be a 5700XT running the newest driver.
57 valid by now, 0 invalid
)
57 valid by now, 0 invalid yet.
Yes, I'm running 1X. Power reported by the card was ~160 W, measured difference at wall socket between GPU idle/load ~180 W. No OC, nothing changed except increased fan speed in order to keep it < 70 C.
Keith, where can I find that internal driver version? I don't have direct access to that machine during weekend, so can check it on Monday only.
-----
Thanks for the heads up
)
Thanks for the heads up Mumak, this new driver is looking good so far. I currently have 28 validated Gamma-ray tasks, 97 pendings and crucially no invalids or error tasks. All tasks thus far have been run x1.
@Keith
As far as I can make out 3004.8 is the internal driver number, taken from Boinc startup messages.
Gav.
Thanks @Gavin, I don't have
)
Thanks @Gavin, I don't have any AMD cards and am only able to inspect AMD users task stderr.txt outputs. Thanks for confirming the internal driver version number reported at BOINC startup by gpudetect.
Still monitoring the cards at Seti Beta to see if the new application fixed the issue there.
Glad to see the new driver has fixed the issue here at Einstein without needing a new app. Looks like the app developers here never implemented that iffy compile parameter to speed the app up.