SAPPHIRE PULSE Radeon™ RX Vega56 8G HBM2 Wrong FAN RPM

[Teral]

Wrong GPU Fan RPM on SAPPHIRE PULSE Radeon™ RX Vega56 8G HBM2 
RPM = 0, but HWiNFO say 1350RPM

 

[Martin]

Please attach the HWiNFO Debug File of situation with RPM=0 and I will check that.

 

[Teral]

Please attach the HWiNFO Debug File of situation with RPM=0 and I will check that.

Done

 

[Martin]

Thanks. Are you sure that the fan is off, or is that just GPU-Z indication?

 

[Teral]

Yes, I checked. Fans do not rotate on the video card, but in monitoring 1351RPM
The fan starts when GPU temperature is more than 55C

 

[Martin]

I see, thanks.
Please try this build and let me know the result: www.hwinfo.com/beta/hwi64_583_3413.zip
Please also attach a new Debug File made using this build.

 

[Teral]

I see, thanks.
Please try this build and let me know the result: www.hwinfo.com/beta/hwi64_583_3413.zip
Please also attach a new Debug File made using this build.

It same.

[font=arial,sans-serif]I think this is a bug of AMD drivers.[/font] [font=arial,sans-serif]As in the monitoring of Radeon Whatman the same wrong RPM.[/font]

 

[Teral]

More info
+debug file

 

[Martin]

Thanks for the feedback.
Yes, all AMD functions report a non-zero RPM in that case, so I'm really puzzled how does GPU-Z report this case.
Can you please try to enable the "GPU I2C Support Force" option in HWiNFO and then create a Debug File? Please note that in this case there's a risk of a system crash with Vega GPUs, so be prepared for this.

 

[Teral]

Done

PS. [font=arial,sans-serif]GPU-Z when it starts also shows the wrong RPM value, but after a couple of seconds it is corrected.[/font]

 

[Martin]

Sorry, I forgot to tell you that after enabling the "GPU I2C Support Force" option, you will need to press the "Reset GPU I2C Cache" button and then start HWiNFO.
Can you please repeat the test run?

 

[Teral]

Sorry, I forgot to tell you that after enabling the "GPU I2C Support Force" option, you will need to press the "Reset GPU I2C Cache" button and then start HWiNFO.
Can you please repeat the test run?

the first attempt to fail during the launch of HWiNFO, the computer hung with a black screen and noisy at maximum RPM fans on the video card - HWiNFO64-1.dbg

the second is a successful launch. But the fan RPM is wrong. - HWiNFO64-2.dbg

 

[Martin]

OK, so that's no way.. Please disable the "GPU I2C Support Force" option back.
Then please run this build: www.hwinfo.com/beta/hwi64_583_3414.zip
while the fan is off and spinning too. Please send me the Debug File created during that run.

 

[Teral]

Done.
Two dbg file.
1 - only fan off : HWiNFO64 -fan off.dbg
2.  start with fan off then RPM up to max, and go to fan off : HWiNFO64 fan off -to spinn at max - to off.zip, and screen shot.

 

[Martin]

Thanks for the results.
I'm sorry, but currently I have no idea how to determine fan stop in this case. It's quite possible that Sapphire uses some custom fan control mechanism about which not even AMD drivers are aware. I will keep trying to determine this, but don't have any other ideas for now...

 

[Martin]

Another theory might be that for these cards a fan speed of ~1330-1370 RPM (~40-41%) means off.
What if I would implement such filtering ?

 

[Teral]

Another theory might be that for these cards a fan speed of ~1330-1370 RPM (~40-41%) means off.
What if I would implement such filtering ?

It seems to me that a similar trick is used by the GPU-Z. When it starts up, it also for a second shows 1350 rpm, but then adjusts the indicators to 0 rpm.

I open ticket in sapphire tech support, maybe they will offer a solution (for example a new version of BIOS)

 

[Martin]

Yes, looks like that. Are you able to determine what fan speed range is reported when the fan is off? Then I could tune this more precisely.

 

[Teral]

I saved the sensor log from the GPU-Z.
I think it checks a few seconds if the rpm does not go beyond the range 1362..1700 then it reports rpm = 0

 

[Martin]

Based on the data you provided so far, I think that a reasonable filter would be that if 1330 <= RPM <= 1370, then it's off.
Are you able to compare HWiNFO fan readouts when the fan is really off (not just off reported by GPU-Z) and confirm this ?