Enabling ADL makes the AMD Fury crash

I have just got a Fury X and testing it. It looks like the entire I2C is not stable and I will have to disable it completely (again) for all Fiji GPUs. It's possible that while on the Fury (non-X) it was semi-stable, on the Fury X it can lead to a system crash much sooner.
But don't worry ;) There seem to be other methods that can provide VR temperatures + Liquid temperature (for Fury X). If these prove stable enough, I will use them instead and the entire I2C will not be used on Fiji to prevent the system crashes.
 
Hi Martin,

I have also recently acquired a Sapphire Fury Tri-X (ie Fiji Pro) and if you require another tester of updated HWiNFO I'd be happy to assist.

I do believe GPU-Z and MSI AB are showing VDDC and not VID, using AiDA64 register dumps I get these VIDs for my card (assuming data is correct, currently it correlates):-

------[ GPU PStates List ]------

DPM0: GPUClock = 300 MHz, VID = 0.90000 V
DPM1: GPUClock = 508 MHz, VID = 0.92500 V
DPM2: GPUClock = 717 MHz, VID = 0.95000 V
DPM3: GPUClock = 874 MHz, VID = 1.08100 V
DPM4: GPUClock = 911 MHz, VID = 1.12500 V
DPM5: GPUClock = 944 MHz, VID = 1.16200 V
DPM6: GPUClock = 974 MHz, VID = 1.20600 V
DPM7: GPUClock = 1000 MHz, VID = 1.24300 V

Reported VDDC in GPU-Z & MSI AB is lower for DPM 7, I think this is due to say how stock LLC (VDROOP) is like on Hawaii cards. Plus just like on Hawaii depending on app loading card I'm seeing varying VDDC even when card in same DPM state. Also if I add GPU core voltage via MSI AB both GPU-Z & MSI AB reported similar increase in VDDC.
 
Thanks for your offer.
Currently I gave up trying to read information from the VRs, as even with the suggested AMD synchronization there's still a risk of a total system crash.
You can try this yourself if you go into Settings, click "Reset GPU I2C Cache" and then depending on HWiNFO version: with v5.20 you can try both modes "GPU I2C via ADL" enabled or disabled, or with the latest Beta v5.21-2810 only when "GPU I2C via ADL" is enabled it will try to access the VRs.
But be careful as those methods have a certain probability of a total system crash...

Anyway, I will try to find out whether it's possible to get VDDC via a different and reliable way...
 
Cheers for info, will try that as need some readings for my own referencing.

Appreciate it if you get VDDC sorted, I've currently had no issues leaving say GPU-Z or MSI AB monitoring card whist say running f@h over a lengthy period. I just prefer using HWiNFO due to how I get other stats as well :) .

As always your support is amazing :) .
 
There is Fury X, Fury Air and Fury Nano.

I have the Nano. My system is runs fine (using 5.2-2800), however under a certain load, I get the GPU/blank screen issue.

Disabled the VRM monitoring sections completely and system is stable under that load.

More details here (sorry for creating new thread).
 
Been using HWiNFO v5.20-2800 since yesterday several times about 20-30min at a times. Not getting any crashes, some error in reads but on the whole AOK.

OS: Win7 x64 UEFI install with Crimson 16.2
Hardware: i5 4690K @ 4.9Ghz, Asus Maximus VII Ranger, Sapphire Fury Tri-X

Was gaining reads for 3DM FS DEMO looped.

tB2nJox.jpg
 
The next HWiNFO Beta will be able to monitor GPU VDDC, GPU Current and Power on Fiji and some others too ;) Without the (problematic) use of I2C.
EDIT: Plus the GPU Memory MVDDC, current and power ;)
 
Works fine, will report back if there are instability issues. Memory current and power often shows zero not sure if it's real (HBM is pretty power efficient, 15W even at max load) or monitoring issue.

attachment.php
 

Attachments

  • amd.png
    amd.png
    12.7 KB · Views: 51
Thanks for the feedback. Yes, the GPU Memory current/power can report 0 when idle. That will change if you put some load on the GPU...
 
Many thanks Martin for your work :) , will be donating via your site for the great app ;) , will also try to encourage others on a Fiji thread I'm doing on OCN to as well.

My results (card had been on f@h prior to this beta screenie so min temps not my normal)

7dgOnbs.jpg


I'm not seeing (VIN/+12V), any reason?

I downed beta, overwrote .exe & .ini, checked ADL is not enabled, then did "Reset GPU I2C Cache" and then also "Reset Original Order" in Layout tab.
 
Thanks for the feedback guys.
GPU VR VIN/+12V is only available via I2C, there's no other way to access it. Due to the many issues when accessing I2C of Fiji (system crashes, reliability) I'm not supporting this. The new values now available are read using a different method, which should be reliable and not cause any troubles.
 
After a few hours GPU Clock reading shows an incorrect value (Maximum: 1311 MHz) the GPU runs at @1040 MHz:

attachment.php
 

Attachments

  • amd.png
    amd.png
    120.6 KB · Views: 31
Probably a rare glitch in the readout. Are you using some other similar tools monitoring the GPU along with HWiNFO ?
 
Now that you mention it I did use MSI AB to show my sensors in-game. I'll try it without additional monitoring apps soon.

Edit: Yes, without MSI AB, the sensor value is OK, shows @1040 as it should.
 
Just installed the new AMD 16.12.1 Driver and the two VRM temperatures (GPU VR and GPU VRM) are missing - removed the registry key to reset the sensors and I2C cache but still not appearing so I think something must have changed in the driver. Using the latest Beta btw.
 
Yup, can confirm this on my Fury X. Seems the ReLive drivers don't support those values anymore. Contacted AMD...
 
Back
Top