Custom Polaris GPUs (RX 480) incorrect parameters

[cez4r]

Thanks. Will wait for the Debug File when you can create it.

My Debug File:

http://www.filedropper.com/hwinfo64

Also, I think the best guy for help would be the GPU-Z creator, W1zzard
@ OCN he helped already a guy in creating amazing tool for RX 4xx family of GPUs:
http://www.overclock.net/t/1609782/watttool-a-simple-tool-that-combines-wattman-with-vrm-monitoring-tweaking-for-rx-400-series

Thanks for your help!

 

[Martin]

Thanks. I don't have solution for this now, but keep working on it...
And btw, what are the power values in both tools in idle ?

 

[cez4r]

And btw, what are the power values in both tools in idle ?

HWINFO:  2.66W-3.55W
GPU-Z:      9.3W-10.3W

 

[Martin]

I've sent questions about this issue to various folks including AMD...

 

[cez4r]

Thanks

 

[VictorVG]

New AMD Radeon Software Crimson™ 16.9.1 HotFix is correct this error?

 

[cez4r]

I won't be testing Crimson 16.9.1 because from different forums I know that OpenCL performance in this version is crippled
I'll wait for Mr Martin's findings.

 

[Martin]

I'm still waiting for answers too...

 

[VictorVG]

Not surprised, because driver ATI has always been disgusting. Ever since the times of the first ATI Radeon drivers so happened that to them even the saints fans. If you want a stable driver working without errors - buy professional card ATI FireGL (professional drivers for FireGL not run on Radeon to - PCI ID different ). A PCI ID is set on the substrate of the chip resistors and code in the Video BIOS. And as far as I can remember all latest card ATI FireGL series after ATI FireGL4 based on ATI Radeon chip (ATI FireGL4 based on IBM chips).

 

[cez4r]

And btw, what are the power values in both tools in idle ?

HWINFO:  2.66W-3.55W
GPU-Z:       9.3W-10.3W

New 16.9.1 drivers - same story ...

 

[Martin]

I don't think the power difference is a driver issue.
I'm still waiting for answers from my sources.

 

[cez4r]

Thanks, no problem

 

[Martin]

OT: Are 16.9.1 drivers really crippling OpenCL performance? I'm using GPUs for computing (BOINC) as well, so I'm personally interested in this.

 

[cez4r]

Initially I saw few posts around the internet about it but I have tried it myself and I cannot confirm it.
Probably the problem is that drivers are not completely removed for complaining people.
I've found that the best procedure now for perfect removing it is doing it with the AMD Cleanup Utility first under Safe Mode, reboot, and then do this again for the second time there, but now with the DDU.
Note: If you're using AMD chipset AMDCU will remove the AMD chipset/AHCI drivers too - so you will need to install them again.

Hope this helps

 

[Martin]

Thanks for the info

 

[cez4r]

@Martin

Hi again!

I have a humble request - would you mind adding 2 things to the HWINFO:

A. VDDCI aka Memory Controler voltage - "Memory Voltage" setting in WattMan (Radeon Settings).
I've found that no software can do this! And this is important voltage because the GPU VDDC voltage level cannot be lower than this.
So, there are a lot of mistakes people do in this regard and it will be very helpful to have such sensor.

B. When it comes to our talk about the GPU Power Only reading for RX 4xx and the discrepancy between HWINFO and GPUZ, there is something far far more important - and a lot of people are looking for it
It would be really really nice if HWINFO would have possibility to display Power Consumption for:
1) PCIE slot, and 2) PCIE connector(s) on a card...
Ad 1) I've heard that "GPU VRM Power in (PIN)" in HWINFO = PCIE slot power usage, just I don't know if it's true?
Ad 2) I've heard that GPUZ in its "GPU only Power Draw" = PCIE connector power usage - but again, I don't know if it's true too...

This PCIE slot power usage display for the Polaris series  is really very important - especially after the infamous initial RX 480 affair

Thanks for Your hard work!
Have a nice weekend
Cheers

P.S. That 'GPU only Power draw' is of no use because the GPU draws current both from the slot and from the connector(s)...

 

[Martin]

A. It's not possible to monitor VDDCI on RX 480 and 470. These boards don't seem to implement the required circuits for monitoring. But I've seen some RX 460 cards that seem to have this (and it's reported by HWiNFO).
B. Similar as A. I have never seen a GPU being able to monitor PCIe power consumption. It would require dedicated circuits and special design.
1) No, Power-In is the power on input side of the VR for the VDDC rail.
2) I don't think so. My current guess is that this power is VDDC power + something additional. I still haven't figured out exactly, and still scratching my head about this

 

[Martin]

I think I just made a breakthrough and will be able to report the total GPU-chip power (as GPU-Z).
Stay tuned

 

[cez4r]

Ok, thanks for answers!

 

[Martin]

Have you checked the new Beta version v5.37-2977 ?
That will show you the GPU chip power similar to what GPU-Z displays. But I believe this value is just an estimation - it's not possible to precisely measure the entire GPU power, as it lacks the required hardware monitoring components on all voltage rails. Actually I don't think there's a GPU that would have a capability to measure all rails and provide exact values.