GPU reading bugs

[tom5576]

Hi

Running hwinfo64 on inspiron 5576 laptop. CPU is FX-9830P with integrated R7 graphics and there is also GPU RX560 chip.

Having several problems here, unsure whether all are driver related or some possibly hwinfo bugs.

1. there is no voltage sensor on integrated graphics. running linux with kernel 5.6 exposes this sensor and readings are correct, but nothing on win7/win10

2. as soon as the AUX power exceed 10W on dedicated GPU the sensor disappear from hwinfo (see screenshot)

3. GPU core current is capped at 69.727A, i think it is system BIOS or driver limit, but not sure since raising voltage increases power draw from wall (for this reason i think it is not VRM limit, at least not VR but could be MOSFET limit) (see screenshot)

Any ideas?

 

[tom5576]

up

 

[Martin]

Try to click "Restore Original Order" or "Fixed Order" in sensor settings, that might bring up the missing sensors.

 

[tom5576]

voltage sensor never existed on AMD R5/R6/R7 graphics even after fresh windows and hwinfo install. until recently i thought it is because there was no sensor on integrated graphics, but then i was testing something in linux and noticed it is actualy there and working as expected

 

[Martin]

Voltage sensor on older AMD GPUs depends on GPU drivers and telemetry is rather limited on those series too. HWiNFO will report GPU VDDC only if the driver supports it.

 

[tom5576]

ok, that confirms 1 and 2 are GPU drivers issue. what about #3 have you ever encountered a situation where the GPU driver caps the allowed current into the core? i cannot go higher than 1350MHz with noticeable impact on performance, even OC from 1300->1350MHz yields smaller performance increase (+10-15 pts in furmark) than 1250->1300 (+40 pts in furmark) or 1200->1250 (+40 pts in furmark)
that is in graphics load scenario.
if using compute load, e.g. oclmembench, then i can go as high as 1560MHz on core and it makes a difference (in this case 69.727A is not even reached)

 

[Martin]

Really not sure what exactly is causing #3, most likely VBIOS, driver or SMU telemetry.

 

[tom5576]

VBIOS has been examined, and seven power related values increased to 100 (TDP, TDC, small power limit, battery power limit etc.) unless there is another undiscovered limitation i'd rule out VBIOS. just hope it is not driver as that could be a nightmare to patch as well as system BIOS SMU. since it is laptop, and space is limited there are only two phases for the core with 2x85A mosfets on each phase. i think they might be limiting here and will try to replace them as soon as i get new 200A mosfets

 

[tom5576]

Voltage sensor on older AMD GPUs depends on GPU drivers and telemetry is rather limited on those series too. HWiNFO will report GPU VDDC only if the driver supports it.

checking power readings using AMD uprof shows this sensor, as well as separate sensors for CU0 and CU1. but there is the temperature offset of some kind as the readings are quite lower than in hwinfo for CPU temp. are they using ADL to poll these sensors, because i've tried enabling ADL in hwinfo but did not notice any difference?

 

[Martin]

ADL is not used in this case and per description of those items it looks like it's not a real measured temperature, but an estimation based on activity data.

 

[tom5576]

ok, APU temp "sensors" in uprof should not be used then, however dGPU temp sensor within the same app is the actual sensor, not just some dummy calculation, as it is showing identical temperature as dGPU sensor in hwinfo.

i have noticed something else during extensive tests:

example 1: furmark -> GPU Core Power + GPU Aux Power = GPU Chip Power (+- 1.5W sometimes, sometimes it is exact sum of the two)

example 2: unreal tournament 4 -> GPU Chip Power is significantly lower than the sum of GPU Core Power + GPU Aux Power, it is even lower than GPU Core Power alone. could it be the cause of low FPS in some scenes?