GTX970 running at 54% TDP & downclocking under stress

[Gouca]

Hi!

Components:

Spoiler

AMD Ryzen 3600 with PBO enabled
MSI B450I Gaming Plus AC (single PCI-E slot so can't test in another)
ASUS GTX970 Strix
2x8GB Ripjaws V 3000 MHz / C15
Seasonic CORE-GC-650W

Following a relocation my PC started behaving weirdly. To keep this short, my GPU, ASUS GTX970 Strix started downclocking when stressed.

• Desktop / 2D use: GPU core 899.4 MHz / VRAM 1752 MHz / VDDC 1.012 mV
• Light 3D use: GPU core 1113 MHz / VRAM 1752 MHz / VDDC 1.075 mV
• Heavy 3D use: GPU core 405 - 560 MHz / VRAM 810 - 1752 MHz / VDDC 1.000 mV

Power states work as if they had reversed. GPU-Z shows PerfCap: PWR (Power) as the bottle neck and reports a maximum of 54% of TDP used when GPU is stressed. Nvidia inspector shows no abnormalities.

• Temperatures are excellent for both CPU/GPU: 30-60*C
• Disassembled, cleaned and reassembled all the components a couple of times
• CPU & RAM are stable, ran 10000% RAM coverage with Karhu and 4 hours of Prime95 without any issues whatsoever
• I've reinstalled Windows after multiple DDU attempts
• I've played around power states and power delivery
• I've tried enforcing OC with MSI AB, no changes during P0

GPU-Z reports that at all times the 8-pin connector for my GPU is only delivering up to 30W of power. PCI-E is reported to deliver up to 75W of power. 8-pin voltages are at 11.5 - 12.0 V according to both GPU-Z and HWInfo.

GPU's U505-chip (apparently main power controller) seems to have been scratched. The funny thing is these issues started before I disassembled the GPU for the first time and the screws were untouched with the warranty stickers undamaged.

I'm trying to figure out if this is motherboard / GPU / PSU related. Can't wait for RMA so I'll have to start replacing components one by one on Monday.
What do you suggest?

 

[droopyRO]

You tried a lot of things, but did you try the GPU on another system ?

 

[Gouca]

You tried a lot of things, but did you try the GPU on another system ?

Unfortunately not, on a summer vacation so don't have other systems here. Just driving home with a new GFX. If it won't work I'll just replace both PSU & MB.

 

[eidairaman1]

Make sure your room is not 80-85+ °F

 

[Chrispy_]

What was the verdict after changing to a new graphics card?

The scratch on the chip is a red herring. Chips are often marked like that as part of the QC process and the plastic packaging on those types of ICs can be measured in mm whilst the depth of the scratch measured in micrometers.

If it's not working with the new graphics card, I'd say try another PSU first; Seasonic claims 3% regulation at full 650W load for your particular model and 11.5V is outside of that 3% range even though you're likely only at 30-40% load.

I have a 9-year-old BeQuiet 680W unit in this box and can still pull 190W from the 8-pin according to GPU-Z at 11.9V. It's a sample size of one so don't read too much into it but 11.5V when you're only pulling 30W sounds sus to me....

 

[Keullo-e]

Make sure your room is not 80-85+ °F

Had to convert that to Celsius, but that isn't bad as I had about 31C max on my apartment on the hottest summer days, still no problems with temps on either on my rigs.

 

[Gouca]

What was the verdict after changing to a new graphics card?

The scratch on the chip is a red herring. Chips are often marked like that as part of the QC process and the plastic packaging on those types of ICs can be measured in mm whilst the depth of the scratch measured in micrometers.

If it's not working with the new graphics card, I'd say try another PSU first; Seasonic claims 3% regulation at full 650W load for your particular model and 11.5V is outside of that 3% range even though you're likely only at 30-40% load.

I have a 9-year-old BeQuiet 680W unit in this box and can still pull 190W from the 8-pin according to GPU-Z at 11.9V. It's a sample size of one so don't read too much into it but 11.5V when you're only pulling 30W sounds sus to me....

Working flawlessly with the new GFX. Turns out oldie GTX970 gave up. Thanks for the info regarding the marking on the chip.

Still curious as to what's exactly going on with the GTX970 since it's in a flawless condition (even went through everything with a multimeter).

Apparently this precise issue is described by several users with GTX970s from various manufacturers.

 

[Chrispy_]

Working flawlessly with the new GFX. Turns out oldie GTX970 gave up. Thanks for the info regarding the marking on the chip.

Still curious as to what's exactly going on with the GTX970 since it's in a flawless condition (even went through everything with a multimeter).

Apparently this precise issue is described by several users with GTX970s from various manufacturers.

Just age probably. I've been doing corporate hardware builds/maintenance for several hundreds of PCs for nearly 20 years now and I've seen it all. Graphics cards are complex beasts and eventually you'll just get a random SMC die and unless you're the OEM it's going to be tricky to track down and identify which exact SMC component needs to be replaced, and even whether the GPU itself survived the SMC failure or the SMCs death allowed too much voltage/current to flow into the GPU in a harmful way.

Nothing lasts forever, and whilst a 970 is still a capable card it's had a good run and outlasted its warranty period by a factor of 200% or more.