Question GPU excessive factory overclock, going into "Idle" on GPU-Z while gaming, usage not 100%

[miedox]

I apologize in advance for being long, but I'm hoping to get some help understanding what my gpu is doing, or what is wrong with it. I've recently upgraded to a 2k 165 Hz monitor, so I began paying close attention to sensor readings on programs like Afterburner and GPU-Z to get the most performance out of my card, but started to notice something wasn't right.

First of all, I have a refurb Zotac RTX 2070 blower, received as a replacement for RMA (second time fml).
Initially upon receiving the card, I noticed that the card overclocked way too much on factory settings. The specs on it says 1620 MHz as boost clock, but the card manages to maintain above 1800 MHz even when temps are hitting 70 C. Again, no OC applied. Well, I thought I just got lucky and got a card that could OC like a boss. BUT problems started to occur when I started to run more demanding graphics with the 2k monitor.

I'll use GTA V as a main example, since it is the most demanding game I play (with high/very high graphics settings), and corresponding "PerfCap Reason" on GPU-Z at various situations.

1. At first minute or so, GPU seems to be fine, (apart from seemingly unreasonable high clocks >1900 MHz) with fps well above 100. PerfCap Reason shows "VRel" + "VOp" (+Pwr occasionally)
2. Then, the FPS drops to around 85, with GPU usage down to 50~70%, and does not manage to come up. At this point PerfCap Reason changes to "Idle" with occasionally oscillating to "Pwr" in more demanding scenes like off-road/mountains. FPS drops even more in those areas. GPU Power is rather higher, anywhere from 60~90%. Clock speed is still above 1800 MHz at this point, and temp is reasonable at 50~70C, but never goes above 75.

Same thing happens in Benchmark

So then, I tried to see what happens with Heaven benchmark. Used: DirectX 11, Quality - High, Tessellation - Normal, Anti-aliasing - x8. Monitoring sensors, I noticed power was maxing out whereas GPU usage was ~70%. So I figured if I boosted the power limit on Afterburner, it would utilize more GPU (power bottlenecking?). So I dragged the power slider to the max at 114, and Heaven managed to stay stable, with GPU usage hitting high 90's. And more importantly, PerfCap Reason did not go to "Idle". It was at "Vrel" + "Pwr" I think.. Temperature was hitting close to 80 C, and I set the fan to max out at 80 so it's basically a hair dryer.

Problem solved right? Wrong...

So I go back into GTA V benchmark. FPS stable around 100~120, and PerfCap Reason NOT going to "Idle" BUT crashed to desktop towards the end. I tried lowering clock speed while keeping higher power limit, but still crashed at the end. Clock speed still manages to go above 1700 MHz and FPS fluctuates greatly. Tried going in game, but also crashed after a few minutes.

Any help would be greatly appreciated since I am unable to use the gpu to it's full potential. Maybe something is wrong with the BIOS that's making it clock too far? I don't wanna have to flash the card, but I really don't wanna go through another RMA... The only thing that keeps the card from going in to this "Idle" state seems to be the power boost, but it's so unstable that it crashes. I already tried using "Prefer Maximum Performance" on Nvidia Control Panel, and turning off Power Saving for PCI Express in Advanced Power Settings. I also tried uninstalling the driver with DDU and installing the latest one.

Thanks in advance!


System:
GPU: Zotac RTX 2070 Blower
CPU: Intel I5-9600k
Motherboard: Gigabyte B365M DS3H
Power Supply: Thermaltake Toughpower GF1 750W 80+ Gold
(XPG 16GB RAM, Samsung SSD 860 EVO)

(I also posted on TechPowerUp, but got impatient on awaiting approval)

 

[Dragos Manea]

Check your CPU temperature. Also that videocard does not show you the VRM temperature which can overheat and drop clocks.

 

[miedox]

Check your CPU temperature. Also that videocard does not show you the VRM temperature which can overheat and drop clocks.

CPU is fine, averaging 50 C.
I don't think dropping clock is the issue here. If temperature was the problem, then clock speed would indeed drop for thermal throttling. But as I mentioned, clock speed stays well above boost clock (+200 MHz) whereas GPU utilization is quite low (50~70%).