[SOLVED] GPU bottleneck or faulty GPU?

Elprokan

Prominent
Aug 3, 2020
11
0
510
Hey so I have this issue with my new GPU rx 590. My fps and clock speed ( MHz ) is not constant and is jumping up and down in several games and I need to figure out if it's because of possible bottleneck or if the GPU is somehow faulty. I'll give you an example, in adrenaline software I changed all states for my GPU except Idle to push highest clock rate 1550 MHz. It works in few games like World of Wacraft, Subnautica or Arc and I can maintain highest clock speed for the entire duration of my gameplay, however other games like Diablo, Valorant, Grim Daw, GTA V struggle to keep constant clock speed and they won't even reach 1550MHz and usually bounce from 600-1200 and because of that I have fps drops. Valorant as an example I can have one second 170 FPS and next it drops to 80. I increased the power limit to +50% and set higher curve for fan speed however the temperature won't exceed 70 degress ( 158 fahrenheit ) so it doesn't seem to be overheating issue. I'm adding two screens from afterburner, one is Diablo where my clock speed bounces and other one is subnautica where it is constant, maybe that can help. GPU usage is 100% and then it usually drops to like 0% for 2-4 seconds and then to like 12% then around 35% and back to 99% - 100% on repeat.

Spec:
GPU - Asrock RX 590 OC 8GB
RAM - Kingston DR4 2x4GB 2666 MHz
CPU - Intel i5 - 2500 3.3GHz
MB - MSI H61M-P23
PSU - Corsair VS650


 
Solution
A CPU bottleneck would prevent maximum performance from the GPU, it would not cause slowdowns and interruptions. (Unless the CPU was thermally throttling or encountering actual issues)

To confirm that look at your CPU usage numbers. If it hits 100% at the time you experience such slow downs then you may need to look more closely at what the CPU is doing.

Bottlenecking websites are utterly pointless. They offer very vague information about systems and the data they use constantly changes. Basically you don't know under what conditions they test, the test itself, or even how they generate the percentage.

All depends on settings, hardware, application.

Eximo

Titan
Ambassador
Try one of the other preset overclocks, starting with silent mode. Better to have consistent performance then raw capacity any day.

If that doesn't work, might consider flashing the GPU with the lastest firmware. I vaguely recall there was something not quite right around the time of the RX580/RX590 launches. I believe it was too much power draw through the PCIe slot causing over power conditions.

Actually, you might try lowering the power limit below 100% and seeing if that makes a difference.
 

Elprokan

Prominent
Aug 3, 2020
11
0
510
Try one of the other preset overclocks, starting with silent mode. Better to have consistent performance then raw capacity any day.

If that doesn't work, might consider flashing the GPU with the lastest firmware. I vaguely recall there was something not quite right around the time of the RX580/RX590 launches. I believe it was too much power draw through the PCIe slot causing over power conditions.

Actually, you might try lowering the power limit below 100% and seeing if that makes a difference.
Thanks for your reply. I tried silent mode or no OC at all just using stock ( automatic ) settings in adrenaline software but nothing has changed. I found this website where you can pair cpu with gpu to check potential bottleneck and my combo ( cpu + gpu ) has supposedly 30% bottleneck which is I think a lot and it explains sudden fps drops for like half a second but I'm still not sure if that's the main cause of my issue.
 

Eximo

Titan
Ambassador
A CPU bottleneck would prevent maximum performance from the GPU, it would not cause slowdowns and interruptions. (Unless the CPU was thermally throttling or encountering actual issues)

To confirm that look at your CPU usage numbers. If it hits 100% at the time you experience such slow downs then you may need to look more closely at what the CPU is doing.

Bottlenecking websites are utterly pointless. They offer very vague information about systems and the data they use constantly changes. Basically you don't know under what conditions they test, the test itself, or even how they generate the percentage.

All depends on settings, hardware, application.
 
Solution