RX 480 hugely underperforming

vinsonc11

Honorable
Aug 27, 2012
114
0
10,690
I have recently purchased a VisionTek reference design RX 480 8gb. After installing an EKWB waterblock, I have used it in my first custom loop with an i5 3570K OCd to 4.6Ghz, 16GB Corsair LP DDR3 and of course an SSD + some hard drives.
46be28b07f.jpg


Before I put the loop together, I tested my system running AMD's 17.3.3 driver, with the graphics card (remember this is reference design) and I got a pretty decent result:
b4b99d678c.png


After putting the loop together, I tested the card again with overclocks and no overclocks. Strangely, the performance of the card seems to have dropped even though the temperatures are much better.
c77291911a.png


I have also overclocked the RX 480 with nice temps to no avail; it was still underperforming drastically.
b6f8e32459.png


I know for sure that my 3570K is not bottlenecking; at 4.6Ghz, it is on par/faster than the i5 6600.
c4f626c74b.png
(3.4Ghz isn't the actual clock, not sure why it doesn't show 4.6Ghz)

ASIC quality of the RX 480 is 73% and I have also noticed while running UNIGINE heaven that the power consumption tops out at 165 watts; even though it is OCed to 1405MHz, it can't reach it most of the time due to the 165 watt limit. I have even OCed it to 1470MHz and it still hasn't been able to get close to the test before I waterblocked the card.


Any idea what to do from here?





 
Solution


Turns out I just needed to undervolt; overvolting...
You said it can't reach the clocks you've set it at, the question is why. Pure guess, the water block isn't the best or right one. Something either isn't making contact like it should, or there is an issue with the setup. It's on the same loop as the CPU so its probably not an issue with the setup. Double check the sink to make sure it's making contact on the GPU and the VRMs, memory, etc like it's supposed to.

What about stock settings? How much worse are they compared to stock with the air sink?
 

I have posted a screenshot comparing stock settings on water vs air, and as shown, the air ones are much better than the water ones.

I am using an EKWB Full cover waterblock, it covers memory, VRMs and the core. I highly doubt that it is to do with the waterblock itself. I have checked that the waterblock is making good contact and the HWmonitor says so too, with an idle temp of 29 degrees and load of 49 degrees on the core + VRM/mem temps slightly higher.
 
So the first two pics are air, and water both stock? If the block does cover everything I'd still stop and make sure it's on correctly. You put TIM where it belongs, screws are tight?

Looking at the last two pics you got a 6.2% increase, with a ~7% bump in clocks which seems right to me. But you said while testing it didn't hit/stay at 1405MHz?

I'm going to remove the best answer seeing as the problem is still ongoing. Please select one after the problem is fixed.
 


First pic is air, second pic is water with the exact same settings. 3d pic is the OC on water. I must have accidently clicked best answer or something, my bad. Block is installed correctly, I am very sure about that. I am just curious to as why it performed worse after I installed the block. Could it be the drivers? I rolled back 17.4.2 to 17.3.3 (as 17.3.3 was the test where results shone), can newer drivers worsen performance?

Similarly, although it is a reference card, does VisionTek use their own third party BIOS on it like sapphire does? If so, could this be the underlying problem?


 
It can't/shouldn't be the drivers because air and first water are the same.

Again, you said they didn't hit what they should have hit, what were the clocks while testing? I'm curious because OCing it farther did up the scores.

Edit: More thoughts. First, I noticed you said a reference 480, and I see there is only a single 6pin plug. The 480 should have come with an 8pin because it's right at the 150W limit that can be supplied by the power it's getting. Stock it should have been fine, but I can see it OC'd failing because it's being starved for power. You said the clocks weren't hitting 1405, what CPU and GPU load are you getting?
 
I'm not sure what's happening, any reply just gets "picked as solution"... strange. OCed at 4700Mhz core clock; it only got up to around 4050Mhz before being limited by the 165 watt barrier that seems to stop it from going higher.
 
This morning I tired OCing to 1400Mhz but it doesn't even reach it due to the 166W limit; it hovers around 1300Mhz at 100% load during unigine heaven but doesn't ever reach 1400Mhz. Is there a way I can increase the power limit? I have updated to the most recent drivers of 17.4.2 and am using Wattman to OC and afterburner to monitor.

I've seen other people such as Jayztwocents reaching 1375Mhz without touching power limits on water; https://www.youtube.com/watch?v=wstG14DmZVk
I've also noticed that even at 1375Mhz, his power draw only goes up to 120W, while I'm struggling to reach 1320Mhz while already pulling 166W
 
I know you aren't going to agree with me on this, but you've had issues with this ever since you put it under water. I still believe that there is an issue there. Can you put the air cooler back on and see if the performance goes back up? Under water it shouldn't have dropped right away. I'm still trying to figure out why OCing the card resulted in a better score. If it was a thermal problem it should have stayed down. But you increased the core clock 7% and saw a 6+% jump which is to be expected.

We seem to be on different sides of "the pond" so I'm forwarding this to the other mods. Hopefully one of them matches your schedule better because I'm heading to bed myself.
 
Thank you very much for the help though, I really appreciate it. I have just rolled back drivers from 1.7.4.2 and gone back to 17.3.3 and I've managed to bypass my original benchmark at 73%. I'm going to go through and check what I did; I'll keep you updated :)
 


Turns out I just needed to undervolt; overvolting wasn't helping. I didn't even reach the 1400Mhz mark but it seems to be not all about the core clock.
8d86e75a12.png

 
Solution