Wish i was covered by consumer rights... Had to order it from Aliexpress though.
Well, expecting some troubles then :c
XFX 7900XT merc 310 high junction temperature
- Thread starter DirtyAlpaca
- Start date
Well, expecting some troubles then :c
DirtyAlpaca
Member
You'll be fine. It's their best consumer block!
Just make sure to use a small GPU support beneath it or mount it vertically since it's ridiculously heavy.
@DirtyAlpaca How do the screws around the gpu-chip look like, when they are torqued right?
I tried it just now with the screws less torqued, and my Hotspot rises under load immediately to 110°C and the GPU throttles to around 900mhz in timespy extreme.
Edit: after a bit of fiddling with it, i found the right amount of torque. The Hotspot now hovers at around 85°C under 100% load with stock Powerlimit (400W).
Tested with Firestrike Extreme and Timespy.
I tried it just now with the screws less torqued, and my Hotspot rises under load immediately to 110°C and the GPU throttles to around 900mhz in timespy extreme.
Edit: after a bit of fiddling with it, i found the right amount of torque. The Hotspot now hovers at around 85°C under 100% load with stock Powerlimit (400W).
Tested with Firestrike Extreme and Timespy.
Last edited:
That's exactly what I'm also worried about. Because I switched from an eisbaer AIO to a custom loop with rigid tubing it would be rather catastrophic to find out about there also being a mounting issue with the Core GPU cooler on the 7900XT (Merc from XFX in my case).
It's not like I could easily remove and reinstall the GPU several times to try multiple ways to mount the cooler with different thermal pastes. Such troubleshooting would already be a pain in the butt with soft tubing but with rigid tubing...? - No way.
Those are exactly my concerns too, yep. Disassembling loop multiple times to troubleshoot GPU waterblock proper installation would be a huuuge pain. Or I can make it easier if i order 2 quick disconnects, but still. Am I supposed to fix possible engineering mistake by investing money into unnecessary for my loop 2 quick disconnects, and also torque screwdriver, which is very expensive here where i live.
I just hope I'll be able to mount block properly by barely tightening those screws by hand, just being veeeery gentle with it. Otherwise I'd be wasting my GPU warranty and a lot of money for nothing it seems.
Last edited:
Is that ok hotspot temp for AMD GPU? I thought water cooled we should expect lesser temps. I have 85C hotspot on my stock air cooled 7900 XTX Red Devil with default settings in Timespy benchmark too. But thats aircooled so I expected way less temps with waterblock
I support the idea to test the video cards on game like Cyberpunk 2077 - the only title i have that can really push 420W out of my ASRock 7900XTX card.
Second step would be for AlphaCool to create a 4k detailed (slow and nice) step by step video how they put the card together!
That will resolve lots of questions!
@ Horushka, try Cyberpunk on 4k like i do, in my case its 410W~ and i get 110C stable on Air and 90C-95C stable on watercooling...
Second step would be for AlphaCool to create a 4k detailed (slow and nice) step by step video how they put the card together!
That will resolve lots of questions!
@ Horushka, try Cyberpunk on 4k like i do, in my case its 410W~ and i get 110C stable on Air and 90C-95C stable on watercooling...
Last edited:
That won't help much. No matter what stress test you use, it's obvious that temperature and hotspot delta are way too high in comparison to stock air cooling. Pushing temperatures higher won't help fixing the issue.
We've seen multiple photos of PCBs bending away from the block where the GCD is. It's easy to understand that the contact between the GCD and the block is poor under these circumstances, resulting in bad heat transfer and high temperature. This problem of the bedning PCB needs to be investigated and hopefully solved.
My theory is that the standoffs are slightly too short. Thus, thightening the screws will bend the PCB. And if you tighten the scews with only 1,5 NM, the PCB will be pushed less towards the standoffs and thus it won't bend as much. Then the contact between GCD and block isn't as bad and temperatures are lower.
Assuming that my theory is correct, it would help to make the standoffs slightly longer. For example with very slim washers between the standoffs and the PCB. Something like these with only 0,1 mm thickness (though I'd prefer them made from Nylon instead of steal):
What might also help is to put some thermal pad on the back of the PCB where the center of the GCD is in order to push the chip towards the block with the "help" of the backplate. Alphacool's thermal pad doesn't cover that spot. But then you would basically put pressure directly on the SMDs and that can easily damage the card.
We've seen multiple photos of PCBs bending away from the block where the GCD is. It's easy to understand that the contact between the GCD and the block is poor under these circumstances, resulting in bad heat transfer and high temperature. This problem of the bedning PCB needs to be investigated and hopefully solved.
My theory is that the standoffs are slightly too short. Thus, thightening the screws will bend the PCB. And if you tighten the scews with only 1,5 NM, the PCB will be pushed less towards the standoffs and thus it won't bend as much. Then the contact between GCD and block isn't as bad and temperatures are lower.
Assuming that my theory is correct, it would help to make the standoffs slightly longer. For example with very slim washers between the standoffs and the PCB. Something like these with only 0,1 mm thickness (though I'd prefer them made from Nylon instead of steal):
What might also help is to put some thermal pad on the back of the PCB where the center of the GCD is in order to push the chip towards the block with the "help" of the backplate. Alphacool's thermal pad doesn't cover that spot. But then you would basically put pressure directly on the SMDs and that can easily damage the card.
DirtyAlpaca
Member
How are you guys holding up?
Final try will be a thermal grizzly pad within the first weeks of September. Had a busy summer and I hope all of you had a great time.
Below is a shot from a not-so-technical scene in Cyberpunk. The card flies with seconds towards its limitations.
Seen the card throttle down to 2400MHz.
User errors
Final try will be a thermal grizzly pad within the first weeks of September. Had a busy summer and I hope all of you had a great time.
Below is a shot from a not-so-technical scene in Cyberpunk. The card flies with seconds towards its limitations.
Seen the card throttle down to 2400MHz.
User errors
Then probably our only choice is liquid metal? And maybe some testing with non conductive wasters on standoffs?How are you guys holding up?
Final try will be a thermal grizzly pad within the first weeks of September. Had a busy summer and I hope all of you had a great time.
Below is a shot from a not-so-technical scene in Cyberpunk. The card flies with seconds towards its limitations.
Seen the card throttle down to 2400MHz.
User errors
View attachment 5464
aglasel6403
New member
Hi,
Also having the exact same issue. Has anyone found a solution? Im about ready to RMA this thing and just buy a different block.
Edit: Ah didnt see there was a new message. RIP. Guess im ordering Byski and returning this junk.
Also having the exact same issue. Has anyone found a solution? Im about ready to RMA this thing and just buy a different block.
Edit: Ah didnt see there was a new message. RIP. Guess im ordering Byski and returning this junk.
direstorm
New member
I'm in the low 90s with Ray Tracing on ultra in a heavy ray tracing part of the city. This is with 1mm shaved off all my standoff screws with regular paste. Eventually, I will try liquid metal again but won't happen until I need to do a thorough cleaning in about a year. Too tired of taking the waterblock apart after the 5th time.How are you guys holding up?
Final try will be a thermal grizzly pad within the first weeks of September. Had a busy summer and I hope all of you had a great time.
Below is a shot from a not-so-technical scene in Cyberpunk. The card flies with seconds towards its limitations.
Seen the card throttle down to 2400MHz.
User errors
View attachment 5464
To me it seems like if you overtighten screws - PCB bends and GCD loses contact with waterblock. If you use 1nm torque - pressure isnt enough to squeeze thermal pads on memory chips and GCD again isnt making contact with waterblock. Maybe we should try thinner thermal pads waterblock side?
Shame we have to experiment like that.
Shame we have to experiment like that.
I have the same issue - XFX 7900xtx Merc, received the block yesterday and installed - followed instructions by the letter. I have been water cooling for years - multiple water blocked cards/CPUs. Same issue as described by multiple posters - bowed backplate - horrible temps. Attached some pics incase I’ve done some thing wrong.
I have 13 days to return or refund via AliExpress - looking at the issues here I'm just wondering whether its worth troubleshooting or just pack it up and send it back.
One thing I'd like to clarify - when tightening the screws for the block to the PCB in order 1, 2, 3, 4 - did you guys crank the first screw all the way down before moving to the next one or tighten them bit by bit in a rotation. I did it progressively to make sure pressure was even (as I've always done for other blocks) . . . just thinking of anything I could change the next time I try to install . . . sigh - pretty disappointing for a $350+ GPU block.
I have 13 days to return or refund via AliExpress - looking at the issues here I'm just wondering whether its worth troubleshooting or just pack it up and send it back.
One thing I'd like to clarify - when tightening the screws for the block to the PCB in order 1, 2, 3, 4 - did you guys crank the first screw all the way down before moving to the next one or tighten them bit by bit in a rotation. I did it progressively to make sure pressure was even (as I've always done for other blocks) . . . just thinking of anything I could change the next time I try to install . . . sigh - pretty disappointing for a $350+ GPU block.
Attachments
Last edited:
Extracted the card from the loop (ugh) - took pictures as I disassembled. Definite bow in back plate - can provide images of where the thermal pad imprints and where they were sitting. Looking at the TIM imprint (Noctua NT-H1) . . . it was barely making contact hence the high temperatures. Only thing I can think to do is build up a thicker layer of TIM? I think Dcox12's theory of the standoffs maybe being slightly too short seems to likely . . . I'll repast and try again but if I get no luck - it's going back . . . . Any thoughts @Vanzin ?
Attachments
Is that hotspot temp particularly bad considering the +15% power draw? theres a limit on what the heatplate can do to certain spikes, and the average temperature (if thats what the GPU temperature is) seems low enough that the waterblock is doing its job just fine and thatr the Hot spot is just going to spike very high when overclocked. My XT sees 77c in games with only 5% power limit increase, 320watts. with 50c on the GPU temp. Is there anyone who can give more reference to their XTX with max power limit? as i know it would easily hit 110c on air. Just the RT cores alone seem to run incredible hot, i drop 25C by disabling RT in games. Id argue youre worrying about nothing here from a temperature standpoint. The bow in the backplate may be a aesthetic over sight but doesnt seem cause any performance issue. You mentioned the rapid swings in temperature but that's normally a positive, it means the cooler is doing its job quickly, dissipating heat after a quick rise or spike.
The issue is it's performing much worse than the stock cooler though - the maximum the hot spot ever got was around 85/90 with fans maxed to 60% - That's in Cyberpunk with RT enabled. With this water block running just normal Time Spy it immediately jumps to 110c. I think that's the thermal throttle point? There's no way a water block should be performing magnitudes worse than an air cooler. Why would I spend more money running a solution that is worse than the stock cooler? This water block was supposed to provide MORE headroom for performance and do it at a lower noise level.
7900XT boost clock as standard is 2450mhz, the base clock is 2050, youre running 2751mhz at 350w and are wondering why its running a bit warm, thats a 30% overclock. the hot spot is the single hottest sensor and its still 13c below max. i have no idea what your issue is here.How are you guys holding up?
Final try will be a thermal grizzly pad within the first weeks of September. Had a busy summer and I hope all of you had a great time.
Below is a shot from a not-so-technical scene in Cyberpunk. The card flies with seconds towards its limitations.
Seen the card throttle down to 2400MHz.
User errors
View attachment 5464
This topic is about the XFX 7900XT Merc 310 and this card neither has a reference design, nor has it the "standard" AMD clocks. This card is factory OC'ed by default (though not 2751 MHz boost).
A board power of 350 watts shouldn't be an issue because that's within spec for a Merc 310 7900XTX and both cooling solutions (XFX stock air cooler and Alphacool's water block) are identical for the Merc 310 series 7900XT and 7900XTX.