AIO on RTX 2080 Super. Help with mounting, temperatures, thermal throttling.

[BillytheImpaler]

Howdy there boys and girls. Long time no chat. I happened across a sore willy deal on a Corsair H150i AIO. I already had an AIO on my CPU so I decided to YOLO it and try to cool my GPU. I'm having some issues mounting it.

Spec:

• Proc: AMD Ryzen 3700X
• Mem: Crucial 2x16GB DDR4-3200
• Mobo: Asrock X570 Phantom Gaming 4
• GPU: Zotac RTX 2080 Super AMP
• Case: Lian Li O11 Dynamic XL
• CPU cooler: Corsair H100i Pro (2x 120mm)
• GPU cooler: Corsair H150i RGB Pro XT (3x 120mm)

I took it all apart easily enough and discovered that the mounting holes in the GPU are spaced 70mm apart, wider than reference nVidia and AMD PCB's.

After some futzing around I decided there was no good way to use the NZXT Kraken G12 I bought for the purpose (Aside from that the H150i doesn't use the round Asetek coldplate assembly so it wouldn't have worked anyway). I used a round file and cut down the Intel bracket and backplate included with the cooler. It fit pretty well, and I got some spectacular temps 37C after an hour of heat soaking with FurMark down from 85C.




I've got the backplate outside the factory decorative backplate because there are some IC's in the area of the GPU that would interfere with mounting it directly atop the back of the PCB. However, my thermocouple indicated that the VRAM was cooking at 95C. Not good.

I broke it down again and glued on heatsinks using Arctic Silver thermal adhesive. I replaced the TIM with fresh Arctic MX-4 and set up some fans to waft air toward the bottom of the card. VRAM and VRM temps are great now, no warmer than about 45C.



Pay no attention to being in a bottom PCI-e slot. Took this pic when experimenting. It's in the top slot now.

GPU temps and performance, on the other hand are not good. My boost clocks are 200MHz lower and GPU Shark indicates I'm at thermal limit at just 65C. The thermal limit is a mystery to me. What's up with that? It ran at ~2000 MHz and 85C on the factory aircooler.




There're only a few things I did differently between the two setups. The first time I had higher mounting pressure. I saw when I went it take it apart the PCB was flexing quite a bit, about 0.5mm across the width of the mounting bracket. I tightened up the cooler until the PCB just started to deflect, then backed it off a bit. I also rotated the coldplate 90 degrees so the hoses don't block airflow to the VRAM modules.

Running Folding at Home simultaneously on the CPU and GPU I've leveled off at 36C water for the CPU with package temps at 72C for the CPU. 29C water with GPU package temps at 51C. GPU-Z only shows 85% utilization on the GPU which may be why it's below that 65C thermal limit I ran into before.

Anything obvious I should be doing to try to mount the cooler; brackets, clamps, something I can just buy or make to help? Any insight on the thermal throttling at only 65C?

Thanks in advance!

 

[LuckyBenski]

I'm guessing some other component is teaching thermal limits. Have you got enough airflow over the VRM area? That midplate doesn't look like it would cool all that well.

 

[BillytheImpaler]

Have you got enough airflow over the VRM area? That midplate doesn't look like it would cool all that well.

A thermocouple directly on the surface of a VRM and on the VRAM heatsink shows temps from 40-45C, that seems plenty low enough. The GPU monitoring software doesn't indicate any power limit state after it gets up to temp. How cool is cool enough for the VRM's and VRAM?

 

[LuckyBenski]

A thermocouple directly on the surface of a VRM and on the VRAM heatsink shows temps from 40-45C, that seems plenty low enough. The GPU monitoring software doesn't indicate any power limit state after it gets up to temp. How cool is cool enough for the VRM's and VRAM?

VRMs are usually rated higher than main chips e.g. 120° or so. So I imagine below 60 is considered very low.

Not sure about VRAM but wouldn't expect it to be any different to main chips - so anything under 80 would be ok.

 

[Finners]

Could it be that it needs an RPM signal from the fans? with the zero speed fan functions we have now it will be expecting the fans to kick in at say 50 degrees but its not happening so its throttling itself for safety

 

[BillytheImpaler]

Not sure about VRAM but wouldn't expect it to be any different to main chips - so anything under 80 would be ok.

Can't find anything useful on Samsung's website, but thrid parties seem to agree that 100C is about the max. I'm well below that.

Could it be that it needs an RPM signal from the fans? with the zero speed fan functions we have now it will be expecting the fans to kick in at say 50 degrees but its not happening so its throttling itself for safety

The initial runs on the test bench were without any fans, that's the source of the phone pic of a screen. After installing the RAM cooling and putting it all back together I installed two case fans blowing directly on the card. Those are attached to the card and register and speed control correctly. You can see that in the screenshot where it indicates 1800MHz. It worked better without the fans installed, at least as far as boost clocks go.

 

[BillytheImpaler]

I should note that I hadn't done any tuning or messing about with the card prior to this. I took it out of the box last year and have been flogging it like a rented mule ever since. The manufacturer lists it at 1650MHz base clock, 1845MHz boost. It's always gone faster than that for me. Did I somehow disturb the magic inside and make it go back to doing what it was supposed to do all along?

 

[BillytheImpaler]

Update: I took it apart again and think the issue causing higher temps than my initial bench run is related to mounting pressure. Pushing on the decorative backplate seems to be bad. I've got some standoffs on the way and I'll try that out.

Still no clue what to do with the "thermal throttling" despite being under the thermal target by 23C.