Aurora R9, DIY project

Thank you @coldfish_91 for the detailed post. I can't wait to see the pics get approved!

We never tested the stock fans. First thing we did after taking it out of box was replacing the fans. Now, idling is about 28-30 degrees on CPU, about 32 degreee on GPU, and nosie is very minimal after we adjusted the curve.  Full load, CPU is about 55-60 and GPU is about 70. 

Had you tested the stock fans, your neighbor at the end of the block might dial emergency services because they thought a plane just landed in the house.

May I know what was used to load the CPU and GPU? 55-60 for 9900K seems a little low, unless there is no OC involved. If you OC all cores to 5GHz with no AVX offset, a small-FFT test from Prime95 should send the rig near to 80C or more. Similarly, squeezing that 250W of GPU you have there would stress the blower beyond its thermal limit.

There are two possible reasons, one is unstable voltage when CPU is consuming more power. Second is unstable PWM signal.  IMO, this problem seems to be related with PWM signal since CPU pump fan is working perfectly under stress test all the time, which shares the same +12V source with top fan but without PWM control. If it was becasue of unstable voltage, then pump fan should also slow but it did not. 

Thankyou thankyou thankyou so much for digging more into this. I've been facing the same issue and suspecting the same as you are about the PWM waveform. Since you know something about PWM control, I'm hoping you have a spare signal analyzer somewhere to probe it? 

In case you're curious, the difference of ~400RPM between 70% where you're pegging the fans at compared to 2400RPM @ 100% translates to ~4C difference in a single fan config. I measured this is by connecting the fan directly to the PSU using a SATA-to-4pin cable and have the fan run at full speed. Capping off at 70% may not be a bad idea because that fan on the top nearest to the bezel will be loud at 100% without the radiator muffling it out.

Phase 2: Upgrade 2080Ti blower to water cooled.

Using the same card or a different card? There are 2 major concerns specific to 2080 Ti: Minimum length for 2080Ti cards is 270mm and the number of VRMs on this card. With the nano case (quote by @Anonymous ), it is going to be tough to find space for the radiator and fan. I got away with my mod because I downgraded to a slim 120x15 fan and hid it in the bezel in front. For a 2080 Ti, that fan won't be enough to keep things cool. If you're using the same card, removing the shroud, you would need to check that something like Kraken G12 mount with the 92mm fan will be cooling those VRMs and GDDR modules sufficiently. If you're going for something else like EVGA Hybrid card, then VRM and GDDR cooling is taken care of.

Good luck!

For graphic card, my card is Dell RTX 2080TI with Samsung DDR so I suppose it is the newer version since 2080Ti had QC issue when it was just launched. Someone said it was because of using Micron memory and the newer one seems to use Samsung. So far, there is no issue with my card. 

Dell RTX2080Ti has TU102-300A-K1-A1 chip which allows factory OC ( I think that means you can potentially flash other vbios on it.).  This card is made by MSI, basically MSI RTX 2080TI AERO edition, but not for retail sale.

TU102-300-K1-A1 are Normal Chips (no factory OC permitted).
TU102-300A-K1-A1 are Binned Chips (factory OC permitted).
Both support manual overclocking

Our plan is to use G12 kit with H55 cooling kit to upgrade this factory card. Everything is ready and we are just waiting for the MSI base plate coming in. One drawback for using G12 kit is VRM would be exposed without proper heatsink attached on them and I think I have a solution for that, which will be posted here after it is done. 

My GPU only hit 70 degree max while doing Time Spy test with stock frequency. And Afterburner OC scanning shows that it should be able to be stable at +200 OC frequency (1550 base/1750 boost) with temp under 80, under stock cooling and without increasing power limit, so I am expecting to have more performance after upgrading to water cooling.

I tried Prime95 small FFT (max power/heat) to test it. 

Here is the result: 1. Enable OC (unlimited power) with frequency at 4.7 , temp is about 70-75 under stress test. The max TDP power could reach up to 190W...... freqency can only be held at 4.7 for all 8 cores for a few secs.

                              2. Disable  OC (stock, limited power at 118W) with frequency at 4.7, temp is about 55-60 under stress test since the power limit throttling will kick in much sooner. 

                             3. Enable OC (unlimited power) with OC to 5.0. Temp will soar to 90 and beyond. I stopped the test once the package temp go beyond 90 so I don't think this rig can handle OC to 5.0 with unlimited CPU power, it is way too much not only for the cooling but also for the MOBO. The VRM around CPU has fewer phases than other aftermarket ones. If you look at the MOBO graph in service manual, it only has 4 phases on the side, and 2 on the top . So at maximum, this MOBO only has 6 phases. If you check other brands, usually there are around 10 phases around CPU, providing way more stable voltage for CPU while OCing. I think this might be the reason why R9 cannot be stable at 4.7 all cores (or 5.0 for all cores)  for too long since there are too much pressure on phases while OCing.  The only way to get 5.0ghz here is to ENABLE OC and adjust freq  to 5.0. But be careful about it, since it will overwhelm the MOBO if you have too much load on all 8 cores.  I will test how it goes for gaming later , but for benchmarking and stress test, it is no no

This rig can handle stock 4.7 with unlimited power under my setup without thermal throttling but as I said, it would lower the frequency even before thermal throttling kicks in so the limiting factor is not temp in my case, it is because VRM and IC design limit cannot handle all core 4.7 for too long on our MOBO, which restrict the performance of 9900K here.  Yes, this MOBO cannot unleash the full potential of 9900K. 

EDIT: @GTS81 GTS81 is right. Under my setup, our MOBO can actually handle the power demand of 5.0 OC ENABLED with auto voltage. Max TDP is about 185W.  It will lower the TDP to 140W after 28-30s as pre-set in BIOS which you can change in XTU at your own risk.  Max temp under Cinebench R20 is 87 if 5.0Ghz OC ENABLED with all fans set at 70%. If we OC to higher frequenecy than 5.0Ghz, the temp/cooling will become the limiting factor (even though it has fewer phases than other brand, our MOBO can handle the power demand for 5.0Ghz OC).  However, at 100% speed,  my fans will still stall somehow immediately after starting stress test if OC is ENABLED, no matter it is 4.7Ghz or 5.0Ghz.  

Also, I made a mistake in my previous post, 9900K can only run at 4.7Ghz at 8 cores when you OC it to 5.0ghz. At 4.7 Ghz, it can only run at 4.38Ghz for 8 cores. 

The only way to get 5.0ghz here is to ENABLE OC and adjust freq  to 5.0. But be careful about it, since it will overwhelm the MOBO if you have too much load on all 8 cores.  I will test how it goes for gaming later , but for benchmarking and stress test, it is no no

Actually if you can hit 5.0GHz with all cores, all you have to do is turn on all the thermal, power, and current limit throttle indicators. This way you can see if your mobo is running out of juice by watching the current limit, no? In Aurora design, I do believe you'll hit thermal limit long before current limit kicks in. You did mention being able to hold on to the OC for only a few seconds. Was it the 28 seconds default Turbo Boost time frame? I recommend that you run the benchmark with a specific power, say 190W sunk, constantly, to see how your CLC behaves at full saturation.

I also noticed you managed to seat the pump in the original location with the sandwich. One of the tubing looks bent at a sharp angle. I was worried about that so I rotated the pump:

For graphic card, my card is Dell RTX 2080TI with Samsung DDR so I suppose it is the newer version since 2080Ti had QC issue when it was just launched. Someone said it was becasue of using Micron memory and the newer one seems to use Samsung. So far, there is no issue with my card

The tendency of this top range card to act up was the main reason why I asked if you're gonna mod this card or going with a different card altogether. Maybe the shroud screws don't have the warranty sticker on them so you can mod without worrying about RMA a few months down the road. I'd add that I'm still searching for the conclusion to what causes the artifacting because I'm biased (working at Micron) and curious. I don't work in the DRAM group so I don't have the details.

G12 mod pics:

Choice of AIO, gasket, and GPUShroud removed and TIM cleaned.

For 5.0Ghz OC, the cooling is not enough since the temp raises so fast. Can your rig handle the temperature with all cores at 5.0 ? 

For 4.7Ghz, somehow the cpu cannot hold all cores at 4.7/4.8 for too long. Can you run Cinebench R20 mutilcore and see what your score is ?  

 I pushed it in so I was a bit worried too. But so far, no issue. I think I will rotate it a bit as well, will reseat it later today. 

It won't be a comparison as my rig is only 9600K. The most I've gotten out of the 6C/6T is 4.8GHz @ 85C but AVX is maxed out at 4.6GHz. But I think our observation is similar: at 190W, the cooling is inadequate to keep the CPU steady.

For 9900K in my rig, the best I can do is 5.0Ghz with 4.7Ghz for all cores, which already brings temp up to about 85 to 90 with stress test.  

If we want all cores at 5.0Ghz, it has to be OC'ed to 5.2-5.3Ghz, in which case neither VRM or cooling can handle it.  Overall, I am very satisfied with the performance of CPU except for the weird fan behaviour. But bottom line is 70% speed does not make significant difference on cooling compared with 100%, so I am okay with it. Still curious about what causes the fan to slow tho.

But through monitoring by XTU, I found that the fan slowers when "hard" power limit throttling (not the random power limit throttling while idling) kicks in. For example in my test of OCing to 5.0Ghz with all cores running at 4.7Ghz,  CPU was running at 185W TDP stable for about 30s, then the power limit throttling kicked in bringing TDP down to 140W, at the same time, the fan slowered and then re-accelerated. My guess is that it could be the "hard cut" on PWM signal when power limit throttling kicks in to bring down TDP but also affects the speed of fan because they share the same PWM chip on our MOBO based on my obervation.   As what I know, VRMs of CPU also relly on PWM control to adjust the voltage.  Just my two cents.

I'm not sure why getting all cores to 5.0 GHz requires OC beyond that. If you look at XTU, you can see the multiplier effect of 1 through 8 cores (assuming it counts physical and not logical cores). I'd expect that the multiplier number between 1 and max cores would be the same hence no need to OC beyond that to get all cores to function at targeted frequency.

The "hard" power limit you mentioned is likely the PL1 time limit where after 28s, the CPU exits Turbo boost mode and drops the frequency. Interesting that you observed the fans spooling during that time because for me, as soon as the CPU temp goes beyond 80C, even within 15s of OC, with a fixed 100% fan speed, the fan starts to trip. And it doesn't speed up again. It just gets stuck at 9% speed and sends my CPU to a fiery death.

The VRMs you were referring to are the ones on the motherboard or inside the CPU die? When I last looked at BDW, we were using FIVR that has integrated VRs in the CPU itself. SKL team decided to replace them with a different VR solution and since then, I've not looked at the on-die VRM solution. Furthermore, the PWM control should be internally derived by the power control unit's register values to regulate the multiple voltage rails within the CPU. Maybe I should go re-read CFL's datasheet. 

Customized baseplate for Kraken G12 kit on 2080Ti is almost ready ! 

I have to cut the plate from MSI a bit tho to give space for the 4-pin connector since only Dell 2080Ti blower fan uses 4pin connection on the right-bottom side but all others use 14-pin on the right side.  

I will use Arctic Silver thermal adhesive to attach these heatsinks on the plate. For the VRM on the far left side, I will attach an active NVMe heat sink on it with PWM fan, which I will receive by the end of this week. 

solution for far side's VRM

update: If you are using regular card from the market, there should be no need to cut the plate. But you still need to file down the corners of A brackets coming with G12 kit to fit the base plate. I have not done this part yet but will post the pic once it is done. 

If I pre-set PWM duty circle at 100%, fan will slow once I start test. If I pre-set fan speed at 70%, it will slow when 28s window finishs and power limit throttling kicks in. I just updated to 1.0.4 BIOS, no change on fan behaviour. 

I have been waiting for someone to hack the G12 mod plate itself and you have done it, @coldfish_91 ! Looking forward to see the approved pics.

EDIT: The pics are visible and I’m hoping there will be more pics soon. Having a hard time imagining marrying the G12 together with the custom baseplate. I assume you want to have a 92mm fan blowing over the fins of the big heat sink array you have on the right? If that’s the case, then won’t the pump side be too far away from the A bracket? Or do you plan to go passive cooling for those fins? Or cut the G12 in half?

@GTS81 I will use 92mm fan to blow air onto those heat sink array on the right side. For the left side of pump, I will have an active heat sink with integrated 20mm fan to cool the VRM.

What do you mean by "If that’s the case, then won’t the pump side be too far away from the A bracket " ?  The position of G12 kit will not be changed. Pump will be at the orginal position and 92mm fan will be right above the mid area of heat sink array, no need to cut the G12 kit. You cannot cover the whole heat sink area with one 92mm anyway but I pick the heat sink with horizontal fins so that air can be spread around the whole area in a better way. 

@coldfish_91 , the H55 pump with the G12 plate pressing down on it results with only a few mm spacing between the G12 and the right side of the card. Maybe it's the picture and the array of heat sinks actually fit in that space. Anyway, looking forward to your next set of pics with the G12 installation.