R9 5950x and RTX 3080 - Is my loop too small? Too little radiator? Too little pump?

[kamikazi]

The case is a Lian-Li PC011 Dynamic that I've modified by cutting the restrictive bottom floorplate out to better accommodate intake fans.. I've also replaced the front glass with a custom panel where one of my radiators is attached. I'm cooling a 5950x with EK Quantum block and a 3080 with EK Vector block. I have a Corsair XR7 360mm radiator (54mm thick) and a Corsair XR7 240mm radiator (54mm thick). The pump is an EK DDC 4.25 pump in the EK Lian-Li Distroplate. I have the pump set at a constant 4000 rpm and I have a curve on the fans. Everything is controlled with an Aquacomputer Aquaero 6. I have two inline water temp sensors, one in the distro-plate and one between the GPU and CPU. I also have some ambient temperature probes attached to the aquaero. Each radiator has Corsair LL120s on it in push configuration. I also have 2 LL120s as intakes on the bottom of the case. Coolant flow goes Reservoir->240 rad->360 rad->GPU->CPU->Reservoir. I'm running PBO on the CPU. I didn't mess with voltage other than to lower it on one of the CCDs by 15 mv. The 3080 EVGA FTW Ultra is at complete stock.

Anyway, my CPU will now spike up to 90C (Tctl/Tdie in HWINFO) with regularity when opening programs, etc. even when the GPU is running idle at 27 or 28C and ambient temp is 23 or 24C. When I was only cooling the CPU with a 360, it would never get close to that. I know it's just a spike and if I run 3D Mark, I actually get a better CPU score now than I did with the single 360. Is this problem solely the difference in my CPU being last in the loop? Is the CPU stretching it's legs with better cooling? In Microsoft Flight Simulator 2000 (Geforce Experience settings at 1440p), CPU gets as hot as the 80s at times, but normally runs in the 60s I think. The GPU tends to get to about 56C on the core (HWINFO reading) and stays there. I'm cooling based on liquid temperatures. With ambient temperature of usually 23 to 24C, the average liquid temperature gets up to around 37.5C. At that point, I'm kind of at equilibrium with all of my fans are running around 1700-1800 rpm, so not exactly quiet (max fan speed is 2150 or so and max pump speed is around 4500 for reference). That gives me an Air-Water delta of 13 to 14C. That seems high to me given the research I've done on other systems. Also, I see people with a 3080ti overclocked with my waterblock saying that they stay in the 40s on the core. Any ideas as to what I should look at first? Am I actually good to go? Pic of the system below.

 

[pendragon1]

maybe the air flow isnt balanced? you have 5 out and only 2 in? add a back intake maybe. loop order is a little odd, i normally have rads before pump but in the end it doesnt really matter much, if at all. like you said, it equalizes under a sustained load. the spikes on the cpu would have me questioning the mount/tim...

 

[Zarathustra[H]]

Yeah, the water to air delta seems a bit higher than it should be.

That said, I have no experience with the CPU, GPU, pump, or radiators you are using.

It's tough to tell from the pic, but it looks like all of your fans are blowing out. Do you have an intake anywhere? They may just be starved for airflow.

Other than that, I might try to investigate what kind of flow you are getting through the system. Something may be either kinked or blocked somewhere.

 

[pendragon1]

what kind of flow you are getting through the system.

forgot to mention that. kinda hard to tell how well its flowing in a plate though, isnt it? i know i judge my tube res's by the little "tornado" that forms...

 

[Zarathustra[H]]

forgot to mention that. kinda hard to tell how well its flowing in a plate though, isnt it? i know i judge my tube res's by the little "tornado" that forms...

Yeah, it is. There are some decent flowmeters, but they aren't that cheap. Since OP has an Aquaero, there are a few options. I have two of these in my build, but they seem to have discontinued them. I can't speak to the other models.

 

[kamikazi]

maybe the air flow isnt balanced? you have 5 out and only 2 in? add a back intake maybe. loop order is a little odd, i normally have rads before pump but in the end it doesnt really matter much, if at all. like you said, it equalizes under a sustained load. the spikes on the cpu would have me questioning the mount/tim...

I have questioned the TIM mount on the CPU, but I did it the same way I did last time. I got a nice even coverage across the whole cpu with Kryonaut. I may be able to crank down the thumb screws a little more. I really wish they were slotted for a screwdriver. I can also remove it and remount without destroying the loop. I left some slack in the tubing, but it will still be difficult. Yes, I have 5 fans out and only have 2 in on the bottom, but the bottom is wide open right underneath the front radiator and air can flow in from the back of the case. The loop order is odd, I'll agree to that. I had a hard time figuring out how to make it work. If I could go back in time, I wouldn't have bought the distroplate, I would have gotten a normal reservoir. I could go to one of the radiators last, but when I did it before, it requires lots of weird fittings and is very difficult to make it mate up.

Yeah, the water to air delta seems a bit higher than it should be.

That said, I have no experience with the CPU, GPU, pump, or radiators you are using.

It's tough to tell from the pic, but it looks like all of your fans are blowing out. Do you have an intake anywhere? They may just be starved for airflow.

Other than that, I might try to investigate what kind of flow you are getting through the system. Something may be either kinked or blocked somewhere.

2 on the bottom are blowing in while the radiators are blowing out. I judge my liquid flow by the bubbles coming in to the top of the radiator. I think I'm getting good flow. The pump is more powerful than a D5 in pressure and flow specs. However, I do have a lot of fittings that do 90 degrees in the system. The blocks are obviously the most restrictive parts, but all the fittings add up.

forgot to mention that. kinda hard to tell how well its flowing in a plate though, isnt it? i know i judge my tube res's by the little "tornado" that forms...

 

[pendragon1]

if your confident in you tim application, then id try tightening it first. yeah thumb-screw only is silly, sometime you cant reach and need to use a screw driver.

 

[Zarathustra[H]]

Just did some googling. Looks like the "High Flow 2" and "High Flow Next" have replaced the flow sensors I have if you decide to get one. You could probably also get away with their MPS 400, but it is not as capable (very narrow flow range). There are also an MPS 100 and MPS 200 but they generally have flow values below what you'd want to see in a loop. I don't know why they even sell them.

Flow sensors are a nice to have. They help you eliminate flow problems. Other than that, they are not incredibly useful on a regular basis.

You can use them to trigger automatic shutdown based on flow though, which is nice. If you suddenly have no flow, something is wrong (dead pump or catastrophic blockage or leak) and you'd probably want your system to stop running asap.

 

[evhvis]

The Corsair LL are not good radiator fans and thick radiators will cause you to have low airflow even at high rpms. Switching to EK-varder, Corsair ML or other higher pressure fans would probably help quite a bit, assuming you must have lights on your fans. The best 120mm radiator fans for high RPM are the Phanteks T-30 while the Noctua NF-A12x25 pwm will be slightly better at lower rpms, but neither have lights. Swapping them out would most likely help quite a bit on the water temps, but it will not be a miracle cure. The HW-labs LX (XR7 adds a corsairs logo to it) is low restriction on airflow for it's radiator size but it is still a thick radiator.

The thumb screws should be screwed in until they stop (run out of thread) and the springs will take care of the mounting pressure. There is no residue in the block and the coldplate, jetplate and in-out are all correctly set up? What are the temp differences between the CCDs in a 32 thread load?

The CPU hitting 90 is quite likely in several use cases if you run PBO and an 8 thread load if your water temp is in the mid 30s (all threads go to first CCD). It should not happen in a 32 thread load though. E.g. if you run cinebench or similar then your temps should be nowhere near 90, but prime95 308 on 8 threads will most likely throttle your CPU.

 

[Zarathustra[H]]

Another few thoughts.

What are you using as your control temperature to ramp up fans?

Are you using the actual chip temps (CPU and GPU) as read by the Aquaero software from the OS, or are you using your inline temperature sensors?

I highly recommend using the loop temp. If you use the direct chip temps what will happen is the following:

1.) During idle, the system will run fans very slow. Loop temp will build up over time until the water in the loop is very warm

2.) Once the CPU or GPU gets some load they will heat up, and the fans will spin up

3.) at this point the water in the loop is already pretty warm. The fans cannot lower that temp instantly. It will take anywhere from a couple to several minutes. Meanwhile your CPU and/or GPU are going to be running much hotter than desired.

4.) Eventually under prolonged load the loop temp will then eventually decrease in temperature allowing the CPU/GPU to get to desired temp.

This is obviously not desirable.

If you control your fans based on the temperature sensors in the water loop instead you will have MUCH better results, and not suffer from this time lag. This requires a little bit more work though.

The way I usually do it is to manually max the fan speed and run a worst case CPU+GPU benchmark. (as an example, simultaneous prime95 and heaven benchmark on loop). Then I allow the system to heat soak and the water temperatures to stabilize (this may take a while) and note what the water temp is when you reach stability. (This is also a nice time to note the temperature deltas between your loop temp and your block temp)

Then use the data you have collected to set your water target temp.

Decide what you want your CPU and GPU temperatures to be. Subtract their respective water->chip delta T's (they will be different for each) from the desired target temps, and set the loop target temp to the lower of the two values you wind up with.

If your flow is good, it shouldn't matter where in the loop your temp sensors are, but since I use PWM to control both my fan and pump speeds, my loop flow rate will often be quite low when the system is idle.

Because of this, I usually put one temperature sensor right after both blocks (which are after eachother in order) This temperature sensor is used to control pump speed. The theory being that if the water exiting the blocks is hotter than desired, the fastest way to correct that if in a low flow state is to increase the flow, and get more cool water in there.

I generally put the other temp sensor right before the two blocks in the loop order. This one controls the fan speed. The theory here being that if the incoming water temperature is getting too high, the only thing to do about it is to boost fans.

Also, since you have an Aquaero, I recommend you don't use fan/pump curves to control your fans and pump. Instead of a curve use Aquaeros excellent "set point" temperature control.

This is a version of what industrial control loops use called PID controller. There is a manual setting to input your P I and D values yourself, but this gets really complicated in a hurry. Process engineers get paid good money to optimize this stuff. If you get it wrong you can have bad oscillations, etc. Luckily the Aquaero has some presets (Slowest, slow, normal, fast, fastest). I'd start at normal, and if you feel it is too peaky, turn it down, if you feel it is too slow to respond, turn it up)

The benefit of this is that in curve modes your fans will often run faster than they need to to accomplish the job. For instance, lets say your target temp is 32C. When at high load, your system will be maxed out at 100% at 32C or above. But when you are at a lower load, your fans will be working harder than they need to to keep temperatures at a temperature that is below 32C, which makes no sense.

The "Set point" or "PID" controller works differently. It constantly monitors the temperature and its changes and will settle in right at the target temp, regardless of system load. (provided the cooling capacity is there). Are you at full load? fans are running fast and you are at 32C. Are you at idle? Fans will be running slowly and you will be at 32C.

It's much more efficient, and great for silent running.

 

[Dan_D]

maybe the air flow isnt balanced? you have 5 out and only 2 in? add a back intake maybe. loop order is a little odd, i normally have rads before pump but in the end it doesnt really matter much, if at all. like you said, it equalizes under a sustained load. the spikes on the cpu would have me questioning the mount/tim...

That doesn't matter as much as you'd think. I've seen systems that have no intake fans with decent'ish temps.

 

[kamikazi]

The Corsair LL are not good radiator fans and thick radiators will cause you to have low airflow even at high rpms. Switching to EK-varder, Corsair ML or other higher pressure fans would probably help quite a bit, assuming you must have lights on your fans. The best 120mm radiator fans for high RPM are the Phanteks T-30 while the Noctua NF-A12x25 pwm will be slightly better at lower rpms, but neither have lights. Swapping them out would most likely help quite a bit on the water temps, but it will not be a miracle cure. The HW-labs LX (XR7 adds a corsairs logo to it) is low restriction on airflow for it's radiator size but it is still a thick radiator.

The thumb screws should be screwed in until they stop (run out of thread) and the springs will take care of the mounting pressure. There is no residue in the block and the coldplate, jetplate and in-out are all correctly set up? What are the temp differences between the CCDs in a 32 thread load?

The CPU hitting 90 is quite likely in several use cases if you run PBO and an 8 thread load if your water temp is in the mid 30s (all threads go to first CCD). It should not happen in a 32 thread load though. E.g. if you run cinebench or similar then your temps should be nowhere near 90, but prime95 308 on 8 threads will most likely throttle your CPU.

I know the LL120s aren't the best, but they are the best of the worst (Corsair) so to speak. The white ones that I have are different from the standard LL120s, they run higher rpms and higher static pressures, but I agree, there are much better radiator fans. I do have a couple of Noctua NF12-A25s that I used in the past. I have 2 running in my entertainment center console. I think I have a couple I could swap in to the 240 radiator though if it gets to that.

Another few thoughts.

What are you using as your control temperature to ramp up fans?

Are you using the actual chip temps (CPU and GPU) as read by the Aquaero software from the OS, or are you using your inline temperature sensors?

I highly recommend using the loop temp. If you use the direct chip temps what will happen is the following:

1.) During idle, the system will run fans very slow. Loop temp will build up over time until the water in the loop is very warm

2.) Once the CPU or GPU gets some load they will heat up, and the fans will spin up

3.) at this point the water in the loop is already pretty warm. The fans cannot lower that temp instantly. It will take anywhere from a couple to several minutes. Meanwhile your CPU and/or GPU are going to be running much hotter than desired.

4.) Eventually under prolonged load the loop temp will then eventually decrease in temperature allowing the CPU/GPU to get to desired temp.

This is obviously not desirable.

If you control your fans based on the temperature sensors in the water loop instead you will have MUCH better results, and not suffer from this time lag. This requires a little bit more work though.

The way I usually do it is to manually max the fan speed and run a worst case CPU+GPU benchmark. (as an example, simultaneous prime95 and heaven benchmark on loop). Then I allow the system to heat soak and the water temperatures to stabilize (this may take a while) and note what the water temp is when you reach stability. (This is also a nice time to note the temperature deltas between your loop temp and your block temp)

Then use the data you have collected to set your water target temp.

Decide what you want your CPU and GPU temperatures to be. Subtract their respective water->chip delta T's (they will be different for each) from the desired target temps, and set the loop target temp to the lower of the two values you wind up with.

If your flow is good, it shouldn't matter where in the loop your temp sensors are, but since I use PWM to control both my fan and pump speeds, my loop flow rate will often be quite low when the system is idle.

Because of this, I usually put one temperature sensor right after both blocks (which are after eachother in order) This temperature sensor is used to control pump speed. The theory being that if the water exiting the blocks is hotter than desired, the fastest way to correct that if in a low flow state is to increase the flow, and get more cool water in there.

I generally put the other temp sensor right before the two blocks in the loop order. This one controls the fan speed. The theory here being that if the incoming water temperature is getting too high, the only thing to do about it is to boost fans.

Also, since you have an Aquaero, I recommend you don't use fan/pump curves to control your fans and pump. Instead of a curve use Aquaeros excellent "set point" temperature control.

This is a version of what industrial control loops use called PID controller. There is a manual setting to input your P I and D values yourself, but this gets really complicated in a hurry. Process engineers get paid good money to optimize this stuff. If you get it wrong you can have bad oscillations, etc. Luckily the Aquaero has some presets (Slowest, slow, normal, fast, fastest). I'd start at normal, and if you feel it is too peaky, turn it down, if you feel it is too slow to respond, turn it up)

The benefit of this is that in curve modes your fans will often run faster than they need to to accomplish the job. For instance, lets say your target temp is 32C. When at high load, your system will be maxed out at 100% at 32C or above. But when you are at a lower load, your fans will be working harder than they need to to keep temperatures at a temperature that is below 32C, which makes no sense.

The "Set point" or "PID" controller works differently. It constantly monitors the temperature and its changes and will settle in right at the target temp, regardless of system load. (provided the cooling capacity is there). Are you at full load? fans are running fast and you are at 32C. Are you at idle? Fans will be running slowly and you will be at 32C.

It's much more efficient, and great for silent running.

Thanks for the write up. I need to tackle that when I'm at home. Right now, I'm using the average of my two liquid temperature sensors to control fan speed. I have the fan speeds bottoming out around 800 rpm as I really can't hear them in the room at that speed. They start moving up with liquid temperature and max out when liquid temp hits 40C I believe. They're around 1750 rpm when liquid temp is 37.5C. That seems to be equilibrium with Flight Simulator running. I'll have to give the Set Point controller a try. I hadn't so far because my liquid temps seem to rise very quickly with any I just assumed my fans would stay maxed out all the time with a set point.

That doesn't matter as much as you'd think. I've seen systems that have no intake fans with decent'ish temps.

That's my thoughts as well. There are plenty of nooks an crannies for air to be pulled into my case. Also, if you put your hands above the two intake fans, they are moving a lot of air compared to the air I feel coming through the radiators. I would guess if you measured the volume of air on the intake fans and compared it to the volume of air exiting the radiators, it probably wouldn't be far off. I can easily test the intake problem by testing with the side panel off.

 

[vegeta535]

Should be enough. I run a 3080ti and a 5950x and my CPU max out in the low 70s. That is with a 54mm 360 and a 30mm 360 running a D5 pump on full blast. I am using the O11 XL with 6 intake and 4 exhaust Lian LI SL120 unifans. Or in them all at around 1000 rpm since that's the highest they go with out being loud. GPU is low to mid 50s.

 

[kamikazi]

Should be enough. I run a 3080ti and a 5950x and my CPU max out in the low 70s. That is with a 54mm 360 and a 30mm 360 running a D5 pump on full blast. I am using the O11 XL with 6 intake and 4 exhaust Lian LI SL120 unifans. Or in them all at around 1000 rpm since that's the highest they go with out being loud. GPU is low to mid 50s.

Do you measure your coolant temperature?

 

[vegeta535]

N.oknlllkp

Do you measure your coolant temperature?

Nope.

 

[kamikazi]

There are some nice charts here when Techpowerup reviewed the 360 rad I use. The created a load of over 500 watts and measured deltas with different speed fans. They got a 11.5 degree delta at 1200 rpm. I've got way more rad with probably less wattage and I'm at a much higher delta at a much higher rpm. Techpowerup review

 

[legcramp]

What's your temp running cinebench multi-core? Have you simply re-mounted your heatsink and check how the thermal paste was spread after lifting the cpu block?

 

[Evil Scooter]

I would not expect to see spikes up to 90c with that setup. How much vcore? As noted above... I'd check the TIM application and waterblock seating right out of the gate.

 

[pendragon1]

I would not expect to see spikes up to 90c with that setup. How much vcore? As noted above... I'd check the TIM application and waterblock seating right out of the gate.

vcore is also something i forgot to mention... mobo could be jacking it up.

 

[legcramp]

Yup at everything stock with PBO on, (which I use nowadays.. set it and forget it..) Ryzen spikes up to 80-90s under normal usage and this is normal for these chips. If someone has a chip that stays around 75-80c no matter what they probably have PBO off or a custom vcore curve etc.

Here's my 5900X w/ 280mm AIO watching YouTube and a few apps, temps were the same when running a 360mm loop last year:

And like pendragon1 said it's probably how the chip at stock settings sets the vcore dynamically... as you can see in my screenshot it ranges from 1.325 to 1.444 depending on what I am doing. If you set your vcore in the bios to a static vcore like 1.325, I bet your temperatures will look great but your boost speed might take a hit somewhere.

 

[kamikazi]

I haven't elevated vcore other than running pbo. I slightly lowered some of the cores on the crap CCD, but just by 15 mVolts.

Thanks for all the responses here. I hope to take a look at this at home tonight or tomorrow and work on trying some things out. I'll report back. I really don't want to rebuild this loop. It was a mess getting it re-built the first time. I did the Mayhem's radiator cleaning on both rads, one was new and the other had been used for about a year and a half. I disassembled the CPU block and cleaned it. The GPU block was new. The new pump I installed leaked and I had to get a replacement. Speaking of that, when I used the old stock pump, I could see how much slower the fluid flowed compared to my old cpu only loop. I'm starting to think that the loop itself is just really restrictive.

 

[Evil Scooter]

^^ Check the AMD processor thread as well.. I garnered lots of good info there and I'm able to keep my 5900x in the 60c's while gaming with a few cores boosting to 4.95Ghz and running benchmarks.

https://hardforum.com/threads/ryzen-5900x-overclocking-help.2005401/

 

[evhvis]

There are some nice charts here when Techpowerup reviewed the 360 rad I use. The created a load of over 500 watts and measured deltas with different speed fans. They got a 11.5 degree delta at 1200 rpm. I've got way more rad with probably less wattage and I'm at a much higher delta at a much higher rpm. Techpowerup review

They are testing in free air though and with different fans so not really comparable. Having case panels, dust filters, negative pressure case setup etc. will make the fans work much harder. Btw. if you haven't done so already, remove the dust filters from front and top as they are not needed when the air is exiting the case.

 

[kamikazi]

I tried to get Unigine Heaven installed to stress the GPU. However, I'm getting the classic MSVCP100.dll couldn't be found error. Every single time I reload Windows, I get this garbage. Then I google and follow all instructions on how to get it to install and it still doesn't. So fun uninstalling and reinstalling all the different versions of Visual C++ back to 2010 and none of them taking care of the problem. This is so infuriating.

ETA: I was able to pull the file off a backup of my old C Drive. However, the new problem is that when running Prime 95 and then starting Heaven, it kicks off some sort of high pitched alarm on my motherboard. If I kill Heaven, it immediately stops. It comes right back on when I restart Heaven. I don't have any alarm set on the Aquaero. In BIOS have all the fan controls disabled since I'm using Aquaero. I can't seem to find any setting for this. I'm thinking when the CPU stays above 85 for any period of time, that may be setting it off. I need to find that setting in Asus BIOS I guess. Any ideas?

 

[pendragon1]

try this vc: https://www.microsoft.com/en-us/download/details.aspx?id=26999
bios' qfan should have a max temp, try raising it.

 

[JSHamlet234]

If you're getting higher temps than before even when the GPU is idle and liquid temps are not higher than what they were before, then I would say you either have a bad mount, or you have an air bubble trapped in the block.

 

[kamikazi]

What's your temp running cinebench multi-core? Have you simply re-mounted your heatsink and check how the thermal paste was spread after lifting the cpu block?

Just under 85C, but because I'm ramping fans based on water temp, the fans don't ramp at all during a run because the water temp is still relatively low.

try this vc: https://www.microsoft.com/en-us/download/details.aspx?id=26999
bios' qfan should have a max temp, try raising it.

I thought I scoured the QFan settings and I just couldn't find anything. But it was late and I was tired, I need to look harder. There's an outside chance my Aquaero is sounding off, but I have no alarms set and I know that it did sound off when I did a firmware update and it was louder.

If you're getting higher temps than before even when the GPU is idle and liquid temps are not higher than what they were before, then I would say you either have a bad mount, or you have an air bubble trapped in the block.

I think I am going to check out my mount. I'm 99% on it, but I need to explore all options. However, my water temp rises fairly quickly in my opinion, so that heat is coming from somewhere.

 

[Dan_D]

Yup at everything stock with PBO on, (which I use nowadays.. set it and forget it..) Ryzen spikes up to 80-90s under normal usage and this is normal for these chips. If someone has a chip that stays around 75-80c no matter what they probably have PBO off or a custom vcore curve etc.

Here's my 5900X w/ 280mm AIO watching YouTube and a few apps, temps were the same when running a 360mm loop last year:

View attachment 439476

And like pendragon1 said it's probably how the chip at stock settings sets the vcore dynamically... as you can see in my screenshot it ranges from 1.325 to 1.444 depending on what I am doing. If you set your vcore in the bios to a static vcore like 1.325, I bet your temperatures will look great but your boost speed might take a hit somewhere.

Where as my temps are crap (partially due to my specific waterblock) but my boosts are good. Mine's hitting 5.1GHz and more all the time. It's "stock" but I'm letting the ASUS board do its thing.

 

[evhvis]

What is your wattage during PBO and what was your water temp? The water temp rising quickly in the beginning is normal and it probably takes 10-15 minutes to reach stable state if you lock the fan speed. On a 5900x with EK velocity block and around 200w pbo with 28 degree water temp gave around 74 degrees on the hottest CCD IIRC. The 5950x shold run cooler with a 32 thread load though as it distributes the watts over more cores so your temps seem to be on the high side.

Have you tried adjusting pump speed to see how it influences your temps? E.g. 50%, 80% and 100% as the EK velocity blocks love high flowrate on AMD. If the difference between 80% and 100% is many degrees then there might be a flow problem, otherwise I would re-paste, re-mount and be very methodical about it. The difference between a good and bad mount shouldn't be more than a few degrees on that block unless you have bad contact, but it is worth trying. Would first check that all the thumb screws are all the way in though, as they give quite a bit of resistance before they run out of thread due to the springs. Basically often had 1 rotation left of thread when they gave significant resistance.

 

[kamikazi]

What's your temp running cinebench multi-core? Have you simply re-mounted your heatsink and check how the thermal paste was spread after lifting the cpu block?

Here is what I got before I made any changes. 88 degrees, 230 Watts. Fans hadn't yet ramped.

 

[learners permit]

Where as my temps are crap (partially due to my specific waterblock) but my boosts are good. Mine's hitting 5.1GHz and more all the time. It's "stock" but I'm letting the ASUS board do its thing.

Whats the cas on them fancy rams ya got Dan?

 

[evhvis]

Here is what I got before I made any changes. 88 degrees, 230 Watts. Fans hadn't yet ramped.
View attachment 440081

Difference between CCDs and TCTL seem normal and 230 watts will make it run hot. My guess is that the paste/mount is probably OK but temps are probably slightly on the high side even for PBO unless you had very high water temp. Did you check temp diffs between different pump speeds to eliminate the possibility of flow issues?

Btw. keep in mind that max temps can be random as long as PB or PBO is involved as there might be single core spikes that dump a lot of watts into a few of the cores. E.g. something might run on 4-5 threads for a little while and your CPU will dump 150+ watts into a tiny area of your CPU. This causes spot heat and that is difficult to transfer efficiently to the cooling system. Average with a steady load is often more informative.

Running cinebench on a loop for all core and running prime95 308 or newer on 4-8 cores, smallest FFTs and hyperthreading on for 5-6 minutes and then reading the average should give a rough estimation of normal sustained all core and worst case temps respectively. Running 4 threads of smallest FFTs with in prime95 gives me the highest temps when running PB only. For PBO it is 6 cores+hyperthreading with prime95 308. It gives me the absolute worst temp on my 5900x and it is the only thing that will bring my CPU into the high 80s. No real world load comes even close to the temps that I get with prime95 308 and older versions of prime95 are nowhere near those temps either. The hyperthreading on seems to be very efficient in producing heat. E.g. highest temps I manage to get with prime95 287 is around 80 degrees and all core+hyperthreading in prime95 308 sits in the low 80s with PBO on.

 

[Dan_D]

Whats the cas on them fancy rams ya got Dan?

 

[kamikazi]

try this vc: https://www.microsoft.com/en-us/download/details.aspx?id=26999
bios' qfan should have a max temp, try raising it.

i believe I found the source of the annoying tones when pushing the system.

I first narrowed down that the sound was only happening when the gpu was under load. After I got fed up and left the computer alone for the night, I searched for tones coming out of graphics cards and in almost every case, they determined it was overloading a UPS. I did recently plug my server with a 3900x into this same UPS with the computer I'm working on. That server is running Blue Iris, so it always has a small load on it. Considering my main computer is pulling almost 600 watts on just the cpu and gpu alone when I'm testing it, it wouldn't take much to bring this UPS to it's knees. I'll verify tonight.

 

[kamikazi]

Another few thoughts.

What are you using as your control temperature to ramp up fans?

Are you using the actual chip temps (CPU and GPU) as read by the Aquaero software from the OS, or are you using your inline temperature sensors?

I highly recommend using the loop temp. If you use the direct chip temps what will happen is the following:

1.) During idle, the system will run fans very slow. Loop temp will build up over time until the water in the loop is very warm

2.) Once the CPU or GPU gets some load they will heat up, and the fans will spin up

3.) at this point the water in the loop is already pretty warm. The fans cannot lower that temp instantly. It will take anywhere from a couple to several minutes. Meanwhile your CPU and/or GPU are going to be running much hotter than desired.

4.) Eventually under prolonged load the loop temp will then eventually decrease in temperature allowing the CPU/GPU to get to desired temp.

This is obviously not desirable.

If you control your fans based on the temperature sensors in the water loop instead you will have MUCH better results, and not suffer from this time lag. This requires a little bit more work though.

The way I usually do it is to manually max the fan speed and run a worst case CPU+GPU benchmark. (as an example, simultaneous prime95 and heaven benchmark on loop). Then I allow the system to heat soak and the water temperatures to stabilize (this may take a while) and note what the water temp is when you reach stability. (This is also a nice time to note the temperature deltas between your loop temp and your block temp)

Then use the data you have collected to set your water target temp.

Decide what you want your CPU and GPU temperatures to be. Subtract their respective water->chip delta T's (they will be different for each) from the desired target temps, and set the loop target temp to the lower of the two values you wind up with.

If your flow is good, it shouldn't matter where in the loop your temp sensors are, but since I use PWM to control both my fan and pump speeds, my loop flow rate will often be quite low when the system is idle.

Because of this, I usually put one temperature sensor right after both blocks (which are after eachother in order) This temperature sensor is used to control pump speed. The theory being that if the water exiting the blocks is hotter than desired, the fastest way to correct that if in a low flow state is to increase the flow, and get more cool water in there.

I generally put the other temp sensor right before the two blocks in the loop order. This one controls the fan speed. The theory here being that if the incoming water temperature is getting too high, the only thing to do about it is to boost fans.

Also, since you have an Aquaero, I recommend you don't use fan/pump curves to control your fans and pump. Instead of a curve use Aquaeros excellent "set point" temperature control.

This is a version of what industrial control loops use called PID controller. There is a manual setting to input your P I and D values yourself, but this gets really complicated in a hurry. Process engineers get paid good money to optimize this stuff. If you get it wrong you can have bad oscillations, etc. Luckily the Aquaero has some presets (Slowest, slow, normal, fast, fastest). I'd start at normal, and if you feel it is too peaky, turn it down, if you feel it is too slow to respond, turn it up)

The benefit of this is that in curve modes your fans will often run faster than they need to to accomplish the job. For instance, lets say your target temp is 32C. When at high load, your system will be maxed out at 100% at 32C or above. But when you are at a lower load, your fans will be working harder than they need to to keep temperatures at a temperature that is below 32C, which makes no sense.

The "Set point" or "PID" controller works differently. It constantly monitors the temperature and its changes and will settle in right at the target temp, regardless of system load. (provided the cooling capacity is there). Are you at full load? fans are running fast and you are at 32C. Are you at idle? Fans will be running slowly and you will be at 32C.

It's much more efficient, and great for silent running.

What is your wattage during PBO and what was your water temp? The water temp rising quickly in the beginning is normal and it probably takes 10-15 minutes to reach stable state if you lock the fan speed. On a 5900x with EK velocity block and around 200w pbo with 28 degree water temp gave around 74 degrees on the hottest CCD IIRC. The 5950x shold run cooler with a 32 thread load though as it distributes the watts over more cores so your temps seem to be on the high side.

Have you tried adjusting pump speed to see how it influences your temps? E.g. 50%, 80% and 100% as the EK velocity blocks love high flowrate on AMD. If the difference between 80% and 100% is many degrees then there might be a flow problem, otherwise I would re-paste, re-mount and be very methodical about it. The difference between a good and bad mount shouldn't be more than a few degrees on that block unless you have bad contact, but it is worth trying. Would first check that all the thumb screws are all the way in though, as they give quite a bit of resistance before they run out of thread due to the springs. Basically often had 1 rotation left of thread when they gave significant resistance.

I hammered it last night with Prime 95 Small FFTs (not sure which version of Prime) along with Heaven looping at 1440P for about a half hour. Fans and pump were both at 100 percent. That gets me the fans at close to 2500 rpm and pump at close to 4500 rpm. Before, I was running the pump at 4000 rpm, but I haven't done an apples to apples test with different pump speeds. I started with ambient at 24C and it rose to 26C during testing. The CPU sat at 90C pretty much constantly. The GPU got up to 57C. Liquid temperature started at 30C or so and was at 38.5C when I stopped. It never stopped creeping up, but was edging up ever slowly. Of course, ambient was as well. The delta between air and water peaked at 12.5C. If I had run longer, it may have grown, not sure. CPU maxed at 220 watts or sow and gpu at close to 380 watts.

Before this test, I had tried running 3DMark Firestrike and it actually got the CPU up to above 90C very early on and the computer black screened very early in the test. The motherboard lit up the DRAM error light and showed a code of 00 which means nothing. I had to turn power off at the power supply to shut it down. Bluescreen viewer showed no error afterward. I have seen the DRAM error light recently when I had installed iCUE after I had installed Aquasuite. Same behavior, computer hard locks and you have to kill power to the power supply to shut down. Holding down the power button does nothing. I had rolled back Windows using system protection prior to insalling iCUE, but it kept doing the same thing. So, I ran a painfully long RAM test using Memtest86 at boot. All 4 sticks passed. So I ran them one at a time in the first RAM slot and they each passed. Then I ran one of the sticks in the only other RAM slot you can run alone and it passed, so I'm pretty sure it's not a RAM issue. A reinstall of Windows fixed it and I hadn't seen the RAM error light since then until I was running 3DMark. However, I did run that Prime and Heaven session afterward and got no errors.

I'm getting discouraged with this system with all the work I've put in and all the problems. I have also noticed that when I have the pump at full tilt, some of the tiny bubbles at the top of the reservoir make their way all the way down to the pump and get sucked in. Those build up over time and will show up as a small bubble in the CPU block that I have to tilt out. However, I don't think I have a bunch of air in the radiators.

 

[evhvis]

The prime version that nearly boils my CPU has "Smallest FFTs..." and a checkbox for "Use hyperthreading...". Versions without these options run much cooler on my system and are around worst case temps, but smallest FFTs and hyperthreading cause significantly higher temps when both of those options are used.

With firestrike the CPU will hit high temps during collection of system information, but during the firestrike benchmark itself the CPU should be cool (it doesn't even reach PB PPT limits during the benchmark, only during loading). Even with 40 degree water temp it should stay way below 90 during the benchmark itself. For reference my 5900x hits low 80s during collection of system info but during the benchmark it was 50 for ccd1, 32 for ccd2 and 56 for TCTL on average with short peaks of 65 degrees in loading screens with PBO on. My water temp was just around 27.5 degrees exiting the CPU block. My guess is that you should be within 60-70 average and peak of 75-80 with your water temps in that benchmark. The temps during collection of system information will most likely go beyond 90, but as soon as it starts rendering images it shouldn't get anywhere near 90.

 

[kamikazi]

The prime version that nearly boils my CPU has "Smallest FFTs..." and a checkbox for "Use hyperthreading...". Versions without these options run much cooler on my system and are around worst case temps, but smallest FFTs and hyperthreading cause significantly higher temps when both of those options are used.

With firestrike the CPU will hit high temps during collection of system information, but during the firestrike benchmark itself the CPU should be cool (it doesn't even reach PB PPT limits during the benchmark, only during loading). Even with 40 degree water temp it should stay way below 90 during the benchmark itself. For reference my 5900x hits low 80s during collection of system info but during the benchmark it was 50 for ccd1, 32 for ccd2 and 56 for TCTL on average with short peaks of 65 degrees in loading screens with PBO on. My water temp was just around 27.5 degrees exiting the CPU block. My guess is that you should be within 60-70 average and peak of 75-80 with your water temps in that benchmark. The temps during collection of system information will most likely go beyond 90, but as soon as it starts rendering images it shouldn't get anywhere near 90.

Sounds like I have the same Prime 95 as you. It has those options, I just chose Small FFTs and use hyperthreading. I noticed that with firestrike as well in the past, it cooks the cpu first then calms down when the bench is actually running. Mine happened to lock up within 15 seconds of the actual bench, so I can't speak to the sustained speeds. My water temp is typically never below 29 degrees C even idling. At idle, my water temp is usually between 5 and 6 degrees above ambient. The room is usually 23 or 24C ambient.

 

[Zarathustra[H]]

I hammered it last night with Prime 95 Small FFTs (not sure which version of Prime) along with Heaven looping at 1440P for about a half hour. Fans and pump were both at 100 percent. That gets me the fans at close to 2500 rpm and pump at close to 4500 rpm. Before, I was running the pump at 4000 rpm, but I haven't done an apples to apples test with different pump speeds. I started with ambient at 24C and it rose to 26C during testing. The CPU sat at 90C pretty much constantly. The GPU got up to 57C. Liquid temperature started at 30C or so and was at 38.5C when I stopped. It never stopped creeping up, but was edging up ever slowly. Of course, ambient was as well. The delta between air and water peaked at 12.5C. If I had run longer, it may have grown, not sure. CPU maxed at 220 watts or sow and gpu at close to 380 watts.

Before this test, I had tried running 3DMark Firestrike and it actually got the CPU up to above 90C very early on and the computer black screened very early in the test. The motherboard lit up the DRAM error light and showed a code of 00 which means nothing. I had to turn power off at the power supply to shut it down. Bluescreen viewer showed no error afterward. I have seen the DRAM error light recently when I had installed iCUE after I had installed Aquasuite. Same behavior, computer hard locks and you have to kill power to the power supply to shut down. Holding down the power button does nothing. I had rolled back Windows using system protection prior to insalling iCUE, but it kept doing the same thing. So, I ran a painfully long RAM test using Memtest86 at boot. All 4 sticks passed. So I ran them one at a time in the first RAM slot and they each passed. Then I ran one of the sticks in the only other RAM slot you can run alone and it passed, so I'm pretty sure it's not a RAM issue. A reinstall of Windows fixed it and I hadn't seen the RAM error light since then until I was running 3DMark. However, I did run that Prime and Heaven session afterward and got no errors.

I'm getting discouraged with this system with all the work I've put in and all the problems. I have also noticed that when I have the pump at full tilt, some of the tiny bubbles at the top of the reservoir make their way all the way down to the pump and get sucked in. Those build up over time and will show up as a small bubble in the CPU block that I have to tilt out. However, I don't think I have a bunch of air in the radiators.

There really should not be bubbles moving around inside your loop after the system hits stability after being filled.

It sounds to me like your issue may be that you still have air trapped in the system.

A few thoughts:

1.) Did you do "case gymnastics" after your fill? Varying pump speed up and down and leaning the case on its sides and front and back up again, while the pump is running to try to get all the air out. You can keep tilting and banging the system around for hours and still not get all the air out. This is usually necessary to get all the air out of the system. With some systems it clears out on it's own over time, but with others depending on the loop and how easy it is for freed bubbles to get sucked back in, it doesn't and needs lots of case movement

2.) What do you use as your coolant? Air bubble problems are a lot less rare if you use a coolant that contains a good surfactant. If you just use straight distilled water you are going to get much more air caught in the system.

 

[JSHamlet234]

I hammered it last night with Prime 95 Small FFTs (not sure which version of Prime) along with Heaven looping at 1440P for about a half hour. Fans and pump were both at 100 percent. That gets me the fans at close to 2500 rpm and pump at close to 4500 rpm. Before, I was running the pump at 4000 rpm, but I haven't done an apples to apples test with different pump speeds. I started with ambient at 24C and it rose to 26C during testing. The CPU sat at 90C pretty much constantly. The GPU got up to 57C. Liquid temperature started at 30C or so and was at 38.5C when I stopped. It never stopped creeping up, but was edging up ever slowly. Of course, ambient was as well. The delta between air and water peaked at 12.5C. If I had run longer, it may have grown, not sure. CPU maxed at 220 watts or sow and gpu at close to 380 watts.

Before this test, I had tried running 3DMark Firestrike and it actually got the CPU up to above 90C very early on and the computer black screened very early in the test. The motherboard lit up the DRAM error light and showed a code of 00 which means nothing. I had to turn power off at the power supply to shut it down. Bluescreen viewer showed no error afterward. I have seen the DRAM error light recently when I had installed iCUE after I had installed Aquasuite. Same behavior, computer hard locks and you have to kill power to the power supply to shut down. Holding down the power button does nothing. I had rolled back Windows using system protection prior to insalling iCUE, but it kept doing the same thing. So, I ran a painfully long RAM test using Memtest86 at boot. All 4 sticks passed. So I ran them one at a time in the first RAM slot and they each passed. Then I ran one of the sticks in the only other RAM slot you can run alone and it passed, so I'm pretty sure it's not a RAM issue. A reinstall of Windows fixed it and I hadn't seen the RAM error light since then until I was running 3DMark. However, I did run that Prime and Heaven session afterward and got no errors.

I'm getting discouraged with this system with all the work I've put in and all the problems. I have also noticed that when I have the pump at full tilt, some of the tiny bubbles at the top of the reservoir make their way all the way down to the pump and get sucked in. Those build up over time and will show up as a small bubble in the CPU block that I have to tilt out. However, I don't think I have a bunch of air in the radiators.

You did go from 220W into 3x120mm rad space to 600W into 5x120 rad space. In doing so, you went from 73.3 watts per 120mm to 120 watts per 120mm, so your air to water delta under combined load will be 63% higher than before based on that factor alone. So if your delta is 12.5C now, that means it was 7.6 before. So that would make you run 5 degrees hotter under a combined CPU GPU load than you did before you brought the GPU and addl' 240 rad into the mix. Hopefully you already knew that going in.

My overall impression is that you're just pushing your chip a little too hard for the current state of your cooling. You can't necessarily run the chip just as hard as you did before if it's 5C hotter. You're also running 4 dual-rank sticks at 3600, which is a heavy load for the IOD. My guess is that your temps are getting too high under combined loads, and it's pushing the IOD off the brink of stability. Since it looks like you can't easily add more rad space, I would dial down the power limits on the CPU to get those 5 degrees back and then call it a day. There's math for that too. If your water is at 37.5 and your CPU is at 90, then the delta is 52.5. 5 less is 47.5. So 47.5/52.5*220W = 199W. So if you lower your PPT to 199, you'll reduce your package temp to 85.

 

[pendragon1]

i believe I found the source of the annoying tones when pushing the system.
View attachment 440160

I first narrowed down that the sound was only happening when the gpu was under load. After I got fed up and left the computer alone for the night, I searched for tones coming out of graphics cards and in almost every case, they determined it was overloading a UPS. I did recently plug my server with a 3900x into this same UPS with the computer I'm working on. That server is running Blue Iris, so it always has a small load on it. Considering my main computer is pulling almost 600 watts on just the cpu and gpu alone when I'm testing it, it wouldn't take much to bring this UPS to it's knees. I'll verify tonight.

lol so it was the ups overload alarm? easy fix there.