Can the PCB Fan Connector still be used with Alphacool Eisblock Aurora ? (RTX 4090 Suprim)

[HoBahr]

I have a general question about the Alphacool Eisblock Aurora Acryl GPX-N RTX 4090 Suprim with Backplate (Article no.: 13459)

When installing this water block, can the circled fan connector on the PCB still be used? I’d like to continue using MSI Afterburner for fan control, but it only supports fans connected directly to the PCB. Is it possible to attach for example a 15cm Alphacool fan cable 4-pin to 4-pin extension (Article no.: 18673) before / when installing the water block so I can afterward connect the radiator fans to the PCB using the extension?

 

[Eddy]

You won’t be able to plug in a connector there anymore. There’s too little space between the acrylic and the header. At least I don’t know of any cable or adapter that would still fit.
With a water cooling system, the fans are usually controlled based on the water temperature. Alternatively, you can just set the fans to a fixed speed. Personally, I’ve never controlled the fans on my own PCs. Whether the GPU ends up at 45, 50, 55, or even 60°C doesn’t really matter with water cooling.

 

[HoBahr]

Thank you for the information. With a 4090, I like to keep the GPU temperature under 57°C under load (higher pump/fan speeds), while avoiding low VRAM temperatures when idle. Since GDDR6X memory is very temperature-sensitive, it needs to stay above a certain threshold. This creates a two-way balance:

1. Preventing the GPU from getting too hot under load.
2. Ensuring VRAM temperatures don’t drop below their safe minimum threshold.

So I could connect the pump and the fans to the mainboard and set fixed speeds but it will be tough to finding a sweet spot which suits both load and idle requirements.
I could use an Alphacool Eiszapfen to read the water temperature, but my mainboard has no temp sensor input. I do not find a temp sensor / fan/pwm controller in the Alphacool shop and the FanControl app does not read the GPU temp without having to run HWinfo at the same time I guess.
That's the reason I thought using the GPU fan connectors for the pump and the radiator fans as I can easily control these with MSI afterburner. Now I am not sure how I should set things up.

 

[Eddy]

Those are pretty strange requirements. Why does the GPU temperature need to stay under 57°C? What’s the point of that? And this thing about the RAM temperatures is something I’m hearing for the first time as well. Where did you get that information? Because that’s complete nonsense. Of course, every chip and every electronic component reacts to temperature fluctuations, but unless you plan to use the graphics card for more than 20 years, none of this really matters.


Controllers for reading temperatures and software control aren’t our market—we leave that to others. Aquacomputer, for example, offers everything you need with their Aquasuite and Aquaero for connecting sensors, reading them out, and controlling everything through software as you wish.
You can find sensors here, for example: https://shop.alphacool.com/en/shop/controllers-and-sensors/temperature-sensor/

 

[HoBahr]

Dear Eddy, as a potential customer evaluating Alphacool's cooling solutions, I was not amused to hear certain technical aspects being dismissed as "nonsense."

NVIDIA's GPU Boost technology relates to temperature, where higher temperatures reduce the possible GPU Boost clock speeds in 15Mhz bins ("steps") when reaching certain threshholds. This is how these cards work, at least from the GTX 1080TI up to the RTX 4090. The new RTX50 series' Blackwell architecture enhances this behaivor: Clocks now adapt 1000 times faster and in a more granular manner than all previous generations.

A non-watercooled 4090 normally runs between 40 to 50 degrees being idle or under light desktop use. Depending on the room temperature it can also be 45 to 55. Whatever the particular case, once the GPU core temperature passes 57 degrees, you'll lose the first 15Mhz bin from your maximum possible Boost Clock. This is just how this technology behaves. You will lose more 15MHz bins the higher the temperature of your GPU core gets.

GDDR6X memory uses on-die power management regulators which are designed assuming typical operating temps. That means, despite being throddled when becoming too hot, the memory chip does not like being “too cold” as well, because the regulator on it will not behave as expected anymore, which will cause crashes. So if for example you achive a nice VRM overclock at 50 degrees, your system may crash once you temperature fall to 40 or below.

This information is not new or nonsense. It has been discussed extensively for years in communities such as Overclock.net, TechPowerUp, and Guru3D just to mention a few. Even a quick search for “4090 cold memory bug” will return countless references to this behavior. Feel free to enjoy exploring it and then think again about claiming "nonsense" to things which may not yet be familiar to you.
Overall, the sweet spot of a 4090 is to keeping the GPU core temps below 57 and the VRAM temps above 50 in all load scenarios. With max. 55 on the core and min. 55 on the VRAM every 4090 user should be super happy.

But this balance only be achieved by some sort of fan/pump control. Some people even use "bad" thermal pads on the GDDR6X VRAM and "good paste" or PTM7950 on the GPU Core for balancing it all out.

Thank you for pointing out possible solutions by Aquacomputer. I will check them out.

 

[Eddy]

The technical specifications are well known to me. If you prefer to follow the information found in forums, I won’t stop you — but practical experience from end customers and especially in the server sector tells a very different story and does not reflect those claims at all. In practice, there are many other factors at play. That’s why, to me, these kinds of discussions are usually exaggerated or simply irrelevant. Based on sales figures, the number of actual issues is so small that we're talking about a failure rate in the range of 0.0X%.

But back to the topic:
If you want to maintain temperatures in that range, you’ll almost certainly need to control it through Aquasuite. It’s one of the very few programs that allows precise control based on water temperatures and target delta values. However, I would not recommend controlling the pump speed — at least not via a curve. That puts unnecessary stress on the pump and shortens its lifespan. Flow rate also has relatively little effect on cooling performance. You have to remember that a water cooling loop has thermal inertia.

For example, when the GPU suddenly goes under full load, it takes several seconds for the water temperature to react — and only then can a control system respond. The water needs time to heat up as it circulates. You can mitigate this a bit by placing the temperature sensor right at the outlet of the GPU cooler, and not — as many do — somewhere else in the loop, like at the radiator or reservoir.

What the ideal delta should be depends on your total radiator surface, water volume, and to some extent your flow rate — as long as it doesn't drop below 60–70 L/h. Beyond that, higher flow doesn’t add much cooling performance. Once you're over 100 L/h, the curve flattens out almost entirely. It’s far more effective to regulate temperature through the fan curve rather than pump speed.

Still, no matter how well you regulate the system, the inherent inertia of water cooling makes it impossible to hold temperatures exactly constant. I’d estimate potential fluctuations of around 5 K, in the worst-case scenario.

 

[HoBahr]

Hello Eddy, I know my card pretty well over the years and also of course its core and memory sweet-spots for optimal results. My first intention was to connecting the radiator fans to the PCB of the card because I can then easily control the fan speeds using MSI afterburner. This would have been my first attempt to reach my goal: 1. setting both a fixed pump and fan speed and look how it behaves under water. 2. trying to work with MSI afterburner's fan curve to try to stay inside the desired sweetspot. The nice thing is that I can "talk" to the GPU core temp using afterburner.
Now, as I know I won't be able to plug in the fans underneath the Eisblock Aurora's acryl, my idea was using maybe a Eiszapfen sensor. The "latency" of water as you mentioned makes using an Eiszapfen and a controller board still not as desireable than using the fan plug on the GPU PCB. Currently I am checking out if other water blocks maybe make it possible to continue using the plug. I am also considering directly soldering the four wires of a fan extension cable onto the PCB or the 4 pins, so that the space between the PCB and the acrylic would be sufficient enough to guide the plug of the extension cable out from underneath there. Another option would be to cut out a portion of the acrylic like it is on top where the 600W power connector is located. But this is something I have no tools for. Thank you for mentioning it is better to find a constant pump speed.

 

[Eddy]

I can’t really help you with HWiNFO — I have no hands-on experience with it myself, aside from occasionally checking temperatures with it.

Just to clarify the point about thermal inertia, so we’re not talking past each other:
When the GPU ramps up under load, you’ll quickly be able to measure a higher water temperature at the GPU outlet — that happens within seconds.
The actual problem is that increasing the fan speed at that moment won’t instantly cool the water down. The water first has to carry the heat through the entire loop before a temperature peak is reached where the fans can actually make a difference.

I hope that makes sense. The water has to first travel into the radiator, release its heat there, and then the fans can carry that heat away via the fins — and that whole process takes a bit of time.

This doesn’t mean your GPU will instantly spike to 60 °C and only cool down much later. Just run a test, and you’ll probably see for yourself how it behaves. A lot depends on your specific loop layout. Some people go straight from the CPU to the GPU or vice versa, others insert a radiator in between.

Controlling the fans directly through the GPU header won’t give you much of an advantage here. While the idea that it allows faster reaction is technically correct, the inherent inertia of the water loop will simply undermine that benefit.

As for modifying the acrylic — I wouldn’t recommend it. Certain wall thicknesses need to be maintained; otherwise, the acrylic can crack. The constant pressure from mounting and the temperature fluctuations could lead to a damaged cooler. Acrylic is a delicate material, which is why we use a fairly thick version of it. In the rear area, it's not as critical since you're far away from the mounting screws. But in any case, you’ll need a proper milling tool specifically designed for acrylic if you really want to go through with this. But any warranty or guarantee would no longer apply if you proceed with that.

Your idea of starting with fixed fan speeds is absolutely the right approach. Set everything to a fixed value and see what your maximum temperatures are under full load. That gives you an upper limit. Then check the system while idle — that gives you your minimum. From there, you can build a proper curve between those two points, and it should work quite well.

The level of precision you're aiming for when switching between idle, partial load, and full load will never work perfectly — no matter what.

 

[HoBahr]

yes .. it will never be "perfect." I just like to be in a sweetspot of maybe 15 degrees where I do not lose much boost clock and can maintain high vram clocks. My plan is to keep the loop as small as possible: just the GPU, pump, and radiator, without the CPU, and with short tubing runs. That way, the overall coolant circulation path remains very short.
For a start, the pump will probably be set to run at around half speed and the fans at low RPM. I’ll first check idle temperatures. Then, creating a sudden high load is simple, and I can observe how the system responds.
The only drawback is that I can’t directly tie fan speed to GPU temperature. I don’t want to rely on HWINFO just to read that value, especially since Afterburner is running all the time anyway. It would be like a waste not to use it and have to run HWINFO in the background all the time.
Since I’ll have to disassemble the card anyway, the simplest solution seems to be removing the 4-pin PWM connectors from its board and soldering in the wires from extension cables.

 

[Eddy]

If you're aiming for a 15 °C delta, then that really shouldn't be a problem.
Have you ever used a custom water cooling loop before? Let me give you an example:
We tested a 5090 tuned to 600 W, and the delta between water and ambient temperature under full load with a single 360 mm radiator was around 15 K. With some cooling optimizations, we even got that delta below 10 K. In an open setup, we never saw GPU temperatures above 50 °C with an ambient room temperature of around 21 °C.
At home, my 4080 and 4090 also consistently stay under 50 °C — even in the peak of summer and under full load (e.g., rendering), they don’t go above 55 °C. Both systems use three 360 mm radiators with Alphacool Rise fans locked at 700 RPM, in cases that definitely don’t have ideal airflow.
Honestly, I think your bigger challenge might not be keeping temps high enough, rather than getting them low.

 

[HoBahr]

Sounds good. Due to the computer's case I can only use a single 240 radiator with 2 fans. I could use a thicker one (45/60/Nexxos) tough if that makes sense.
And yes .. since you gain more performance with high VRAM clocks on the 4090 opposed to only core clocks, I want to be able to keep my VRAM as it is right now even under water - which means its temps can not fall below I think it's 47 degrees in my case. could be 45 .. not sure. I have to look. This would result in maybe 40 degrees GPU core temp i think. I am not sure yet as I am not at home. Keeping the GPU core temp from 40 (desktop use / idle) to 55 (full load) would be perfect as the VRAM would still be functioning maxed out and I would not loose Boost clock as well.
As mentioned .. I only have space for a 240 radiator. I could limit the GPU voltage to 1.050mv (max around 550 watts) or go full 1.100mv (666+ watts) depending on the BIOS I use. But that is not so important. I would be fine with 1.050mv and being in the sweetspot instead of gaining 30 to 60 Mhz more but ending up not being able to keeping it balanced.
Do you have any images of a MSI 4090 Suprim X and a Eisblock Aurora just to be sure this really will not fit ?! On the green lit image it looks like it almost does work out without any soldering:

 

[Eddy]

I can’t provide pictures of a specific card. We don’t have every GPU constantly lying around — that would be a bit much, considering we currently support around 200 graphics cards with all the latest coolers. Also, many coolers are developed based on the manufacturers’ 3D data. Shipping cards back and forth for every model would be madness.
A 60 mm thick radiator only really makes sense in a push-pull setup. If you only mount fans on one side, they’ll need to run at very high RPMs to get the most out of the radiator. In that case, I’d actually recommend going with the XT45, but also in push-pull — that’s usually a better-balanced combination. Thicker doesn’t always mean better in every situation.
That said, a 240 mm radiator is quite limited when it comes to handling that kind of heat — especially if you want the system to run quietly.

 

[HoBahr]

To give you the whole picture: I’m using an MSI Suprim “Liquid X,” the AIO version. Under max. load, the GPU reaches around 65–70°C, with VRAM in the 80s. Even tough it is "liquid," its not performing better than the air cooled version. Anyway - since the card is several years old, it’s time to replace the thermal pads and paste, as the not impressive cooling performance has declined a tiny bit.

Back on February 13, I bought an Alphacool Eisblock Aurora for it. The intention was that I sell the 4090 and get a 5090. I would pass the Eisblock along to the new owner especially since the stock AIO pump makes strange noises (which is also "normal" for these cards).

Anyway .. I’ve decided to skip the 5090 and wait for the 2nm 6090 instead, which should offer way better value for the money. That leaves me with the Eisblock, and I’m now considering how much I can improve thermals - ideally by around 15°C - by getting rid of the MSI "Liquid" stuff and pairing my 4090 PCB with the Eisblock and a dedicated pump and a different radiator instead of relying on the stock AIO setup.

That’s the situation. I thought using the Eisblock and a better Pump and Radiator will easily beat the MSI AIO sby about 15 degrees. The MSI AIO setup only comes with a 240 radiator as well.

 

[Eddy]

You will achieve better temperatures, that I can guarantee. However, I cannot guarantee that you will see such a large improvement with just a 240mm radiator. In fact, we usually recommend at least a 360mm radiator. Regardless of the thickness of a 240mm radiator, it will always perform worse than a 360mm. For example, a 240mm UT60 will never reach the cooling performance of a 360mm ST30. I somewhat doubt that you will reach your target temperatures.

It also sounds like you have a very compact case, which is disadvantageous for airflow. Even with water cooling, proper airflow is necessary. If you only run intake fans, the heat will build up inside, forcing the fans to work harder. If you only run exhaust fans, the question is whether enough fresh air can be drawn in. I can’t judge your exact case setup.

That’s why with our GPU AIOs we only offered models with 360mm radiators. The 240mm models were simply too weak by our standards. 600W of waste heat is about the same as what a small household radiator generates over roughly 1 meter in length (older models, not modern ones). That is an enormous amount of heat—don’t underestimate it.

 

[HoBahr]

It's a gaming PC underneath a TV stand (which can be rolled around in the room.) The fans are blowing out of the case and they have enough inlets to pull fresh air. Everything is The AIO is not performing that bad. I can push 670 watts through it and the cpu core won't reach more than 73 degrees (maybe 75 after hours of runtime.) Usually you are in the low to mid 60ies in normal gamesplay. Again .. if I could lower the temps by about 15 degrees I would constantly be in a sweetspot situation.
Unfortunately, there is no space for a 360 radiator.

 

[Eddy]

As much as I’d like to, I simply cannot guarantee you a 15° improvement in this case. The radiator is definitely the limiting factor. One option would be to install two 240mm ST30 radiators in a sandwich configuration. This works well as long as you route the loop into the first radiator first, and then into the rear one—not the other way around. We use this method in server racks as well. But keep in mind, that would be 60mm for the radiators plus another 50mm for the fans in thickness, since you would need at least two rows of fans in this setup. That already adds up to about 11cm in thickness.

With ST20 radiators, you could save about 2cm in total thickness. But even then, such a configuration would still perform better than a single thick radiator.
Beyond that, I can’t really think of other options if your available space is so limited.

 

[HoBahr]

This is my case:




This is the same case, with a single NexXxoS XT45:


Do you think maybe one ST30 and one ST20 would make sense ?

 

[Eddy]

I think so, that should work. It will just be a bit tricky when it comes to assembling everything. The fans need to go between the two radiators, but I’m not exactly sure how that can be managed in this case. For our server cases, we have special brackets for this. I’m not sure how it can be solved here. That’s the last remaining issue with the whole setup.

 

[HoBahr]

I think so, that should work. It will just be a bit tricky when it comes to assembling everything. The fans need to go between the two radiators, but I’m not exactly sure how that can be managed in this case. For our server cases, we have special brackets for this. I’m not sure how it can be solved here. That’s the last remaining issue with the whole setup.

I know a metalwork/locksmith shop where I could get a bracket. I think I end up with 4 fans. Like one pair pushes toward the ST20. Then, in between the ST20 and the ST30, there is another pair of fans which pulls air coming from the of the ST20 and pushes air toward the ST30. So the ST20 would have a push/pull situation and the ST30 has a push situation only. Maybe swapping the position of the ST20 and the ST30 also makes sense so that the thinner radiator is the one with only a push configuration. Not sure about that.
One remaining issue to me is the pump and how to fill the loop. Which pump will make sense and what would be the smallest reservoir and easiest way to fill the loop.

 

[Eddy]

I might be wrong, but I would probably set up the ST30 in push & pull and the ST20 only in push. It’s easier to push air through the ST20 than through the ST30. That way, the first fans are pushing while the second fans are also pulling air out of the ST30 at the same time. I imagine this could be the better option, but I don’t have any practical experience with it. We’ve only ever tested identical radiators in a sandwich setup.

In any case, I would always try to go with the Apex pump. DDC pumps are becoming obsolete, the DC-LT is small but also very weak, and the new DDCZero pump is not yet available. That one would actually be ideal.
So, my best advice is to measure carefully and check what we offer. A 120mm Core Flat Reservoir (14561) might be an option. But I don’t know if it would fit. You don’t necessarily have to mount it with the push mountings; it can also be mounted directly onto the case.