All-in-one liquid coolers are commonplace for CPUs these days, but slapping a closed-loop cooler on one’s graphics card still takes a bit of elbow grease. Finding the right third-party cooler bracket and a compatible graphics card can require a fair bit of research, and disassembling a perfectly fine graphics card to Frankenstein a liquid-cooling bracket to it can be intimidating for the inexperienced. Corsair wants to take all of the guesswork out of liquid-cooling a GeForce GTX 1080 Ti with its Hydro GFX. This card boasts a pre-installed liquid cooler that’s ready to rock out of the box.
The Hydro GFX GTX 1080 Ti is actually a collaborative effort between MSI and Corsair. MSI provides the GeForce GTX 1080 Ti card itself, while Corsair provides the liquid-cooling hardware and radiator fan that keep the GP102 GPU cool. For its part, Corsair pulls a version of its H55 120-mm closed-loop cooler from its arsenal and equips it with one of its new ML120 fans. The ML-series design relies on an unusual magnetic levitation bearing to suspend its rotor, and it claims to bridge the gap between high-static-pressure and high-airflow designs.
Even though the ML120 is space-age tech, I’m let down by the fact that Corsair and MSI didn’t see fit to include an integrated fan controller that would tie the ML120’s speed to GPU temperatures. Instead, that duty is delegated to the owner’s motherboard or an external fan controller. Problem is, most motherboards made today can’t track graphics-card temperatures or tie that information to the duty cycles of their onboard fan headers, leaving Hydro GFX owners with a couple of less-than-ideal options for keeping fan noise in check.
One way would be to set a constant fan speed for the Hydro GFX’s associated fan header in a motherboard’s firmware. This method admittedly avoids rapid up-and-down changes in fan speeds, but it also might result in higher-than-necessary noise levels at idle. Another would be to manually raise fan speeds before entering a game and cranking them down again after finishing. Some gamers might have the patience to do this on the regular, but it’s so inconvenient that I can only imagine the most noise-sensitive would ever bother.
For folks like me who simply can’t live without temperature-controlled fan speeds from their graphics cards, Corsair offers a unique solution in its $70 Commander Pro. This USB fan controller can both tie fan speeds to GPU temperatures and natively control PWM fans—a combination that is, to my knowledge, impossible to find in any other product.
The problem with the Commander Pro is that it’s another $70 on top of the already eye-watering $850 (or $820 on discount, as of this writing) one will have to shell out for the Hydro GFX. Even those ready to buy a GTX 1080 Ti might take pause at that prospect, since practically every other graphics card on the market from the Radeon RX 550 to the Nvidia Titan Xp already offers a built-in fan controller. Sure, the Commander Pro also gives its owner control over RGB LED lighting strips and adds some other handy features to the bargain, but would it really have been so hard to include the unit’s fan-control logic in the Hydro GFX to begin with?
I won’t be stripping this GTX 1080 Ti completely down to its GPU in this review, because the card I have on hand is a loaner from Corsair that will be returning to the company’s review-sample pool when I’m done with it. Removing the plastic shroud from the card does let us get a glance at the Hydro GFX’s bones, though.
A smooth metal plate provides heat dissipation to the memory chips surrounding the GPU and some of the card’s power circuitry. From the seven phases visible underneath the blower fan, it’s safe to say the basic PCB layout is nearly identical to that of the GTX 1080 Ti Founders Edition’s.
Unlike many graphics cards available these days, the Hydro GFX sticks to white accent lighting for the MSI logo on the cooler shroud and the translucent rotor of the ML120 fan. Corsair has made a name for itself as the purveyor of all things RGB LED, though, so some might find the exclusively white lighting a drawback for their otherwise-colorful builds. White is one of the hardest lighting colors to coordinate using RGB LEDs, and a close-but-not-quite match might annoy picky DIYers.
The Hydro GFX offers a bevy of display outputs: one DVI-D port, one HDMI 2.0 port, and three DisplayPort version 1.4-ready outs. If you want to hook up triple displays and a VR headset all at once, the Hydro GFX is ready.
Though I’m reproducing Corsair’s nominal clock speeds for the Hydro GFX here out of completeness, the Pascal architecture’s GPU Boost 3.0 technology mostly makes specified clock speeds irrelevant. Even so, the Hydro GFX offers three clock speed modes: silent, gaming, and OC. (You’ll need MSI’s Gaming App to take advantage of these profiles.) Silent mode claims the same boost range as the GTX 1080 Ti Founders Editon. Gaming mode steps that range up 75 MHz to 1657 MHz, while OC Mode further boosts the boost range to 1683 MHz. With any luck, these figures will bear little relationship to the card’s delivered clock speeds, though.
I’ve already revealed the Hydro GFX’s $850 price, and I gotta admit: I find that number a bit hard to swallow. Nvidia’s Founders Edition GTX 1080 Ti lists for $700, and the Hydro GFX essentially takes that reference board design and mounts an affordable Corsair H55 liquid cooler to it. The liquid cooler will almost certainly make the Hydro GFX an overclocking star, but the reference-ish board design may not be a good enough supporting cast. The Founders Edition GTX 1080 Ti couldn’t sustain both a GPU and a memory overclock without running into power limits in my experience, and MSI’s Aero power-delivery subsystem isn’t much different from that card’s.
On top of the fan-control foibles I already described, it’s hard to ignore the fact that air-cooled GTX 1080 Tis with overbuilt board designs and beefy, highly effective heatsinks sell for about $100 or even less than the Hydro GFX. These cards may not be able to milk every bit of potential from a given GP102 chip, but they can probably get close for more than 10% less money. Let’s see whether the Hydro GFX can take GP102 to great enough heights to justify its price tag.
This review marks a change in the way we test Nvidia graphics cards outside of initial reviews of a new chip. While manufacturers can claim meaningful increases over the clocks of Nvidia’s Founders Edition cards, the company’s GPU Boost 3.0 algorithm mostly eliminates the differences in delivered clocks between custom cards. Our tests have shown that the resulting differences in clocks between hot-rodded GeForces are often within 50 MHz of one another, if not even closer still. Those differences of a couple dozen MHz simply aren’t meaningful in evaluating real-world performance. We already know that the GTX 1080 Ti Founders Edition delivers a high bar for both performance potential (as measured by average FPS) and smoothness (as measured with our Inside the Second frame-time methods), so anything extra is just icing on the cake.
We’ll still observe the GPU and memory speeds these cards can deliver in both stock and overclocked configurations, of course, but we aren’t going to waste precious time seeing how those tiny differences translate into a couple more average FPS here or there. Noise levels, cooling performance, and power consumption tell us a lot more about living with one of these graphics cards than raw performance numbers do, so that’s where we’re choosing to spend most of our limited testing time.
We used the following settings on each graphics card to evaluate real-world performance at stock speeds:
Our thanks to Intel, G.Skill, Gigabyte, HyperX, and Corsair for helping us outfit our testing rigs with some of the finest hardware available. Along with the GeForce GTX 1080 Ti Founders Edition, we’re also including the formidable Gigabyte Aorus GTX 1080 Ti Xtreme Edition 11G in these tests to put the Hydro GFX’s performance in perspective.
Our testing methods are generally publicly available and reproducible. If you have any questions or comments regarding our test approach, feel free to join us in the comments thread on this article or in our forums to discuss them.
To evaluate the clock speeds these cards can reach at the stock settings we chose, I fired up the 2016 reboot of Doom with all settings maxed at 4K. This title can run at high framerates while occupying large amounts of graphics memory, so it’s a great stress test as well as a great game.
With its “gaming mode” clock profile and its radiator fan fixed at 1700 RPM, the Hydro GFX likes to hang out around 1911 MHz. It’ll occasionally boost up to 1923.5 MHz, and it can sometimes dip to 1898 MHz. A clock speed bin up or down now and then is fine performance. The clock-speed range of the Aorus GTX 1080 Ti is about a bin behind by my observations. That card settled around 1898 MHz as its midpoint, and it occasionally dipped to 1885 MHz and boosted to 1911 MHz. To be clear, the real-world performance of these cards shouldn’t be noticeably different. Varying workloads and swings in ambient temperature will likely have greater effects on performance than one custom cooler or another will at stock speeds for the GTX 1080 Ti.
Nvidia’s Founders Edition GTX 1080 Ti delivers lower boost clocks than either of these custom-cooled cards, but that’s because of its strictly dual-slot cooler and 84° C thermal limit. I observed boost speeds ranging from about 1733 MHz to 1759 MHz once the card hit that limit. Even so, those speeds are well in excess of Nvidia’s rated 1582 MHz boost clock. Both the Hydro GFX and Aorus Xtreme Edition 11G should be slightly faster in real-world use than the Founders Edition, but again, we’re only talking a few frames per second more on average.
It’s good to know how these graphics cards perform at stock speeds, but we don’t imagine many builders buy hulking custom-cooled GTX 1080 Tis to run them at stock clocks only. To turn up the heat on this trio, we turned to MSI’s Afterburner app. We first maxed out the power limit and temperature limit for each card before gradually increasing the core clock and memory offsets in Afterburner. We stopped when we saw signs of instability or crashing when we ran our Doom load.
I look at these numbers in a couple of ways. On the one hand, it’s impressive that putting the GP102 GPU under water let me eke out a sustained 2050 MHz boost clock from the Hydro GFX. The GTX 1080 Ti Founders Edition tops out at 1987 MHz, and the Aorus XE 11G hits 1974 MHz. Thing is, for all its fancy cooling hardware, the Hydro GFX is only running about 3.1% to 3.5% faster than its air-cooled competition, and the differences in delivered performance between these cards is likely to be even narrower still—certainly narrow enough that one likely wouldn’t notice any difference between them in real-world use. Even so, Corsair can claim the highest overclocked speeds we’ve seen from a GTX 1080 Ti thus far, and a close victory is still a victory.
In an improvement that may be a result of more mature drivers, both the GTX 1080 Ti Founders Edition and the Hydro GFX had no trouble maintaining both their core and memory overclocks when I pushed them to the limit this time around. Seems my concerns about the power-delivery circuitry on the Hydro GFX were unfounded.
If the Hydro GFX has one ace up its sleeve compared to the air-cooled GTX 1080 Ti competition, it’s in GPU temperatures under load. At stock speeds, the Corsair card opens a whopping 17° C gap on even the massive Aorus cooler, and it completely freezes out the Founders Edition card with a 32° C advantage. Overclock the cards, and the Hydro GFX still maintains a 17° C gap on the Aorus and a 30° C gap on the Founders Edition.
While this huge temperature drop versus air coolers is the most impressive feature of the Hydro GFX, it sadly doesn’t have much real-world use for Pascal-powered graphics cards. Nvidia locked down voltage tweaking for these cards, and that restriction remains in place on the GTX 1080 Ti. If I had even a small range of additional voltage to play with on this card, I might be able to push it to truly impressive extremes. As matters stand, it’s hard to justify the Hydro GFX for its cooling performance alone. The cooler shouldn’t stand in the way of extracting the most performance from a given GP102 chip, at least. It should also allow the Hydro GFX to hang at or near its maximum boost clock for the majority of the time under load.
While it’s fun to see how far we can push the core clocks of various GTX 1080 Tis, cooling performance and noise production are where custom-cooled graphics cards really set themselves apart these days. To test noise levels, I used the Faber Acoustical SoundMeter app running on my iPhone 6S Plus. I positioned the phone 18″ from each graphics card on an open test bench.
At idle, both the Founders Edition and the Xtreme Edition 11G card are noticeably quieter than the Hydro GFX. The Founders Edition card has its variable-speed blower to thank for the victory, while the Aorus card can stop its fans entirely at idle. That’s not to say the Hydro GFX couldn’t run with these two if we hooked it up to a fan controller, but we wanted to test how loud these cards get stock-for-stock. Using Corsair’s installation instructions, the Hydro GFX isn’t loud at idle in absolute terms, but it’s definitely more noticeable than the competition thanks to its fixed fan speed.
If you do decide to hook up a Commander Pro to the Hydro GFX or manually slow the ML120 fan at idle, the rather coarse-sounding pump of the H55 all-in-one cooler will be the ultimate noise floor for the card at about 34.2 dBA on our test bench. Again, not loud, but not silent, either. The wide PWM range (500 RPM to 2600 RPM) of the ML120 does let us practically silence the 120-mm spinner at idle with the right tools, though.
Under load, the centrifugal blower on the Hydro GFX spins up alongside the ML120 fan, so the card’s total noise levels creep up just a bit compared to its idle numbers. The Aorus card remains the quietest of the bunch. The Founders Edition has to work quite a bit harder to keep its GPU cool, so it’s the loudest of these three by far.
Those standings hold when we consider overclocked noise levels, as well. The Hydro GFX spins up its blower ever so slightly more when we put the spurs in it, so it gets slightly noisier than the Aorus card. It’s worth remembering that the Corsair card is running slightly faster than the Aorus, though. Despite its midpack overclocking result, the Founders Edition starts to toe the threshold of tolerability when we take noise levels into account at just under 50 dBA.
Absolute noise levels are only one part of the picture, of course. The character of a sound can make or break a heatsink, too. Corsair’s ML120 fan deserves high praise for its noise character. At 1700 RPM, it produces a broad-spectrum and slightly high-midrange whoosh that sounds like air moving for the most part. It’s by far the best Corsair fan I’ve ever laid ears on, and it’s a fine match for the Hydro GFX’s first-place overclocking performance. The ML120 isn’t the only fan going on the Hydro GFX, though, and the card’s centrifugal blower can make a slight high-pitched hiss when it’s really turning. This hiss is barely noticeable, though.
How troublesome a noise is will always be a matter of personal preference, though, and I still have to give the edge to the Aorus card’s massive air cooler here. Its three 100-mm fans aren’t that much smaller than Corsair’s single 120-mm unit, and they produce a much lower-pitched sound that’s easier to tune out than the combination of the whoosh from the Hydro GFX’s ML120 fan and the mild hiss of its blower. Neither card qualifies as offensive to my ear, but the Hydro GFX will always make itself known at the settings we chose to test it with compared to the Aorus’ low-pitched sound.
To see how much juice these three cards are sucking down, we ran a spot-check with our Doom load and our trusty Watts Up meter. We measured total system power draw for our test rig minus its monitor, which was plugged into a separate outlet.
At idle, these three GTX 1080 Tis behave about identically, as we’d expect. Under stock load, the cards are within a few watts of each other, as well; that’s a feather in the cap for the Aorus and the Hydro GFX since they’re boosting higher than the Founders Edition card. Overclocked, the cards remain pretty closely matched. The Hydro GFX delivers its world-beating speed without making our test system draw a ton more power than the overclocked Founders Edition card does, and it lets our rig consume a bit less power than the Aorus card draws. On the whole, though, a GTX 1080 Ti is basically a GTX 1080 Ti when measured at the wall socket. This trio of cards has a basically identical impact on our test system’s power draw.
Closed-loop liquid cooling has made waves for CPUs and graphics cards alike of late, so I was excited to see what such a cooler could do atop Nvidia’s GP102 GPU. By mating one of its H55 coolers with that chip, Corsair can claim the highest overclock and lowest load temperatures we’ve seen from a GeForce GTX 1080 Ti so far, but those impressive achievements come with caveats.
For one, I wish Corsair and MSI had drawn on their combined expertise to make fan control on the Hydro GFX a self-contained affair. Running the card’s radiator fan at a fixed speed works, but it can make a system louder than it has to be at idle. Corsair offers a $70 fan controller that can tie the Hydro GFX’s radiator fan speed to GPU temperatures, to be fair, but it’s hard to justify that extra cost when most every other mainstream graphics card includes some form of fan control on board. Building that feature in doesn’t seem like too much to ask of an $850 graphics card.
Corsair and MSI’s innovative approach to cooling the Hydro GFX can’t overcome the fact that Nvidia limits voltage control on Pascal cards, either, so this pixel-pusher’s impressive thermal headroom mostly goes unused when we start turning the screws. At $150 more than Nvidia’s Founders Edition card, the chart-topping speeds I was able to pull out of the Hydro GFX come at an eye-watering cost. At stock settings, the Hydro GFX performs no differently than our custom air-cooled point of comparison, either, so this is a card you essentially must overclock in order to realize its full value.
Presuming you’re ready to put the spurs in this card from the get-go, the Hydro GFX has several good things going for it. Appropriately mounted, the remote radiator will move heat directly out of a case, and the lightweight chassis of the card won’t place lots of strain on PCIe slots or cover up more than two such slots. Corsair’s superb new ML120 fan is also one of the best-sounding I’ve ever heard at any speed, so it’s a perfect match for the Hydro GFX.
Even with those virtues, custom graphics cards really set themselves apart on noise, vibration, and harshness these days, and the Hydro GFX just isn’t $100 or so better than the other custom air-cooled GTX 1080 Tis we’ve used on those points. Unless you’re willing to take a costly shot at the silicon lottery for a chance at the absolute highest clock speeds possible out of a turn-key GTX 1080 Ti, or unless you need to move hot air directly out of your case, other, cheaper GTX 1080 Tis seem able to deliver only slightly worse overclocking potential and better noise characteristics. Still, there may be something to be said for having a shot at the best GTX 1080 Ti performance out there, and if you’re willing to play the silicon lottery, the Hydro GFX could reward the leap handsomely.
– they’re already running the pump and blower fan off the GPU’s PWM fan header [with PWM control only going to the blower, unless I Jeff mentioning hearing the pump noise ramp]. There is a finite current limit for that circuit; would it be able to handle an additional fan? Or maybe push/pull? Neither Corsair nor MSI can predict the current draw of a given fan, or fans, that a customer might install. Additionally, making the blower run at the same speed as the radiator fans is a poor solution, in that it leads to situations like drfish related earlier with his 980 Ti.
– they could add a second PWM fan circuit to the card: it’s not needed in the original design, so there’s a PCB respin. It likely won’t be used on any other offerings of this card, especially not so far after launch, so there’s little chance they’d do it for such a low volume product. Even better, where would you place the header? And how does the extra load affect the card’s power envelope?
– Even if they did add a second fan circuit, you’ve then got to spin additional firmware for the card in order to control the two fans. Again, this wouldn’t be likely to get reused in any other version of this card.
– the cabling is currently quite clean: two tubes run to the radiator, and that’s it. Add a fan cable and it begins to look messy, as well as adding to the BOM. Then consider that, as they have it, swapping fans out or adding more is very simple for an end user to do.
The only realistic way that I can see to do it better, using their off the shelf parts, is to implement a variation on the “full cover block” theme and have the Corsair CLC’s coldplate mount to a plate that covers the RAM and [more importantly] VRM in addition to the core, and run the fan cable to a 4 pin fan header mounted close to where the tubes exit the shroud.
Otherwise this style of card remains another commercialisation of a DIY hack and it should be bought and sold under the knowledge that it’ll require some DIY tinkering.
I’m not seeing a 1080 Ti with a CLC cooler on Asus’ website, would you mind sharing a link? I’m intrigued.
A quick look at the EVGA one I can find suggests that they’re using an extra fan header for controlling external fans
My bad, I was thinking of the [url=https://www.asus.com/us/Graphics-Cards/ROG-POSEIDON-GTX1080TI-P11G-GAMING/
There’s a low of power coming in on the 6+8pin connectors. If they didn’t want to create a custom PCB to handle the fan power, they could have ran another wire directly from the main power input for the fan. They could then “split” the PWM signal for the fan control. Or just put a connector on the board with just the PWM signal on it and supply an adapter to get power from a SATA power or something and the PWM signal from the card. This wouldn’t be pretty but at least it would work.
You may have missed the last sentence of my first point, and drfish’s below. The control would need to be separate for a robust solution.
The NZXT Kraken does this. It’s powered by 3-pin 12v but it has a 4-pin header and maintains speeds through its own embedded software – you don’t even have to plug in the USB header or install the CAM software. The fan is just controlled by liquid temperatures.
That sounds like a pretty straightforward implementation. Do any current Corsair branded CLCs offer similar functionality?
Thinking further, while linking radiator fan speed to water temp does seem a pretty good fit for the task, it does add another couple of degrees of disconnect and points of failure in the temperature sensor -> fan speed control loop for specific points of interest, such as VRMs or core.
I don’t know how such a system might play out in practice, but I’d love to have the time and resources to tinker!
Still amazed when i see the charts showing entire system power with a 1080 ti is well within a 450 Watt PSU (346W to 396W at PSU side). That’s enough headroom to have the CPU still do some work too.
I would be intrested to know how this stacks up against the EVGA GeForce GTX 1080 Ti SC2 HYBRID GAMING. A card that appears to at least have GPU PWM fan control. [url
Those temps seem a little high. I’m doing 44c max (torture testing CPU/ GPU simultaneously) on a heavily OC’d liquid cooled 1080 at 110% core volts and 2.15GHz. This is in a machine with fans no faster than ~5-600 RPM.
Heck, $850 is more than I spent on my 1080, water block and full custom water cooling system combined, and I’ve had 4 generations of hardware under the same rads, pump, reservoir.
I’ve been using SpeedFan to handle GPU fan speeds for years. I’ve used it for CPU fan speeds as well but most motherboards handle this well enough on their own now. It’s not the most intuitive tool but it’s very powerful. The fan speed control on my old PowerColor HD7770 was terrible out of the box. I used this to turn it off under normal conditions and set a curve for it that generally maxed out at what used to be the minimum and it still ran at reasonable temperatures. Of course it’s dead now. So maybe the VRMs or something needed more cooling than I gave it. But the point is that there are ways to do things like this.
But I agree that they did a terrible job of integrating the cooler. Other then the pretty packaging, they basically just slapped a liquid cooler onto the card without any thought for how they would integrate. Not only is the fan on the radiator not controlled by the card but the fan on the card seems to act like it’s actually cooling the whole card and likely spinning faster than it needs to. Go back to the drawing board guys. And next time, bring some engineers in there instead of letting the sales team design it.
The blower seems to mainly force cool air out of the case, I doubt it will do much for cooling the memory. And it blows a bit along the heatsink, for the VRM’s I suppose.
From what I understand, voltage is bios locked in the 10 series, and the bios is encrypted. You *can* replace the bios with another model’s with a slightly higher voltage, but not much else.
You have to edit the DEVICE_ID entry in the ini file for afterburner to match your particular make of card (a small selection of cards are already defined when you download the program), but you can also use wildcard entries ‘??????” etc and afterburner will unlock voltage controls with any vendor.
Gotya, I just put the “OC mode” bios on my 1070, and my 5x 1080TIs I just pushed the powerlimit to like 105-110%, depending on card location, any higher and I get in the 70c range, too hot for my taste. I was told turning up the voltage in afterburner will only move it to it’s max voltage in bios but not exceed it (say the card ‘s bios is 1.09v max while another may be 1.07v), to move up you would need to change bios, and then the card has a 1.25v limit that any bios cannot exceed (maybe voltmoding the chip can).
I was just asking whether TR had tried any of the low hanging voltage boost fruit when OC’ing the Hydro.
EDIT: I should point out that I currently have +0mV extra voltage on my 1080Ti, because the card is more than fast enough to feed my 1440p display at 120 fps+ without any excessive OC, so I have not explored the limits of overvolting in afterburner.
Forcing users to manually control the fan speed is a huge oversight. It’s also something I had just assumed that these CLC-equipped graphics cards did. So I learned something.
I bought MSI’s liquid cooled GTX 1070 with the exact same H55 cooler, shroud and fan to mine with. In my experience mining really cuts down on the life of a graphics card.
The GTX 470 that got me my first Bitcoin now has memory issues when playing games, the Radeon HD 6870 gets all its shares rejected, a Radeon HD 7950’s VRM burnt out, and a one of the fans on a Radeon HD 7970 GHz suddenly jammed itself.
I spent the 100 premium on the cooler gladly, if it means I don’t have to worry about my card conking out in the near future.
For comparison, here’s what [url=https://techreport.com/review/31562/nvidia-geforce-gtx-1080-ti-graphics-card-reviewed/15
Yeah, that’s either incredibly lazy or incredibly sleazy on Corsair’s part. All they need to do is run a PWM wire from the GPU’s fan header. This (overpriced) $8 adapter, for example, will do it:
The thing is, it’s not just them. My EVGA 980 Ti Hybrid has a power header for the radiator fan. However, and I haven’t exhaustively researched this (because I worked around it), it appears that the speed of both fans is only influenced by the core temp and they always run at the same percentage speed. Since the rad is very good at keeping the GPU cool, the blower on the memory/VRMs barely moves any air at all and I’ve found that to cause artifacting (especially in [i
Agriculture Sprayer Parts High Quality Supplier
That’s assuming the header is actually there. Jeff had good reason to not totally take the card apart, but it could be that MSI left the header off entirely. It’d have to be under that block somewhere.
“Even though the ML120 is space-age tech, I’m let down by the fact that Corsair and MSI didn’t see fit to include an integrated fan controller that would tie the ML120’s speed to GPU temperatures. ”
With gaming GPUs drawing two to three times as much power as gaming CPUs, it makes more sense to put water cooling there. Not using the built-in fan speed controller on the graphics card does not make sense. I’ve adapted case fans to run off of the graphics card’s fan controller in the past.
Stainless Steel Castings, Gray Iron Casting, Ductile Iron Casting - Yogem,https://www.yogemcasting.com/