3rd Party GPU AIR Coolers

[OliverQueen]

Not too impressed by the standard 3 fan solution on my "cheap" PNY RTX3080ti card (was only one I could get at time of purchasing last year anywhere near MSRP). I don't feel just changing the thermal compound on the core would help that much.

Looking around all I can find that may be compatible is the Rjaintek (sp?) but now noticed the OCUK do not list the RTX30xx version anymore.

Anybody got any experience in these types of 3rd party air coolers & how well they work over stock or just stick to what is on the card, replace the TIM/Pads with better than stock & put up with a stable small non-factory OC (power & voltage are locked at 100% so cannot even add an offset to them in Afterburner) until build new machine with 40xx or whatever is current at the time?

 

[D1craig]

I had one of these coolers years back. They are pretty big buy are really good at cooling. I couldn't get the vrm or vram heatsinks to stick though, they kept falling off.

They reduced my temps by atleast 15c. I think mine was prolimatech mk26. You had to stick 2 120mm fans to it so it needed a lot of room...

 

[Cyber-Mav]

Seems to be a big lack of aftermarket air coolers for gpus, all i see these days is water blocks.
Arctic cooling used to do a bunch of them but for some reason stopped turing cards were the last supported ones for the accelero xtreme.

Maybe its to do with aib cards having huge 3 fan cooler options that ate into the market? Cheaper to buy higher end card with better cooler vs buy cheap card and slap on an aftermarket cooler?
Also noticed these days stock cooler integrated vrm and vram contact plates so thats another issue for aftermarket coolers to deal with and my experience of those stick on heatsinks supplied by arctic were very bad.

 

[bimbleuk]

Not too impressed by the standard 3 fan solution on my "cheap" PNY RTX3080ti card (was only one I could get at time of purchasing last year anywhere near MSRP). I don't feel just changing the thermal compound on the core would help that much.

Looking around all I can find that may be compatible is the Rjaintek (sp?) but now noticed the OCUK do not list the RTX30xx version anymore.

Anybody got any experience in these types of 3rd party air coolers & how well they work over stock or just stick to what is on the card, replace the TIM/Pads with better than stock & put up with a stable small non-factory OC (power & voltage are locked at 100% so cannot even add an offset to them in Afterburner) until build new machine with 40xx or whatever is current at the time?

As you are using Afterburner then applying an OC will only worsen the situation. You want the temps down in order to allow higher clocks. You can apply a small OC while also using the Curve undervolt tool. Try the below to see if you can get the power usage down which should then allow the GPU to boost higher.

Open the curve tool (small graph icon left of Core Clock)
Shift + L Click (hold) on the 0.975 graph square and drag the whole curve up to the normal boost clock (~2700 for my RTX 4080)
Shift + L Click (hold) just above the square and drag to the right blue highlighting everything to the right.
Shift + Enter (twice) to flatten everything to the right targeting that mV for all higher clocks (this doesn't always work for some reason, try again)
Back on the MSI AF interface click the Enter/Check button to apply your curve.

Then while running Valley 1.0 or Superposition monitor the power usage before and after. My RTX 4080 tends to drop anything from 30-50W when under load. Temps go down, fans stay lower and frequency boosts higher.

Spoiler: RTX 4080 UNDERVOLT CURVE

 

[OliverQueen]

As you are using Afterburner then applying an OC will only worsen the situation. You want the temps down in order to allow higher clocks. You can apply a small OC while also using the Curve undervolt tool. Try the below to see if you can get the power usage down which should then allow the GPU to boost higher.

Open the curve tool (small graph icon left of Core Clock)
Shift + L Click (hold) on the 0.975 graph square and drag the whole curve up to the normal boost clock (~2700 for my RTX 4080)
Shift + L Click (hold) just above the square and drag to the right blue highlighting everything to the right.
Shift + Enter (twice) to flatten everything to the right targeting that mV for all higher clocks (this doesn't always work for some reason, try again)
Back on the MSI AF interface click the Enter/Check button to apply your curve.

Then while running Valley 1.0 or Superposition monitor the power usage before and after. My RTX 4080 tends to drop anything from 30-50W when under load. Temps go down, fans stay lower and frequency boosts higher.

Spoiler: RTX 4080 UNDERVOLT CURVE

With the OC I use in Afterburner, the temps don't change from stock (hence why only a small one). I also have the temp limit set to 83C as at 84C it is unstable with artifacts/tearing (V-Sync makes no difference) & drop-outs before it can throttle back the clocks.

At idle, the fans are set to 31% (auto) & even dropping the speed by 5-10%, there is a significant increase at idle of 8-12C (which I don't feel is right but that's what it does!).

Even at stock - under load - I have to ramp the fan speed to 85% for it to be under control temperature wise (hence why I know the stock triple fan cooling solution is barely the minimum required for the thermal output) even with good airflow through the case (have tried several fan combinations for negative, positive & neutral pressures as well as fan sizes, etc). I did find that with the NH-D15 on the CPU, the temps dropped by 1-2C on the GPU compared to the AIO it replaced which I put down to the increased air flow from the 2 fans on it pulling hot air away a little quicker than before. With the on/off 'heatwave' we had over the last month, I had to leave the side panels off & have a small tower fan blowing onto the GPU to keep it under control (not removed the cooler to replace the TIM/pads yet so know that not messed something up there YET!)

With the PNY card the voltage adjustment box is completely blanked out & cannot even apply an manual (in box typed) undervolt (have tried), but I will have a play with the curve settings when I get a spare few hours to get stuck in, really put mind to it & see what can come up with results-wise.

Even if I get same performance with lower temps from a 3rd party cooler, I would be happy with that as it performs fine with the 9900K even with the Gen3 bottleneck on the applications/games I use.

Cyber-Mav , that's is what I noticed as well compared to the past. Maybe the lack of a real reference design from Nvidia since they started doing the FE cards so each AIB card could have components in slightly different places on the PCB compared to another. Same applies to aftermarket GPU AIO's (which I would also consider). If I was buying a new GPU now, I wouldn't go with the cheapest option unless was in the same predicament as I was at time of buying the 3080ti, which was that it was the only card I could find in stock anywhere that was close to MSRP when the prices started to rapidly drop in Spring 2022. If I could have got an Asus Strix or EVGA FTW3 card at a reasonable price difference, I would have gone with one of those instead. The other reason for wanting a different cooler is that got no control at all over the lighting on the card as it is not addressable or able to be turned off without physically unplugging from the board (I dislike RGB tbh but having no control over the factory set rainbow run it does should be expected I suppose from a basic "cheap" card makes it infinitely worse).


D1Craig, that is a good temperature drop & something that I would be very happy getting from an aftermarket solution. I would fit it out with Noctua fans to match up with the rest in the system so definitely do away with any RGB!

 

[varkanoid]

The NZXT G12 was brilliant with an AIO Water cooler I guess the market just isnt there anymore. My GTX 580 way back when I tie wrapped a AIO cooler to it and added VRM heatsinks it was brill knocked off 20c temps.

The Arctic Cooling ones were also often better than the fitted one. It all seems to have died off after the RTX 2xxx series which IIRC was the last compatible series with these aftermarket coolers.

You could always go back to the tie wrap method and fly with the seat of your pants.

 

[bimbleuk]

With the PNY card the voltage adjustment box is completely blanked out & cannot even apply an manual (in box typed) undervolt (have tried), but I will have a play with the curve settings when I get a spare few hours to get stuck in, really put mind to it & see what can come up with results-wise.

This doesn't need the voltage control to work though you can enable the Voltage Monitoring from the properties so that you can see in the Afterburner interface that your upper mV limit is applying.

I've played with several nVidia GPUs from 1660 to 4080 recently and they all tend to run in the 1000 - 1075 mV range so dropping that by ~100 mV can have a noticeable affect on power and temps. If you run a load on the GPU and see what the typical boost clock is on you card just use that as your frequency target when lifting the graph at .950 or .975 mV which would be a good place to start. If you want that ultra low power usage for some reason but with a lower boost clock you can go lower but I've never needed to do that.

My RTX 4080 OC comes from lifting the whole graph to ~2800 MHz rather than 2700 MHz. That wasn't my aim but it's a small change that seems to run fine so I left it. You can also then add say +500 MHz in the memory box after you've tested the undervolt stability.

 

[D1craig]

To my nooby ears it sounds like a TIM replacement could be a good idea.

 

[OliverQueen]

This doesn't need the voltage control to work though you can enable the Voltage Monitoring from the properties so that you can see in the Afterburner interface that your upper mV limit is applying.

I've played with several nVidia GPUs from 1660 to 4080 recently and they all tend to run in the 1000 - 1075 mV range so dropping that by ~100 mV can have a noticeable affect on power and temps. If you run a load on the GPU and see what the typical boost clock is on you card just use that as your frequency target when lifting the graph at .950 or .975 mV which would be a good place to start. If you want that ultra low power usage for some reason but with a lower boost clock you can go lower but I've never needed to do that.

My RTX 4080 OC comes from lifting the whole graph to ~2800 MHz rather than 2700 MHz. That wasn't my aim but it's a small change that seems to run fine so I left it. You can also then add say +500 MHz in the memory box after you've tested the undervolt stability.

The undervolting didn't drop the temps consistently but did make the card seem to run with some stutter & also lower framerates (even with just -50mv offset applied across the whole graph & the card seemed to take longer to reach the boost clock), but I managed to get the card running stable at stock voltage (& with just 1C temp increase) with +175 on the core/+450 on VRAM (+200 core/+500 VRAM wasn't stable & Diablo 4 would crash as soon as entered game with those settings even though Furmark, 3DMark & Kombuster would run stable for as long as I let them run). Might be me not fully understanding the curve settings but I essentially placed a -50mv across the curve & flattened it at 2005Mhz/1050mv (which is the maximum stable boost I see when under load & stock voltage at 2005Mhz is 1100mv - 1250mv). When I get more time I will run some stress tests & decrease the offset by 1mv every 5-10 minutes until start seeing either crashes, artifacts or/and framerate drops.

With the higher manual overclock I did see a significant increase in FPS in Diablo 4 (jumped from ~140fps to ~205fps at 5120x1440p) with no changes to the in-game graphical settings.

I will replace the TIM with Kryonaut (if I have any left) or Noctua H2 (which I know I have a full syringe of) but will have to reuse the thermal pads on the memory & VRM's as know I have none at all (need to order some but not at top of list presently), once the parts I am waiting for from OCUK arrive to finish the case swap (which hoping will help with lowering temps overall as is, with the increased airflow from the Lancool II Mesh Performance (replaced stock fans with Noctua Chromax 120mm & Redux 140mm fans) over the Corsair Crystal 570X (running LL120 & LL140's) it is currently residing in).