1 hour of cyberpunk 2077, and I have a Delta T of 9.5 degrees (room temp 22, water temp 31.5). Water temp was very stable. GPU was really low temp of 46, stable throughout (just moving up or down 1 or 2 degrees). CPU (i9 10900K) was also stable mid 50s (55-57 largely). So this setup works cool and silent. Oh, and that was with a mild overclock on the GPU.
Well next was some testing today to see what sort of fan and pump levels were required.
I did a 15 minute test, with my CPU slightly undervolted, and GPU at stock. I used CPU-Z to max out the CPU, and then I ran MSI Kombustor at 2560x1440, letting both run for 15 minutes with HWinfo open on sensors.
I will report the Delta T below, which was the max difference between the ambient room temperature, and the max water temperature. In each test I had the starting water temp between 23.0 and 23.3 degrees celsius, and then picked out the max water temp of the test, subtract the ambient room, and that is the Delta T I am reporting here (the room temp varied by 1 degree across tests).
I will also note the max GPU temp and max CPU temp (package), but note that the VRM temps hit 100 degrees or 101 degrees in every one of these 4 tests, and yes, I do have the VRM heatsink, but just not a lot of internal air-flow as I am temporarily down to 2 case fans.
Pump at 50%, Fans at 50%
Delta T = 11 degrees celsius
Max GPU = 49.8
Max CPU = 72
Pump at 50%, Fans at 75%
Delta T = 8.1 degrees celsius
Max GPU = 47.6
Max CPU = 69
Pump at 50%, Fans at 100%
Delta T = 7.2 degrees celsius
Max GPU = 47
Max CPU = 69
Pump at 75%, Fans back down to 50%
Delta T = 9.7 degrees celsius
Max GPU = 48.9
Max CPU = 71
So unsuprisingly, having 9 fans on the radiator changing their power makes a lot more difference than changing the flow rate by increasing the pump power.
The sweet spot so far is the pump at 50% with the fans at 75% for gaming, but I will set up a curve to control this so that when I am surfing the web the fans can be off, and the radiator can work in passive mode, then ramp up for gaming. Clearly a lot of cooling power here, but the VRM temps are the problem and I think these were responsible for the core clock reducing from 5.0 to 4.9 and 4.8 at about the 12 minute mark in most tests, I may add a second top fan in intake mode (I have the space drilled already, from my earlier work), and see if that helps.
Sorry, but an active backplate isn't going to do much. Not sure there is much more to try, but then I thought that last time before I came up with this idea, so let's see.
@Hewligan , you can add a VRM/MOSFET block on your water loop to keep those temps more under control. Or, I wonder if a couple of Peltier units added to the VRM heatsink would do the trick (I know they're a bad idea for CPU cooling, but might be enough to keep the VRM under control).
Nothing would beat finding an old air conditioner and turning it into a phase change cooler, like Linus did in that video I posted on the previous page though
Peltier units are a bad idea because they add heat and need a heatsink as well as insulation so that water doesnt condense on them. These devices must be used in conjunction with a heat sink to avoid a fire.
It was me who asked for the back picture, because I wanted to see the capacitor layout used.
Your picture shows 5 PS-Caps and 1 MLCC cap.
That would make this is a mixed design. At least it's not a full PS-Cap design.
Wish there was more MLCC caps used. The PC-Caps are the largest point of failure when using clock frequencies above 2 Ghz on the GPU. This has been confirmed by EVGA, who had to delay their launch because the al PS-Cap design could not pass their own internal validation.
Hewligan
1 Rookie
•
89 Posts
1
May 2nd, 2021 08:00
1 hour of cyberpunk 2077, and I have a Delta T of 9.5 degrees (room temp 22, water temp 31.5). Water temp was very stable. GPU was really low temp of 46, stable throughout (just moving up or down 1 or 2 degrees). CPU (i9 10900K) was also stable mid 50s (55-57 largely). So this setup works cool and silent. Oh, and that was with a mild overclock on the GPU.
Vanadiel
6 Professor
•
7.1K Posts
1
May 2nd, 2021 12:00
You could try as push-pull configuration to get even better temps.
That's an eye catching setup, for sure.
Can't wait to see what you come up with in a few weeks. I am hoping it is one of those desks with a computer inside.
bertro514
1 Rookie
•
363 Posts
0
May 2nd, 2021 17:00
@Hewligan that is VERY impressive!! I second @Vanadiel's suggestion for your next cooling mod. Or, go with a refrigerated setup: https://www.youtube.com/watch?v=nwCtvpgwm5o
Vanadiel
6 Professor
•
7.1K Posts
0
May 3rd, 2021 04:00
They should have one of those voting options, where the community could vote on what's next. It would be interesting!
Hewligan
1 Rookie
•
89 Posts
3
May 3rd, 2021 07:00
Well next was some testing today to see what sort of fan and pump levels were required.
I did a 15 minute test, with my CPU slightly undervolted, and GPU at stock. I used CPU-Z to max out the CPU, and then I ran MSI Kombustor at 2560x1440, letting both run for 15 minutes with HWinfo open on sensors.
I will report the Delta T below, which was the max difference between the ambient room temperature, and the max water temperature. In each test I had the starting water temp between 23.0 and 23.3 degrees celsius, and then picked out the max water temp of the test, subtract the ambient room, and that is the Delta T I am reporting here (the room temp varied by 1 degree across tests).
I will also note the max GPU temp and max CPU temp (package), but note that the VRM temps hit 100 degrees or 101 degrees in every one of these 4 tests, and yes, I do have the VRM heatsink, but just not a lot of internal air-flow as I am temporarily down to 2 case fans.
So unsuprisingly, having 9 fans on the radiator changing their power makes a lot more difference than changing the flow rate by increasing the pump power.
The sweet spot so far is the pump at 50% with the fans at 75% for gaming, but I will set up a curve to control this so that when I am surfing the web the fans can be off, and the radiator can work in passive mode, then ramp up for gaming. Clearly a lot of cooling power here, but the VRM temps are the problem and I think these were responsible for the core clock reducing from 5.0 to 4.9 and 4.8 at about the 12 minute mark in most tests, I may add a second top fan in intake mode (I have the space drilled already, from my earlier work), and see if that helps.
Vanadiel
6 Professor
•
7.1K Posts
0
May 3rd, 2021 08:00
This is a liquid cooled GPU, right?
If it is an RTX 3090 could could try an active backplate. It's something new they recently made, Linus tech tips has a video about them.
For an RTX 3080 the memory is at the same side as the GPU, so an active backplate would make no difference.
Phase change cooling would be your next option to reduce heat even more...
Hewligan
1 Rookie
•
89 Posts
2
May 3rd, 2021 10:00
@Vanadiel I have the RTX 3080, so just a liquid cooled front, with a thick passive backplate. I have the https://www.aquatuning.co.uk/water-cooling/gpu-water-blocks/gpu-full-cover/fullsize/27822/alphacool-eisblock-aurora-acryl-gpx-n-rtx-3090/3080-with-backplate-reference?c=22412 (AlphaCool Aurora Eisblock) and for the CPU I have an https://www.aquatuning.co.uk/advent-special/4.-advent-special/21507/alphacool-eisblock-xpx-cpu-deep-black (AlphaCool XPX). The reservoir is in the top HDD bay and is an https://www.ekwb.com/shop/ek-quantum-volume-flt-120-d-rgb-plexi (EK FLT 120).
Sorry, but an active backplate isn't going to do much. Not sure there is much more to try, but then I thought that last time before I came up with this idea, so let's see.
bertro514
1 Rookie
•
363 Posts
0
May 3rd, 2021 14:00
@Hewligan , you can add a VRM/MOSFET block on your water loop to keep those temps more under control. Or, I wonder if a couple of Peltier units added to the VRM heatsink would do the trick (I know they're a bad idea for CPU cooling, but might be enough to keep the VRM under control).
Nothing would beat finding an old air conditioner and turning it into a phase change cooler, like Linus did in that video I posted on the previous page though
speedstep
9 Legend
•
47K Posts
1
May 5th, 2021 09:00
Peltier units are a bad idea because they add heat and need a heatsink as well as insulation so that water doesnt condense on them. These devices must be used in conjunction with a heat sink to avoid a fire.
https://usa.banggood.com/DIY-Kits-Thermoelectric-Peltier-Refrigeration-Cooling-System-Water-Cooling+-Fan+-2pcs-TEC1-12706-Coolers-p-1608070.html
https://usa.banggood.com/Enhanced-Version-12V-12A-144W-DIY-Double-Head-Semiconductor-Refrigerator-Radiator-Cooling-Equipment-p-1174934.html
Hewligan
1 Rookie
•
89 Posts
1
May 8th, 2021 11:00
Someone asked a while back for a photo of the rear of the RTX 3080. Well, I took it apart today to clean, and here it is:
I also moved to an external pump / res as I will be moving to a new home office in a week and want to have my rad and pump on show.
This was it mid setup, before I tidied it up and placed it all level again.
Vanadiel
6 Professor
•
7.1K Posts
1
May 8th, 2021 14:00
It was me who asked for the back picture, because I wanted to see the capacitor layout used.
Your picture shows 5 PS-Caps and 1 MLCC cap.
That would make this is a mixed design. At least it's not a full PS-Cap design.
Wish there was more MLCC caps used. The PC-Caps are the largest point of failure when using clock frequencies above 2 Ghz on the GPU. This has been confirmed by EVGA, who had to delay their launch because the al PS-Cap design could not pass their own internal validation.
They went with a mixed design of 4 PS + 2 MLCC.
Maddogcrypoze
1 Message
0
September 8th, 2021 19:00
Im really curious about these mods but only one of the pictures is showing up