180mm of rads enough for cpu/gpu?

nascentparadigm

Caliper Novice
Original poster
Jan 4, 2019
25
22
One of my dreams with the A4 has been to do a custom loop for both the GPU and CPU. While air-cooling both components is most likely the better solution (and has been working well for me for the last six years), I've just been fascinated by the engineering challenge of getting a custom loop to fit completely within the confines of the case.

I tried and failed to build one in 2018 (EKWB pwm pump was just way too loud, and couldn't find an AC/DC adapter for the HDPLEX-400 that didn't make a horrible electrical whine), but looks like there are lots of new products have come into the market that might finally make this possible, such as the Modultra LOBO, which saves a lot of space by combining multiple components into one.

My main hangup before going onto the purchasing state is, how much TDP can I afford? The current plan is to fit two Alphacool 92mm ST25 radiators with Noctua NF-A9x14 fans, one behind the PSU using the 120mm AIO bracket and 120mm>92mm fan adapter, and the other underneath the PSU like lots of folks have done with the Asetek. Combined this gives me about 180mm of total radiator surface area.

Currently I have a 3700x and a RTX 2070, which together is about 240tdp. Would love to move up to a 5800x3d (so I don't have to get a new mobo and all that entails) and a 4070ti, but that pushes the envelope to ~390tdp. Been looking through some threads and the general recommendation has been to have a 65tdp CPU but I haven't really seen any kind of numbers regarding GPUs so I'm not sure what would be pushing the limit there.

Just curious what other folks experiences are with pushing components in their systems, if I should keep what I have and get waterblocks for them (although the only block for the 2070 seems to be from Bykski), or do some more modest upgrades, say to a 5700x (or not, it's like exactly the same as the 3700x) and 4060ti (which is not super well reviewed...) or 4070.

Thanks in advance!
 

nascentparadigm

Caliper Novice
Original poster
Jan 4, 2019
25
22
Playing around in Blender some more - with the very generous stp files people have authored online - if I really want this to be as dense as possible, I think I could actually get a 120mm rad behind the GPU, and a 92mm underneath it. So that'd bring total rad dimensions to 210 (a small but perhaps not insignificant improvement?)
 

Skripka

Cat-Dog Perch Manager
May 18, 2020
443
544
Traditionally, the recommendation has been a 120mm radiator for each high-draw part in the loop. By unit area a 180mm might do you, but you will be running somewhat high air-water delta-T under load.
 

nascentparadigm

Caliper Novice
Original poster
Jan 4, 2019
25
22
I did some real bad back of the napkin math, looking at this setup for comparison - Based on the sampled noise profile in the video I think that would be my target, and for sake of comparison I came up with this horrible equation (assuming the same fans are used in both setups):

TotalTDP / (RadLength * RadThickness * FPI) == efficiency

Optimum's would be (with a 4090 + 7800x3d)
570 / (480 * 20 * 22) = 0.002698863636

Mine would be (4070 + 3700x):
265 / ((92 * 25 * 15) + (120 + 20 + 22)) = 0.003035509737

I think that puts it relatively in the same order of magnitude at least, so hopefully it should be able to handle that load.

If I upgrade to the 4070 ti however that number becomes 0.004009163803, meaning my system would dissipate heat at about 2/3 the capacity of Optimum's.

...again, this is napkin math so I have no idea if it is valid whatsoever.
 

nascentparadigm

Caliper Novice
Original poster
Jan 4, 2019
25
22
Actually my bad math was worse than I thought, I only accounted for the length + depth of the cards, but excluded the width.
So the corrected equation is:
TDP / (Rad(L * W * D) * FPI) = efficiency

Assuming the optimumpc ratio is the goal (ie grading on a curve so its value is 1), here's the following relative scores of various combinations (first column is gpu/cpu combo followed by TPD, columns 2-3-4 are the relative scores using the above equation):
RTX 4060 | 3700x9x99x1212x12
1800.7931.1881.583
RTX 4060ti | 3700x9x99x1212x12
2300.6210.931.239
RTX 2070 | 3700x9x99x1212x12
2400.5950.8911.188
RTX 4070 | 3700x9x99x1212x12
2650.5390.8071.075
RTX 4070 Super | 3700x9x99x1212x12
2850.5010.751
RTX 4070 ti | 3700x9x99x1212x12
3500.4080.6110.814
RTX 4080 | 3700x9x99x1212x12
3850.3710.5560.74
RTX 4060 | 5800x3d9x99x1212x12
2200.6490.9721.295
RTX 4060ti | 5800x3d9x99x1212x12
2700.5290.7921.056
RTX 2070 | 5800x3d9x99x1212x12
2800.510.7641.018
RTX 4070 | 5800x3d9x99x1212x12
3050.4680.7010.934
RTX 4070 Super | 5800x3d9x99x1212x12
3250.4390.6580.877
RTX 4070 ti | 5800x3d9x99x1212x12
3900.3660.5480.731
RTX 4080 | 5800x3d9x99x1212x12
4250.3360.5030.671

You'll notice there's a 12x12 column, 'cause I might have figured out how to fit two 120mm radiators in the case - put one fan/rad combo on each side using the AIO bracket, and then sandwich two HDPLEX-250 GaN supplies in between them. Not sure I feel comfortable with the HDPLEX though, it didn't work out for me so well before (but that had more to do with the AC brick than the PSU itself).
 

Frenzy

Trash Compacter
Feb 14, 2020
47
18
Currently I have a 3700x and a RTX 2070, which together is about 240tdp. Would love to move up to a 5800x3d (so I don't have to get a new mobo and all that entails) and a 4070ti, but that pushes the envelope to ~390tdp. Been looking through some threads and the general recommendation has been to have a 65tdp CPU but I haven't really seen any kind of numbers regarding GPUs so I'm not sure what would be pushing the limit there.

Just curious what other folks experiences are with pushing components in their systems, if I should keep what I have and get waterblocks for them (although the only block for the 2070 seems to be from Bykski), or do some more modest upgrades, say to a 5700x (or not, it's like exactly the same as the 3700x) and 4060ti (which is not super well reviewed...) or 4070.

Just to chime in a bit about my own train of thought i had about this matter with my DAN A4 with 3700x and 1080. Although i hadnt quite considered using the A4 for a loop, was looking more at Formd T1 etc.

A custom loop is relative expensive, and it would be a bit of waste to spend it on old hardware i.e. the 3700x and 2070. Considering the waterblocks are not transferrable to a newer system later on. Given the thermal constraints better take more modern but lower spec hardware which is more efficient. The current 4070Super looks quite decent power vs. performance.

Also just taking the 65W TDP of the 3700x of 65W is not that accurate. Depending on your workload at full tilts its more around the 85W (say rendering) and during gaming depending on the GPU pairing (considering 2070 and 1080 perform similar'ish), the 3700x is running ~30% load (and GPU and 100%).
Ow and heavy GPU undervolting is also possible ofcourse to keep the Watts low.

I'm looking to upgrade to 4070ti super , however slightly concerned the 3700x can't keep up and potentially slot in a 5800x3d. But ill stick to air cooling as i'm also planning to take the A4 on the plane from time to time. Undervolt the 4070ti super and replace stock fans and shroud wtih 90x90 noctua fans
 
Last edited:

nascentparadigm

Caliper Novice
Original poster
Jan 4, 2019
25
22
Ok, following up on this, I have two different designs I'm considering:
  • 2x Alphacool ST20 120mm radiator with 15mm fans >
  • 1x Alphacool XT45 120mm radiator with a 25mm fan >
Which one do you guys think would be more performant?
 

SFFMunkee

King of Cable Management
Jul 7, 2021
687
701
Ok, following up on this, I have two different designs I'm considering:
  • 2x Alphacool ST20 120mm radiator with 15mm fans >
  • 1x Alphacool XT45 120mm radiator with a 25mm fan >
Which one do you guys think would be more performant?
Possibly too niche to even provide speculative suggestions on performance. Don't forget to account for actually building these designs too.
I've fallen into the trap of making an intricate and extremely tight design only to find I couldn't actually build it because I hadn't accounted for bend radii of tubes/cables and access to things like the CMOS reset etc.