Water cooling is dead....except it is REALLY not!

[Smoked Brisket]

Every time this discussion comes up it is always about cpu's. Why is there never any conversation about gpu's and the benefits of custom loops. I know, personally, that the loudest, hottest, part in my builds was/is the gpu. Can any water cooler here tell me they want to go back to ramped up fan speeds and times when they were afraid to max out every setting on msi afterburner? Even under badly planned loops you can push gpu's past manufacturer settings without fear, and its silent compared to the obnoxious air solutions out there. Yes there are hybrid cards, but they are not even close to as effective and still cost as much. Drunken rant over.

 

[Supercharged_Z06]

My main rig has gone through numerous hardware upgrades over the past 15+ years... everything has been replaced multiple times, but the one thing that has remained constant no matter what gets ripped out and upgraded is that a custom loop of some sort to cool both the CPU and GPU has always been involved. I like to overclock both my CPU/GPU rather [H]ard... air cooling simply would never work for me.

 

[1337Goat]

Water cooling also requires fans.
You can't escape fans so easily.

 

[Smoked Brisket]

Water cooling also requires fans.
You can't escape fans so easily.

The loudest fans are gpu fans ramping up. No contest. I can dig up hundreds of forum posts claiming that the first GTX titan ran hot. I can dig up hundreds of any first generation card `forum posts that ran hot. Please dig up one single piece of evidence that a gpu card under water ran hot no matter how shitty the loop was and then you can talk shit.
Forgive my language.

 

[1337Goat]

The loudest fans are gpu fans ramping up. No contest.

I guess you're right. I've been using an R9 Fury X with the built in watercooler. =p

 

[Smoked Brisket]

I guess you're right. I've been using an R9 Fury X with the built in watercooler. =p

Can you link that? It is a classic and we would all do well to see a real picture of it! for example here https://www.pcworld.com/article/293...ard-review-amds-thoughtful-4k-powerhouse.html
when it was supposed to be the first 4k graphics card. That was a bad attempt at an AIO graphics card. Sorry but it was. I dont want to bust Aegir's balls but it was.

 

[1337Goat]

Can you link that? It is a classic and we would all do well to see a real picture of it!

The R9 Fury X is a serious beast. It was money well spent.
And the sheer collector's value is immense. It's some seriously sleek engineering.

 

[Smoked Brisket]

I guess you're right. I've been using an R9 Fury X with the built in watercooler. =p

The R9 Fury X is a serious beast. It was money well spent.
And the sheer collector's value is immense. It's some seriously sleek engineering.

Ok brother, you are correct in the sense that you have the single graphics card that is silent, powerful. and removes the need for water cooling. Looks like I was wrong all along.

 

[Zarathustra[H]]

I don't know who is suggesting it is dead.

There are more parts available, and seemingly more people doing custom loops than ever.

Granted, most of them seem to be youtubers doing it for looks, but still.

It's not as fun as it used to be - IMHO - as there is less DIY inventiveness, but still I see a lot of builds out there.

As far as usefulness goes, I mean CPU overclocking is on its last legs. Through a combination of smaller process nodes, and binning closer to the actual capability of the silicon, AMD CPU's barely overclock at all anymore, and Intel is probably not far behind them, so water cooling is not a game of hitting max CPU clock anymore.

I agree, water cooling has a huge impact on GPU temps. AIO's do better than air coolers, but a good custom loop smokes them both, both in terms of noise levels and temps.

To get a balance in fan noise and GPU temp, I usually do a load test on a new loop, determine how much hotter than the loop temp the GPU core runs, and then set the target on the loop temp such that the GPU never hits 40C. This is where I find I get the best boost clocks. I start loosing boost over 40C. So I usually wind up with a loop target temp of ~32 to 33C. This seems produce temps that are find for the CPU

If I really want to (for some crazy reason) and max all of my fans, I can have my GPU, a Pascal Titan at max overclock running with max load temps of 24C It's loud though and there is no point to it.

So, I guess custom water cooling for me is a matter of getting the most out of my GPU, while also keeping everything nice and quiet at load.

 

[VanGoghComplex]

If I could only cool one, I would choose GPU over CPU any day, hands down. Seeing those nice low temps on the CPU tickles my jimmies, but it doesn't translate to much in terms of performance. GPUs, however... a 1080ti at 35°c is practically a different card from an identical one at 70°c.

 

[Ready4Dis]

I don't know who is suggesting it is dead.

There are more parts available, and seemingly more people doing custom loops than ever.

Granted, most of them seem to be youtubers doing it for looks, but still.

It's not as fun as it used to be - IMHO - as there is less DIY inventiveness, but still I see a lot of builds out there.

As far as usefulness goes, I mean CPU overclocking is on its last legs. Through a combination of smaller process nodes, and binning closer to the actual capability of the silicon, AMD CPU's barely overclock at all anymore, and Intel is probably not far behind them, so water cooling is not a game of hitting max CPU clock anymore.

I agree, water cooling has a huge impact on GPU temps. AIO's do better than air coolers, but a good custom loop smokes them both, both in terms of noise levels and temps.

To get a balance in fan noise and GPU temp, I usually do a load test on a new loop, determine how much hotter than the loop temp the GPU core runs, and then set the target on the loop temp such that the GPU never hits 40C. This is where I find I get the best boost clocks. I start loosing boost over 40C. So I usually wind up with a loop target temp of ~32 to 33C. This seems produce temps that are find for the CPU

If I really want to (for some crazy reason) and max all of my fans, I can have my GPU, a Pascal Titan at max overclock running with max load temps of 24C It's loud though and there is no point to it.

So, I guess custom water cooling for me is a matter of getting the most out of my GPU, while also keeping everything nice and quiet at load.

Start losing boost clocks over 40c? CPU or GPU? I thought most didn't start doing any limiting until like 60-65c?

 

[Ready4Dis]

If I could only cool one, I would choose GPU over CPU any day, hands down. Seeing those nice low temps on the CPU tickles my jimmies, but it doesn't translate to much in terms of performance. GPUs, however... a 1080ti at 35°c is practically a different card from an identical one at 70°c.

Different card? They perform exactly the same...
"While NVIDIA doesn’t publish this information directly, the GTX 1080 Ti’s preset thermal throttle point is 84C, which is a degree or two higher than on NVIDIA’s previous cards. As a result, the card reaches equilibrium at a slightly higher temperature than NVIDIA’s other cards."
https://www.anandtech.com/show/11180/the-nvidia-geforce-gtx-1080-ti-review/16

"Nvidia's GTX 1080 Ti, for example, has a throttling point of 84 degrees. If you keep the temperature below 80 degrees you leave yourself with a bit of breathing room".
https://www.techspot.com/amp/article/1638-what-is-thermal-throttling/

Granted I'd prefer sub 80C temps personally, but you won't notice a difference in performance between 35c and 70c... Just notice a difference in noise and $$. Unless they are both lying (as well as others who came to the same conclusion independently), I'm not sure why you think there is some huge magic delta if neither is throttling. The only thing I can think is you MAY be able to maintain a slightly higher clock at lower temps, but nothing like the night and day difference you are suggesting.

 

[Keljian]

Watercooling can smoke air cooling, if you have sufficient radiator space, it is also substantially more expensive.

Typically AIOs don't do better than high end air cooling unless you crank the fans right up in which case they become much louder and they only give you 5-8C more headroom. AIOs do have more thermal mass though which means that it takes a lot longer to heat them up, which for peaky loads can be good. But with some judicious tuning of fans air can be just as quiet.

I submit that you could, in theory, use ion wind cooling for either air or water, but I haven't seen it done in a while

 

[Ready4Dis]

Watercooling can smoke air cooling, if you have sufficient radiator space, it is also substantially more expensive.

Typically AIOs don't do better than high end air cooling unless you crank the fans right up in which case they become much louder and they only give you 5-8C more headroom. AIOs do have more thermal mass though which means that it takes a lot longer to heat them up, which for peaky loads can be good. But with some judicious tuning of fans air can be just as quiet.

I submit that you could, in theory, use ion wind cooling for either air or water, but I haven't seen it done in a while

The thermal mass helps a lot when you have a lot or even a few short spikes. Don't get those annoying oscillating *fan* cycles!!

Edit: fixed auto correct issue

 

[Keljian]

The thermal mass helps a lot when you have a lot or even a few short spikes. Don't get those annoying oscillating can cycles!!

If you have oscillating fan cycles then you're doing it wrong. I've set my d15 up so that the fans use the average cpu temperature over 10 secs, spikes/pulses gone.

The exception to this is when you regularly have a load that is 2-5 minutes long, in which case you might have a reason for water cooling. I stress might. Fusion 360 or Inventor may be actual use cases for this.

My software compiles are usually done in seconds and don't register a blip.

This chart is interesting and on topic (from here: https://www.gamersnexus.net/hwreviews/3596-ek-aio-drgb-360-240-review-liquid-coolers ):

Esentially if you have a constant load over about 80 seconds (1m40s) but less than 260 seconds (4m20s) then it becomes worthwhile to have an AIO instead of high end air..

 

[Zarathustra[H]]

Start losing boost clocks over 40c? CPU or GPU? I thought most didn't start doing any limiting until like 60-65c?

That's on my Pascal Titan X. In order to maintain constant boost clocks in the 2030Mhz to 2080Mhz level I need to maintain temps at or below 40C. Voltage seems to have no effect. I'm guessing that's just what the early GP102's were like.

 

[Zarathustra[H]]

If you have oscillating fan cycles then you're doing it wrong. I've set my d15 up so that the fans use the average cpu temperature over 10 secs, spikes/pulses gone.

True, you have stopped your fans from oscillating, but you are also allowing a short rise in CPU temp above target. If you are anywhere near its limit, that could be a problem.

I personally don't believe on controlling fans based on silicon temps in water loops, because this allows the coolant to heat up a lot when you are at idle, so when you add a sudden load, you overshoot the desired silicon temp dramatically while the fan struggles to remove all that stored heat in the coolant.

The good thing here is that Delta T (temp difference between silicon and coolant loop) is remarkably stable given a fixed system at full load. Doesn't really matter what your temps are in your loop or ambient, if you are at full load, your silicon temp can almost always be predicted by <loop temp> +k, where k can determined experimentally.

This is why I believe strongly in fan control based on loop temp. Set your loop temp controls to <desired silicon temp> -k (Use the worst case between your CPU and GPU) and you wont even need to worry about using any fan curve smoothing or averages. The thermal mass of the coolant gives the temperature changes enough inertia that everything winds up being nice and smooth without having to worry about temp spikes above your target.

 

[VanGoghComplex]

Different card? They perform exactly the same...
"While NVIDIA doesn’t publish this information directly, the GTX 1080 Ti’s preset thermal throttle point is 84C, which is a degree or two higher than on NVIDIA’s previous cards. As a result, the card reaches equilibrium at a slightly higher temperature than NVIDIA’s other cards."
https://www.anandtech.com/show/11180/the-nvidia-geforce-gtx-1080-ti-review/16

"Nvidia's GTX 1080 Ti, for example, has a throttling point of 84 degrees. If you keep the temperature below 80 degrees you leave yourself with a bit of breathing room".
https://www.techspot.com/amp/article/1638-what-is-thermal-throttling/

Granted I'd prefer sub 80C temps personally, but you won't notice a difference in performance between 35c and 70c... Just notice a difference in noise and $$. Unless they are both lying (as well as others who came to the same conclusion independently), I'm not sure why you think there is some huge magic delta if neither is throttling. The only thing I can think is you MAY be able to maintain a slightly higher clock at lower temps, but nothing like the night and day difference you are suggesting.

The card doesn't throttle until 84°c, sure, but throttling is not what I'm talking about. Throttling is intentionally lowering clocks to below rated levels in order to get thermals under control. Open up a hardware monitor and run a benchmark - you'll see. The card does lots of clock adjustment based on temperature before it gets to the point that it's throttling. Those constant changes in clock speed are noticable as little jitters in frame times and a degradation of your minimums.

With a watercooled card, you can overclock the snot out of it and it will hold those boost clocks rock solid. GPUs aren't CPUs: their performance/temperature relationship isn't as simple as throttling/not throttling.

 

[VanGoghComplex]

Water pumps also make noise.

Crappy ones / poorly installed ones, maybe.

 

[Ready4Dis]

If you have oscillating fan cycles then you're doing it wrong. I've set my d15 up so that the fans use the average cpu temperature over 10 secs, spikes/pulses gone.

The exception to this is when you regularly have a load that is 2-5 minutes long, in which case you might have a reason for water cooling. I stress might. Fusion 360 or Inventor may be actual use cases for this.

My software compiles are usually done in seconds and don't register a blip.

This chart is interesting and on topic (from here: https://www.gamersnexus.net/hwreviews/3596-ek-aio-drgb-360-240-review-liquid-coolers ):
View attachment 254157

Esentially if you have a constant load over about 80 seconds (1m40s) but less than 260 seconds (4m20s) then it becomes worthwhile to have an AIO instead of high end air..

This is common on my server, annoys the crap out of me, but it's pretty locked down the way it handles fan changes, although I was able to write my own script to override default behaviour. Also, allowing spikes is ok at lower temps, but at higher temps you want fast responses. Blender is also known to spike for short durations while working (I do some unreal engine/VR dev, normal compiling mostly c++, and some 3d modelling). The thermal mass of water is much better for this, among other things. Sadly, I can barely fit the aio in my ITX case as it is, probably not going to fit a custom loop with reservoir.

 

[Ready4Dis]

The card doesn't throttle until 84°c, sure, but throttling is not what I'm talking about. Throttling is intentionally lowering clocks to below rated levels in order to get thermals under control. Open up a hardware monitor and run a benchmark - you'll see. The card does lots of clock adjustment based on temperature before it gets to the point that it's throttling. Those constant changes in clock speed are noticable as little jitters in frame times and a degradation of your minimums.

With a watercooled card, you can overclock the snot out of it and it will hold those boost clocks rock solid. GPUs aren't CPUs: their performance/temperature relationship isn't as simple as throttling/not throttling.

Maybe it's a custom card OC thing when it auto boosts... As seen here:
https://www.tomshardware.com/reviews/msi-gtx-1080-ti-lightning-z,5162-5.html

It DOES throttle above 40C, it's an aib partner OC board, so they did some custom throttling on it. This is what I was actually looking for, I thought I had seen something like this before. In this case, lower = faster as you guys noted. It went from 2000mhz to 1950... Hardly a night and day difference, although slightly more consistent frame maybe (I imagine you'd be hard pressed to notice it fluctuate as it was a pretty straight line, not bouncing up/down)? About 2% difference though, so it's not nothing either.

 

[VanGoghComplex]

Maybe it's a custom card OC thing when it auto boosts... As seen here:
https://www.tomshardware.com/reviews/msi-gtx-1080-ti-lightning-z,5162-5.html

It DOES throttle above 40C, it's an aib partner OC board, so they did some custom throttling on it. This is what I was actually looking for, I thought I had seen something like this before. In this case, lower = faster as you guys noted. It went from 2000mhz to 1950... Hardly a night and day difference, although slightly more consistent frame maybe (I imagine you'd be hard pressed to notice it fluctuate as it was a pretty straight line, not bouncing up/down)? About 2% difference though, so it's not nothing either.

It's not that an air cooled 1080ti isn't fast. It's just that it's not consistent. It's easy to notice once you've spoiled yourself and become accustomed to high refresh rate monitors, G-sync and the like.

All my rant is to say that while watercooling a CPU has benefits, in my experience, watercooling the GPU creates more tangible ones.

 

[Keljian]

True, you have stopped your fans from oscillating, but you are also allowing a short rise in CPU temp above target. If you are anywhere near its limit, that could be a problem.

Actually this is not the case (with high end air) as displayed in the chart, there is a 1m29s delta before reaching a steady state, a 30 second burst is not going to move anything greatly. That said, I am not seeking ragged edge overclocking, so ymmv.

I agree with your comments with regards to water cooling though.

 

[Zarathustra[H]]

It's not that an air cooled 1080ti isn't fast. It's just that it's not consistent. It's easy to notice once you've spoiled yourself and become accustomed to high refresh rate monitors, G-sync and the like.

All my rant is to say that while watercooling a CPU has benefits, in my experience, watercooling the GPU creates more tangible ones.

Agreed. Dones't even take high refresh if you are at 4k Ultra settings though. You need every ounce of boost just to stay out of the upper 50's framerates as much as possible.

 

[exlink]

It's far from dead. I just finished building my first custom loop in probably over a decade. Only reason I stopped was because it was cumbersome maintaining the loop while living in a dorm room.

Put things in perspective:

Air Cooled
3700X Idle - 45*C
3700X AIDA64 Load - 92*C
1080 Ti FTW3 Idle - 43*C
1080 Ti FTW3 Doom Eternal - 73*C

Water Cooled (Custom Loop)
3700X Idle - 42*C
3700X AIDA64 Load - 83*C
1080 Ti FTW3 Idle - 33*C
1080 Ti FTW3 Doom Eternal - 45*C

My GPU benefited from the water cooling a lot more, but getting my CPU down to actual reasonable temps under AIDA64 load is also a huge plus. My GPU's load temp while gaming under water is essentially the same temp as it was at idle with air cooling. Not to mention my entire rig is nearly silent now as well.

 

[M76]

I don't know who said it was dying. But it is also not true that the only way to build a low noise system is with a custom loop.

 

[THRESHIN]

I'd argue that a custom loop can actually be cheaper.

What's that you say? Threshin is smoking crack again? While that may be true, it's besides the point.

A custom setup has a substantial up front cost. It can be huge compared to an air cooler. However, you'll keep those parts for many years. My custom setup is nearing 15 years old. Pump, rad, reservoir and GPU block have not changed in all that time. CPU block I changed about 6 month ago due to mounting issues with the old block. I've also had to replace tubing when moving to a new case.

Then consider that you have huge cooling capacity. That means higher overclocks and your hardware can last you longer before an upgrade.

 

[Zarathustra[H]]

I don't know who said it was dying. But it is also not true that the only way to build a low noise system is with a custom loop.

At idle most cooling methods can be tuned to be quiet.

At full load it can be challenging without things getting so hot they throttle, or using very low power components.

 

[somebrains]

I’ll wait for Ampere power/temp complaints so we can rehash gpu cooling being the more beneficial path.

On the cpu side 10600k, 10900k, upcoming amd xt all look like install open loop candidates to me.

 

[d3athf1sh]

The R9 Fury X is a serious beast. It was money well spent.
And the sheer collector's value is immense. It's some seriously sleek engineering.

Can you link that? It is a classic and we would all do well to see a real picture of it! for example here https://www.pcworld.com/article/293...ard-review-amds-thoughtful-4k-powerhouse.html
when it was supposed to be the first 4k graphics card. That was a bad attempt at an AIO graphics card. Sorry but it was. I dont want to bust Aegir's balls but it was.

not trying to burst aegir's balls either but is that the same one that's for sale on ebay at this link? https://www.ebay.com/itm/AMD-Radeon...794702?hash=item23d9509e8e:g:S0AAAOSw~cRe0LKK

 

[1337Goat]

not trying to burst aegir's balls either but is that the same one that's for sale on ebay at this link? https://www.ebay.com/itm/AMD-Radeon...794702?hash=item23d9509e8e:g:S0AAAOSw~cRe0LKK

Yeah, I'm trying to sell it. =p

 

[Zarathustra[H]]

not trying to burst aegir's balls either but is that the same one that's for sale on ebay at this link? https://www.ebay.com/itm/AMD-Radeon...794702?hash=item23d9509e8e:g:S0AAAOSw~cRe0LKK

Seeing that it is a used card card that roughly ties a new $200 GTX 1660 (Non-super, non-TI) but at ~2.5x the TDP, it certainly does seem like there is at least a little collector aspect to it driving up the price


I mean, because if you were shopping strictly based on price to performance in the $200 category, and there were two cards that performed similarly, would you choose the new 120w card with warranty, or the used 275w card with no warranty?

 

[1337Goat]

Seeing that it is a used card card that roughly ties a new $200 GTX 1660 (Non-super, non-TI) but at ~2.5x the TDP, it certainly does seem like there is at least a little collector aspect to it driving up the price


I mean, because if you were shopping strictly based on price to performance in the $200 category, and there were two cards that performed similarly, would you choose the new 120w card with warranty, or the used 275w card with no warranty?

Yeah, there's for sure collector value to it, considering it's the highest end that AMD offered at the time.
I set the price based on what was recommended to me here, so don't look at me like I'm some sort of mungus.

 

[Keljian]

I'd argue that a custom loop can actually be cheaper.

What's that you say? Threshin is smoking crack again? While that may be true, it's besides the point.

A custom setup has a substantial up front cost. It can be huge compared to an air cooler. However, you'll keep those parts for many years. My custom setup is nearing 15 years old. Pump, rad, reservoir and GPU block have not changed in all that time. CPU block I changed about 6 month ago due to mounting issues with the old block. I've also had to replace tubing when moving to a new case.

Then consider that you have huge cooling capacity. That means higher overclocks and your hardware can last you longer before an upgrade.

I have a D15, per my sig, I have had it for 5 years across 3 chips, it was $120 (AUD) at the time, it is now $149 retail. Noctua recently replaced the fans under warranty.

The reason I went D15 in the first place is because my d5 pump failed, and it would have cost the same to replace it as it did to get the D15.

Note I recently priced up a basic ddc water cooling setup on Aliexpress including 360 rad/block but without coolant and it came to $380AUD delivered. Add coolant and it ends up just over $400

Please tell me how water cooling can be cheaper than my setup.

 

[86 5.0L]

The loudest fans are gpu fans ramping up. No contest. I can dig up hundreds of forum posts claiming that the first GTX titan ran hot. I can dig up hundreds of any first generation card `forum posts that ran hot. Please dig up one single piece of evidence that a gpu card under water ran hot no matter how shitty the loop was and then you can talk shit.
Forgive my language.

you obviously havent seen my delta fan cooled tower

 

[THRESHIN]

I have a D15, per my sig, I have had it for 5 years across 3 chips, it was $120 (AUD) at the time, it is now $189 retail. Noctua recently replaced the fans under warranty.

The reason I went D15 in the first place is because my d5 pump failed, and it would have cost the same to replace it as it did to get the D15.

Note I recently priced up a basic ddc water cooling setup on Aliexpress including 360 rad/block but without coolant and it came to $380AUD delivered. Add coolant and it ends up just over $400

Please tell me how water cooling can be cheaper than my setup.

keep in mind i'm going on memory here and prices are in CAD

swiftech MCP655 (a D5) used for $80, used reservoir for $20, swiftech MCW60 GPU block and apogee GT for less than $40 each also used, replaced CPU block recently with a new heatkiller IV for $80 (got the cheaper version), new black ice pro 3x 120mm rad for about $60, tubing costs what $20? coolant i run deionized water from the automotive parts store which costs less than $3 a jug. so we're looking at $400 CAD for argument's sake and this is over a period of 15 years. so i'm less than $27/year.

and that is how you do it cheap. largely thanks to the items i purchased on these forums.

 

[Keljian]

keep in mind i'm going on memory here and prices are in CAD

swiftech MCP655 (a D5) used for $80, used reservoir for $20, swiftech MCW60 GPU block and apogee GT for less than $40 each also used, replaced CPU block recently with a new heatkiller IV for $80 (got the cheaper version), new black ice pro 3x 120mm rad for about $60, tubing costs what $20? coolant i run deionized water from the automotive parts store which costs less than $3 a jug. so we're looking at $400 CAD for argument's sake and this is over a period of 15 years. so i'm less than $27/year.

and that is how you do it cheap. largely thanks to the items i purchased on these forums.

You missed fittings, but hey, let’s say you got those for free.

A comparison: NH-D15 $120/5 years = $24, and if it lasts another 5 years/I don't get sick of it, that's $12/year. In the unlikely event it lasts me 15 years like the water cooling setup, that’s $8/year.

Note this is AUD, in CAD it's about $22/year over 5 years, or $11/year over 10.

I really want you to prove me wrong so I can go and spend up on watercooling, but looking at the numbers I can’t see how it works.

 

[Keljian]

Note the total for my aliexpress order including tax was $377AUD shipped just for the cpu cooled, but brand new parts.

A good 360mm AIO is about $200-250.

I haven’t pulled the trigger yet, still not making sense...

If you have watercooling gear, and you've had it for a while, and you write off the original sunk cost (pump,res, rad, maybe case too) then getting a cpu block every few years isn't a big deal. If you're starting from scratch it's a different story.

 

[Dayaks]

Mine is the best and I can hear it.
Do crappy ones last 12 years?

I put my pump/fans in my basement where no one can hear them.

 

[Nimisys]

Note the total for my aliexpress order including tax was $377AUD shipped just for the cpu cooled, but brand new parts.

A good 360mm AIO is about $200-250.

I haven’t pulled the trigger yet, still not making sense...

If you have watercooling gear, and you've had it for a while, and you write off the original sunk cost (pump,res, rad, maybe case too) then getting a cpu block every few years isn't a big deal. If you're starting from scratch it's a different story.

Cpu blocks honestly don't need changed out all that often. My original block has gone from a 2600k to a 1700x with nothing more than a new mount.

Bigger issue is the GPU, full coverage blocks are expensive and frequently only work with a single generation and/or model. The universal blocks became less effective as the need for keeping the rest of the components on the card cool increased beyond passive heat sinks.

However the pump, res, radiator(s), tubing type, only really need changed in the event of failure, or more likely, feeling the desire to do something / upgrade itch.