Question ZOTAC GAMING GEFORCE RTX 3090 ARCTICSTORM requires some additional water HW.

[istavnit]

I just bought

ZOTAC GAMING GEFORCE RTX 3090 ARCTICSTORM​

In specification it says ArcticStorm Waterblock *additional liquid cooling hardware required.

I have ryzen 9 with liquid cooling.
Could I use that same cooling system (came with my CLX PC) that also came with RTX 2080 no liquid cooling?
Also

This is information about the card: https://www.zotacstore.com/us/zt-a30900q-30p-o

This is My motherboard: https://www.asus.com/us/motherboards-components/motherboards/prime/prime-x570-p/
How can I tell if the 3090 I bought will work with it?

 

[DSzymborski]

Unless you have a custom cooling loop, which will look something like the below, you cannot use that GPU. It's hard to be more specific because we only know what kind of cooling came on your PC if you actually identify it; you're almost certainly the only person in this thread who will ever have been in your house or looked at your PC.

 

[istavnit]

Look inside my PC 1
Look inside my PC 2
Link to motherboard spec
750W Power Supply

Can I reuse my CPU radiator, pump and chain additional component (GPU) in cooling chain? What are the cheapest but still reliable components I need to buy to water cool this thing?

 

[DSzymborski]

Look inside my PC 1
Look inside my PC 2
Link to motherboard spec
750W Power Supply

Can I reuse my CPU radiator, pump and chain additional component (GPU) in cooling chain? What are the cheapest but still reliable components I need to buy to water cool this thing?

That's simply an AIO. You can't do a custom loop with those parts.

There's no such thing as "cheap" when you're doing a custom loop. The better solution would be to return the card if you can.

At a minimum, you're looking at $100-$150 for a quality pump, $50-$75 for a reservoir, $30-$50 for fittings, $20-$30 for soft tubing, $15-$25 for a drain valve, $15-$25 for fluids and a proper water bottle. Its also requires some comfort with this kind of work and much more upkeep.

 

[COLGeek]

Your best option would be to return the GPU and go with an air-cooled model.

If you want to keep, reach out to these guys to see if one of their offerings could replace the water block on what you bought: https://modmymods.com/water-cooling-kits/gpu-water-cooling-kits.html

This is not an inexpensive endeavor as previously mentioned.

No responsible member is going to talk you through kludging an AIO not designed for your needs.

Last option, build a full custom loop.

 

[istavnit]

Thank you all for your advice. The card is returnable, I will most likely return it.
The reason I was interested in the card in the 1st place was to run local Neural Net models predictive models that need the 24GB. I am not looking for overclocking so air cooled will probably meet my needs. Thanks again!
My motivation for the purchase was 2 year warranty. Do you think a trip to MicroCenter could help to settle this? I do have a room on top of my case for a 3X fan setup (not installed, perhaps it is for additional radiator that could be used for GPUs. Is there a downside for liquid cooling? if installed by a "pro"?

 

[bill001g]

There is a huge list of downside to liquid cooling which means you really must have a good reason to do it. A lot of people do it for appearance more than function.

A couple of the big ones.
It might leak and destroy your machine
It has many more parts that can fail and are then complex to replace.
The coolant used has a limited life and can actually clog things up which also can't be easily fixed.

Pretty much the only thing that can fail on a aircooled GPU are the fans and if you are concerned some can be replaced easier than others.

You would have to dig around to try to find some benchmarks that show how much increase in performance you can get on water cooling. Last time I looked there was very little difference between GPU in general that had the same clock settings. Mostly it was how loud the fans might be and how big the card physically was.

 

[istavnit]

Just to circle back on the subject as @bill001g mentioned GPU overclocking gains for ML workloads is miniscule, while pairing water with electricity sounds very meh.
source: https://medium.com/@timyee90/does-o...ove-deep-learning-training-speed-2488f9cf4dbd

Quote: ​

The Verdict:

GPU clock and memory frequencies DO affect neural network training time! However, the results are lackluster — an overall 5.15% when running the CPU at 4.8GHz(+500MHz), the GPU Core clock at 2000MHz(+100MHz), and the GPU Memory clock at 4,000MHz(+200MHz). Each variable contributed to the overall performance increase, but a 5.15% overall reduction in training time is hardly worth writing home about.​