Watercooling

[af_newbie]

Is it worth it to watercool your GPUs?

[1713556852]

1713556852

[tritium]

money wise, no its not worth it.

it also isn't a magic heat remover, the heat needs to still be disapated from the radiator so the room temperature will be the same. it is much quiter if you have decent fans, and the card temps will be much lower

personally i do watercool

[portron]

I've got four 7970's, and for that matter, 4 - 5870's both on GD70's in HAF X cases.

Stock fans on both the cards and the case(s) temps don't above 78c at the highest, 58c at the lowest.  

I don't have pics of the 7970 but here are the 5870's:



Just make sure you get the turbine style fan design that blows out the back of the case and draws from the front (reference design)... instead of the fans being in the middle.

Keeping Dual GPU cards cool, packed that tightly together in a HAF X case is another story...

[Phlegmsandwich]

I watercooled my 3x 5970 rig.  I did it for noise reasons.

I have it for sale here

https://bitcointalk.org/index.php?topic=63300.0

[Inspector 2211]

Thanks.

so you keep the case open?  I'm thinking of getting "Cooler Master RC-942-KKN1 HAF X ATX Full Tower".
Is it an overkill?

I don't keep my cases open in order not to spoil the "flow-thru"effect from front to rear.
I have these Corsair cases, R500 or whatever they are called, very wide ones.
Four graphics cards in each of them.
In front of the graphics cards, there are two 12cm 110cfm fans on top of each other.
Zip-tied to the graphics cards and power cables.
The bottom fan cools the two bottom cards.
The top fan cools the two top cards.
As the room heats up to the 90s, this configuration barely keeps the cards from throttling down.
Without downclocking the RAM, some cards actually do throttle down.
Since I downclock RAM, I don't have that problem anymore.
Or only very rarely.
I just ordered two 135cfm fans from SuperBiiz, advertised special, to find out whether they cool even better.
As a test subject, I will use the rig that has the least MH/s.

[tritium]

if you are watercooling you don't need high airflow through the case or an open case (although it doesn't matter if you do), just mount the radiator either outside the case or venting out. its better to have fans on both sides so that you push and pull.

[waterboyserver]

I run a single 7970 with one full card EK waterblock. My mining operation is in the same room that I sleep; the benefit is the reduction in noise (200mm radiator, quiet pwm pump, and low fan speed), watercooling also allows you to overclock a card far more if energy cost is not an issue (I can reach over 800 MH/s on a single 7970 watercooled on 1275 MHz core, which draws 240 watts).

[rjk]

I run a single 7970 with one full card EK waterblock. My mining operation is in the same room that I sleep; the benefit is the reduction in noise (200mm radiator, quiet pwm pump, and low fan speed), watercooling also allows you to overclock a card far more if energy cost is not an issue (I can reach over 800 MH/s on a single 7970 watercooled on 1275 MHz core, which draws 240 watts).

Nice. You should add it to the mining hardware comparison on the wiki, if you haven't already.

[nebulus]

I have 3 7970, these fucking things get so hot causing the whole system to freeze up. They are also are pretty fucking loud. I put my rack into a closet with a dedicated fan. That room got temped up to about 95F, it was too hot to walk into. You will definitely need an adequate way to cool them. I am looking for one myself.

[Glasswalker]

Here is one of the Bitcoin Syndicate (http://www.btcsyn.com & https://bitcointalk.org/index.php?topic=64638.0) mining rigs we're operating.

It's a quad 6970 system, liquid cooled. Now this system was a bit more expensive to build, but it runs stable and cool. I could probably push it harder than this by a fair bit too, it's not overclocked at all, this is all at stock clocks. But ultimately we'll be using this box for GPU Compute R&D working on big data and predictive analytics work later on once it's done it's mining commitment to the Syndicate (I'll likely personally pay to replace it with BFL or FPGA gear in the next 2-3 months to fully replace it's MHash to the syndicate so I can shift it back to doing OpenCL R&D).

Right now it's got a 1500W power supply 90+ efficiency. Haven't measured it's actual wall power draw yet. But it's not tweaked to be the most efficient. It's probably drawing over 1KW for the whole system right now, but that doesn't matter much since at this location electricity is free. (just a ballpark)

The main goal was allowing high throughput without impacting the lifespan of the cards. It's also silent compared to similar air cooled gear. (those fans are all very quiet fans).

Anyway, here are the pics (sorry for the poor quality, taken with my cell phone, in non-optimal lighting, and it has no flash):

[mila]

Here is one of the Bitcoin Syndicate mining rigs we're operating.

if it's rd, I'd like to know where does the difference in reported mining come from.
I compare the screenshots with the info from the stats page (5-7 %)

thumbs up for the noise reduction.

[Glasswalker]

As was mentioned on the syndicate thread, the difference in reported values is due to the difference between raw miner hash rate, and actual "effective" hash rate (from the pool, based on luck/accepted shares/rejects/stales and so on). The effective hash rate will normally (but for some reason not always) be lower than the miner reported hash rate. And the effective hash rate tends to fluctuate more, where the miner hash rate is very stable and barely fluctuates at all.