Watercooled octane render workstation TITAN+3x 780 6GB

[smicha]

OMG, just installed the 480 and on passive mode (I don't have cables to connect fans to my controller) temps went about 5C down. I cannot believe my eyes - 31-33C on all GPUs while rendering in Octane. And none of the Noctua fans is on!

So Glimpse, yes - previous measurements were on one old Phobya 560, now I added the beauty 480 and temps are crazy low.

[glimpse]

OMG, just installed the 480 and on passive mode (I don't have cables to connect fans to my controller) temps went about 5C down. I cannot believe my eyes - 31-33C on all GPUs while rendering in Octane. And none of the Noctua fans is on!

So Glimpse, yes - previous measurements were on one old Phobya 560, now I added the beauty 480 and temps are crazy low.

ok, I can expect 40-50C.. but don't say 30C =DDD 'cos it's just too low..to be true =DDD

You've OVERclocked or UNDERclocked? =DDD - no really, that's impressive, Smicha!

P.S. don't know if You've seen, but AquaComputer actually sell these rads in passive configs, just "a bit bigger" configuration =)
P.P.S. here's the idea for Your next build : https://www.facebook.com/photo.php?fbid=746348708752496 =DDD time to rest..

[smicha]

Need to get cables for noctuas. On 560 the 4xGPUs temps were increasing - so 560 is good for cpu+2xgpus. Adding 480 lowers temps but after 10 minutes temps are also rising. So there is no way to keep temps stable on passive 480. I expect about 40C but with +250mhz on core overclock. I can get no sleep looking at the rad on top of SMH10

More photos soon

PS the last link you've sent - liquid nitrogen? why not

[glimpse]

Yeah, that was a pot for Liquid Nitrogen + custom power delivery =) bit of an insane stuff =DDD

talkin' about passiveness I hope I could get 3x CapeCora equivalent rads next week =)..tBlack will go silent =p

[smicha]

Noctua NF-F12 fans on aquacomputer, and NF-S12A as an exhaust. Primarily they were in pull config, but too noisy then (I assume their high static pressure makes them better in push).







Icy Dock fan replacement

[glimpse]

Nice! =) just an idea, simply put them under radiator, & they push the air through it & exhaust out of the case =)

I've noticed the same that radiator is oder when it sucks air =) but I guess that has to do with it construction as blades are closer to rad if You use it for pull =)

[smicha]

Nice! =) just an idea, simply put them under radiator, & they push the air through it & exhaust out of the case =)

I've noticed the same that radiator is oder when it sucks air =) but I guess that has to do with it construction as blades are closer to rad if You use it for pull =)

Thanks Glimpse. My first thought was to put them under the radiator and push air outside the case. This has one drawback which made me push the air into the case and avoid negative pressure that sucks lots of dust into the case http://www.caselabs-store.com/demciflex-filters/
Anyway I am also not sure of how much cleaner and warmer the case will be. If this will be an issue I'll swap it.

[Jonas]

Super nice pics!

[schodt]

Awesome picture Smitcha - you inspired me to go the same route. Just placed an order on 4x waterblocks, rads etc. for my system today Can't wait to go noise-free!

[smicha]

Econometric analysis of efficiency of Octane Rendered watercooled workstation.

Let's go with some more scientific approach of how good is the water-cooled workstation. Please forgive me if I do too many shortcuts here. I'll answer all questions with pleasure.

One GTX 780 is used.

I've measured Octane speed for 5 variables:
- DL speed (PT and PMC to bo done) in ms/s - as a dependent variable; also in logarithmic form for elasticity interpretation
Independent variables
- Core clock in mHz, from downclocking to 862mHz to overclocking up to 1320mHz (1293 stable), also in log form
- Memory clock in mHz - two measurements: 3005 and 3506mHz, also in log form
These logarithms give me ability to measure of how much I gain in speed performance in overclocking core or memory clocks (ceteris paribus, all other variables constant)
- Binary variable - alpha shadows on 1, off 0
- Binary variable - coherent mode on 1, off 0

So here you are guys.

Interpretation

For levels constant was not significant - removed, all variables (independently) are significant (p-value almost equals zero), join variability very high with high Fischer stat (significant),

Now some important info for Octane Users:

- every 10% in CORE overclock lead to increase of DL speed by 7.7%
- the rest if the effect (2.3%) is due to memory overclock

SO it is much better (3x) to focus on core overclock.

Coherent mode turned on increases (on average) the DL speed by 10%
Alpha shadows reduces it by 10%