Stren's Titan Water Block Roundup

[stren]

wow, only 20c. Seems like they run cool enough that you wouldn't need a heat spreader or anything back there. IR should work..look forward to your additional results..

Yeah they do get hotter when you run an air cooler as they would roughly track the GPU temps, i.e. ~70C when the gpu was at 75C on air.

wo looks nice

only 20C? Any chance of you testing a universal block on this card?

Nice results here! As an ati fan i'm not going to buy titan but good wb testing is my soft spot :-) also any universal block would be nice...

send from Deathstar via HAL9000

Yeah we'll see how time goes, with multiple blocks and multiple runs and the extra time to let MX2 cure then I might want to be selling this card before I get to that point. Plus I'm off on vacation for 10 days on wednesday so I'm trying to get out what data I can before I leave...

[stren]

Alright then. I got done with the first run on the Koolance block which is a real puzzler. I'm definitely going to have to run this block again to make sure this is legit:



>> edit after removing the koolance block I noticed the TIM wasn't spread as well as it could have been and was still quite thick. This either means that I didn't tighten the block enough (unlikely but always possible) or that the contact area for the GPU is actually too high and isn't making good contact with the IHS.

[avabaska]

stren
>This either means that I didn't tighten the block enough (unlikely but always possible) or that the contact area for the GPU is actually too high and isn't making good contact with the IHS.
Will be new results?

And question about EK. This thing got like baffle before VRM zone. Sure it's made to reduce the resistance. But is that good for VRM cooling? Titan can take from PSU 330+ W (with modded BIOS) just under 3DMark11. Don't know about VRM's margin of safety on this card, but I never heard about deaths of these cards even under air cooling. So maybe nothing worry about.

Ordered simple EKs + backplates. But Aquacomp's blocks are most beautiful, as usual. I think they must make XXL Edition, like EK did!

[stren]

stren
>This either means that I didn't tighten the block enough (unlikely but always possible) or that the contact area for the GPU is actually too high and isn't making good contact with the IHS.
Will be new results?

And question about EK. This thing got like baffle before VRM zone. Sure it's made to reduce the resistance. But is that good for VRM cooling? Titan can take from PSU 330+ W (with modded BIOS) just under 3DMark11. Don't know about VRM's margin of safety on this card, but I never heard about deaths of these cards even under air cooling. So maybe nothing worry about.

Ordered simple EKs + backplates. But Aquacomp's the most beautiful, as usual.

Yeah I'll be retesting all the blocks, I'm trying to run every block once first in case something dies.

I did forget VRM and memory temps on the EK block, and Koolance had a backplate so I couldn't measure that one, but I have those temps on the AquaComputer block that I'm running now. Yeah we'll see about the VRM's. I'm measuring about 50C over ambient on the VRM's right now with the AC block. I think adding MX2 to both sides of the thermal pad really helps (which this run doesn't have), but I need to get firm data on that to be honest. I also noted that a thermal pad on the VRM sense resistor really helps to stabilize the card's power measurements over temperature. That resistor does not have a zero temp co and could be causing people to power limit unnecessarily.

[naennon]

for the koolance vid titan

do NOT use pads on vram chips
block will not get enough contact to the gpu
simply use thermal grease on vram chips - much better

thats what i meant with koolance temperatur problems - just figured out

[stren]

for the koolance vid titan

do NOT use pads on vram chips
block will not get enough contact to the gpu
simply use thermal grease on vram chips - much better

thats what i meant with koolance temperatur problems - just figured out

Cool I'll try this next - did that help your temps?

[Martinm210]

Interesting flow behavior on the Koolance. Perhaps there is some block expansion happening as flow and internal pressure increases causing that. I know I have seen some odd things in pressure drop curves flattening out like the GTX radiator series that was presumably due to the very thin tubes expanding under pressure.

Love the work and think the one mount concept first is very wise. Take care of that card and take your time. Enjoy the Vaca.. much deserved break!!

[avabaska]

stren
>...but I have those temps on the AquaComputer block that I'm running now. Yeah we'll see about the VRM's. I'm measuring about 50C over ambient on the VRM's right now with the AC block.
+50C from ambient directly on one of the transistors? And I guess it at stock clocks (with boost only)? I think it's a good idea to use a backplate with pads (need to pick up thickness) for VRM zone + slow fan for cooling that backplate. Did it with GTX 580s and will repeat with Titans.

>I think adding MX2 to both sides of the thermal pad really helps (which this run doesn't have), but I need to get firm data on that to be honest.
EK wrote about this in manual. But is that OK if I have MX3? And is there any benefit? If it's only 1-2-3 degrees...

>[i]I also noted that a thermal pad on the VRM sense resistor really helps to stabilize the card's power measurements over temperature. That resistor does not have a zero temp co and could be causing people to power limit unnecessarily./i]
Sorry, I don`t understand - resistor on Titan`s PCB?
I only OC`ed this card under air (stock) in open case, fan was at 5000+ RPM and + two 120 mm side Slip Streams (1600 and 1900 RPM). Card consumption was up to 330W and back side of PCB (on the opposite side of VRM) was about 65-70C (infrared thermometer). Ambient ~24C.

naennon
>do NOT use pads on vram chips
block will not get enough contact to the gpu
But according Koolance manual, we must use them. That's strange. But also they wrote:

Multiple types of heat transfer pads are included with some Koolance blocks.
You might not require both (check the pad diagram). This is done because card
manufactures change the component height tolerance from time to time.

But it looks confusing, I prefer EK way

[stren]

Interesting flow behavior on the Koolance. Perhaps there is some block expansion happening as flow and internal pressure increases causing that. I know I have seen some odd things in pressure drop curves flattening out like the GTX radiator series that was presumably due to the very thin tubes expanding under pressure.

Love the work and think the one mount concept first is very wise. Take care of that card and take your time. Enjoy the Vaca.. much deserved break!!

Thanks Martin - interesting - yes it reminded me of a laminar to turbulent transition plot from back in the day. But I thought we always had turbulent flow these days so your explanation might be better!

stren
>...but I have those temps on the AquaComputer block that I'm running now. Yeah we'll see about the VRM's. I'm measuring about 50C over ambient on the VRM's right now with the AC block.
+50C from ambient directly on one of the transistors? And I guess it at stock clocks (with boost only)? I think it's a good idea to use a backplate with pads (need to pick up thickness) for VRM zone + slow fan for cooling that backplate. Did it with GTX 580s and will repeat with Titans.

+50C above ambient measured on the back of the board underneath where the VRMs are. This was the worst spot of the 6 phases, others were cooler. This raised to ~57C at very low flow. This is without TIM on the thermal pads though so I bet it could be improved.

>I think adding MX2 to both sides of the thermal pad really helps (which this run doesn't have), but I need to get firm data on that to be honest.
EK wrote about this in manual. But is that OK if I have MX3? And is there any benefit? If it's only 1-2-3 degrees...

Any non conductive TIM should be fine. I don't have a definitive numerical answer for you, but the EK block (where I did use TIM on the vram pads) was about 15C better on the VRAM chips on the back of the card than the AC block (where I didn't use TIM on the pads). Now that could be block differences but it might be the TIM.

>[i]I also noted that a thermal pad on the VRM sense resistor really helps to stabilize the card's power measurements over temperature. That resistor does not have a zero temp co and could be causing people to power limit unnecessarily./i]
Sorry, I don`t understand - resistor on Titan`s PCB?
I only OC`ed this card under air (stock) in open case, fan was at 5000+ RPM and + two 120 mm side Slip Streams (1600 and 1900 RPM). Card consumption was up to 330W and back side of PCB (on the opposite side of VRM) was about 65-70C (infrared thermometer). Ambient ~24C.

There is a resistor (or at least I assume it's a resistor) for each phase next to the regulator labelled R22. I assume this means it's a high power 0.22 ohm resistor used to measure the output current of the regulator. But this is a guess. Most instructions tell you to not put a thermal pad on it, but when I didn't on the Koolance block the power measurements were very sensitive to block temperature and would raise 5% over the test. Given that with a stock bios you can easily hit the power limit when overclocking then adding a thermal pad to that resistor might help extend your ability to overclock it.

naennon
>do NOT use pads on vram chips
block will not get enough contact to the gpu
But according Koolance manual, we must use them. That's strange. But also they wrote:

Multiple types of heat transfer pads are included with some Koolance blocks.
You might not require both (check the pad diagram). This is done because card
manufactures change the component height tolerance from time to time.

But it looks confusing, I prefer EK way

Yeah I saw this yesterday too (I didn't have a manual shipped with the block as it was shipped so quickly). So they kind of agree except Koolance should have specified that you may not even have to use a pad at all if the spread is still too thick.

Compiled the AquaComputer GPU core results - it's a bunch more restrictive as we knew from the earlier plots. In a single CPU, GPU, Rad, 4xQDC loop it could drop flow from ~1.05GPM (for EK and Koolance) to 0.85GPM. At higher flows it does about the same as the EK block, however at lower flows it does better. Between the two I'd recommend the AC for multiple cards in parallel and the EK for a single card, or cards in series.

Note the lowest pump setting for the AC block is off the scale of my flow meter. The float was not pegged so I made an estimate of 0.28GPM, but this number will be far from exact.



That's about all I'll have time for before I leave for vacation. I'll be back by the end of the month and will run the XSPC and then the EVGA. Hopefully the other blocks will arrive shortly after I get back.

[Martinm210]

Sweetness...finally some good data to understand lower flow rate performance too. I had always assumed the larger fin designs of GPU block would be more sensetive to lower flow rate. That AC block clear down to .25GPM is still doing extremely well. Just another reason not to get overly worked up about thinking we need high flow rates.

[frank anderson]

Thx Stren, and Hello Martin

Those blocks are absolutely beautiful, loving the AC back plate.. definitely ordering a couple titan's this year (maybe 3 if it scales well in surround ), water blocks will definitely be used so keeping a eye on this thread.

Looks like EK's and AC's are going toe to toe this round while Koolance is trailing just a step behind, so in the end it may just boil down to aesthetics and which company you prefer, those nickel plated AC blocks just brings another meaning to the word "Sexy".... ummmmm

@shoggy, is there a SLI and Tri SLI block available that is compatible with the heat pipe or this is still in the works?

[naennon]

sure it helped me out with my temps @ stren

do it my way

[tiborrr]

Love your work, keep it up! Have a nice vacation, recharge your batteries.

[avabaska]

+50C above ambient measured on the back of the board underneath where the VRMs are. This was the worst spot of the 6 phases, others were cooler. This raised to ~57C at very low flow. This is without TIM on the thermal pads though so I bet it could be improved.

Thanks for the answer and I'm very sorry for meddling in your topic
Unfortunately, I can measure only the back side of PCB (so it can be + up to 10C to this temp = transistors temp) and there was up to 68C under heavy load (300+W) with water temp 33C:



Also there was a faint air cooling (92 mm, 1000 rpm):



And I did not succeed with pads for the back side of PCB, because had only hard pads from Phobya and a backplate began to deform when tightening the screws. So I decided not to install backplates, instead of this I used heatsinks for memory and just a fan for VRM zone. Not so effective, ugly but it works.

Any non conductive TIM should be fine. I don't have a definitive numerical answer for you, but the EK block (where I did use TIM on the vram pads) was about 15C better on the VRAM chips on the back of the card than the AC block (where I didn't use TIM on the pads). Now that could be block differences but it might be the TIM.

It possible, but 15C on VRAM chips I thought if we have a strong pressing, like this (GTX 680):



with a pretty hard pads (like EK's), there is no need of any other thermal interface. But if it will be a copper or aluminium plate, sure we need something like MX2 or 4.
But look how squeezed they are... Maybe thermal paste it's a good idea. I'm starting to think, that I need to reinstall blocks (laziness...) or wait for the next upgrade

There is a resistor (or at least I assume it's a resistor) for each phase next to the regulator labelled R22. I assume this means it's a high power 0.22 ohm resistor used to measure the output current of the regulator. But this is a guess. Most instructions tell you to not put a thermal pad on it, but when I didn't on the Koolance block the power measurements were very sensitive to block temperature and would raise 5% over the test. Given that with a stock bios you can easily hit the power limit when overclocking then adding a thermal pad to that resistor might help extend your ability to overclock it.

That's really strange

Between the two I'd recommend the AC for multiple cards in parallel and the EK for a single card, or cards in series.

But our temps will be higher. BTW, I have not noticed any appreciable difference between three GPUs at ~230 l/h. Even at 175 l/h the difference was very small. Maybe even at 150 l/h it's better to use a series connection? In parallel, temp of all cards = last card in series connection. But if we have sufficient (or minium) flow only with one card and want to add two cards more, maybe we need use parallel.

[Martinm210]

AC block seems to be leading both aesthetics and performance so far although it does appear to be a tiny bit more restrictive. Have a great vacation, look forward to more later.

[juthoss]

Titan's block from Bitspower

[.::NeO::.]

Titan's block from Bitspower

Fugly.

wtf were they thinking with those huge black screw at the top?

also, the internals seems a bit too simple and poor

[avabaska]

Cooling that memory chips on the back side - useless Max temp (55C) got chips near to VRM zone. But 55C it's nothing. So backplate it's only for aesthetics

[frank anderson]

I have owned bitspower blocks (580 GTX models) in the past, they are surprisingly well built, I put those things through torture, literally... Remove / Installed about 5 times, ran coolants, mayhem dyes, flushed with tap water then distilled uncountable times, in the end I use distilled and silver coil only for long use, the quality of the nickel never budged and it never leaked, they are pure eye candy if you pop some LED's in them..

Performance wise is a bit lacking, but beats stock air cooling by far,if you are after pure aesthetics, picture below is a old shot of mine.







for my new cards, I have ordered some AC blocks with the passive back plates

I wouldn't say back plates are purely aesthetics, for the temps it may not help much or be negligible as far as performance is concerned, but for the stability of the card, it really does help a lot, going tri or quad graphic cards + blocks + bridge, it weighs quite a bit.

[stren]

Stren is back! XSPC is on the bench then followed by the hydrcopper. So hopefully those results will be in by the end of the week. I'm following up with BP, alphacool and watercool to make sure I get sent the blocks soon now that they are all ready it seems. Phobya should be ready next month.

[paulbagz]

Snip

Ouchies, most expensive SLI bridging ever! (Looks awesome as though )

-PB

[Koolance]

Hey Stren,

I think you have a faulty block from us (warped or otherwise not flat), so we'll ship out another one. You mentioned a less than desirable paste spread. Do the standoffs sit flush when placed onto a flat surface, or does the block rock a little? In our tests on both Titan and GTX 780, the GPU was noticeably cooler with VID-NXTTN than with EK's. This "should" be expected since we're using denser and thinner fins, so I think this certainly sounds like a contact issue.

A little off-topic, but our Titan block will be the first to use our new upcoming acetal SLI bridge connectors (as opposed to our usual sliding pipes). That's the reason the fitting area can be removed from the main block and has its own o-ring seal.

Tim

[stren]

Hey Stren,

I think you have a faulty block from us (warped or otherwise not flat), so we'll ship out another one. You mentioned a less than desirable paste spread. Do the standoffs sit flush when placed onto a flat surface, or does the block rock a little? In our tests on both Titan and GTX 780, the GPU was noticeably cooler with VID-NXTTN than with EK's. This "should" be expected since we're using denser and thinner fins, so I think this certainly sounds like a contact issue.

A little off-topic, but our Titan block will be the first to use our new upcoming acetal SLI bridge connectors (as opposed to our usual sliding pipes). That's the reason the fitting area can be removed from the main block and has its own o-ring seal.

Tim

Hey Tim - Yes there's a very slight wobble when I put it on a flat surface only in one direction though. And I love that you're doing bridge connectors too! If you decide to send a new one can you also send the short screws so that I don't have to use the backplate? That way I can measure the back of the board VRM/VRAM temps Otherwise once I'm done with the first test on all the blocks I have I'll go back and run without thermal pads to get some better data.

[Koolance]

Ok, we'll ship today with two GTX 780 hardware packs (shorter screws).

Thanks,
Tim

[stren]

Thanks Tim

For now I've removed the Koolance block off the chart until the new block gets here. I completed the XSPC run. It's less restrictive than the AquaComputer (though not as much as the EK) but performs the best so far. Now that we've zoomed in we can see that the data is a bit noisy. This is partly because this is a single run with ambients moving (my garage is not climate controlled) and because I'm using MX2 which although it has a short curing time still has some small curing effects despite doing an overnight pre run burn-in.