Hondacity's HeatKillers GTX285 in Triple Sli (simple test)

[Hondacity]

Another fine weekend as I got some points from HWbot.

But anways, just sharin about tri sli, in parallel.

Simple Setup:

Single loop: Dual 355s > Koolance 350 > HeatKillers > HWlabs 480 GTX with 4 Scythe Gentle Typhoons 28dB

My flow is alittle below 1.5gpm...look at k350 + dual 355's 480gtx, dunno what the tri sli gpu's introduce as restriction...but I don't care. Flow is overrated.




MY STFF, Watercool did not commission me in anyway, FU who thinks otherwise.



Anyways..lets get to the idle temps

notice all three gpu's idling, no worries here...



Load temps, here I used OCCT's power supply test, cpu and gpu are stressed, psu is also stressed, i love my Enermax Evo1250



Well its obvious parallel gpu blocks work.

HWlabs 480GTX + 4 Gentle Typhoons work too and its silent without undervolting
Part2


****cpu ang gpu's are on separate loops****
The important stuff : The RAD and FANS, quad for the cpu, dual quad for the tri-sli



First 20minutes


40minutes later


gpu1 graph from occt

gpu2 graph from occt

gpu3 graph from occt

--------------------------
************idle temps aren't really that interesting, nvidia throttles the gpu speed to around 300mhz versus max set, temps go down at 300mhz

************restriction on these blocks aren't bad.
Enjoy

Hondacity

[Chruschef]

very nice and even temps across all GPUs, nice

[SNiiPE_DoGG]

This is how I read this - we have TOO much flow in our systems right now as in an amount that is on the flatplane of the thermal performance.

1.5 gpm across the loop means a mere .5 GPM is going through each GPU block. I think we need to re-evaluate our stance on restrictive blocks in a loop with others because right now its looking like FC gpu blocks can perform just as well as series with a third of the flowrate we commonly give them in a series loop.

thoughts?

[systemviper]

Love those HK blocks...

I think flow is as important as it ever was, i mean there is the optimal flow rates to cool your system based on
ur components. If it's slower or faster, there always is a sweet spot...

Plus the OP is using a killer pump / radiator set up, so his temps should be very nice
using a 480 and Typhoons.

but I think that as you would drop his flow, he better have a big honking rad on there
because the flow thru the gpu's goes down by 3. and when it starts getting to slow
then it will easily start to heat up.

Say he only had 1 GPM that is .33 for the cards,
or even .75GPM takes it down to .25GPM per gpu

So i would think that with parallel cooling the flow matters even more then in series.

keep the numbers coming, sweet work.

[SNiiPE_DoGG]

well what I was refering to is that I have my GPU on a DDC3.2 with an FC block on a dedicated loop; I would probably get the same temps more or less if I was using a D-TEK DB-1 for a pump lol

[Jah]

Those HK blocks sure looks sweet, especially with that 3 way sli bridge.
The HK FC blocks seems to doreally well with low flow, but it differs rather much from the "standard" design of GPU blocks, would e interesting to see how a EK style block would perform with such low flow.

[ShoNuff]

Honda thanks for posting the results. I was a bit skeptical at first but based upon what a few others (you, ballzor and Radical53) have posted I was hoping that a parallel solution would work for a Tri-SLi set up. Your results are better than expected.

Thanks for sharing the data.

[BDW88]

Thanks for all the work you do Honda!

[jcool]

I am loving the HK Tri-SLI solution since they first released it for G80. I don't even care if parallel is bad for flow, I just love that 3 way connector.

[b@llz0r]

I don't even care if parallel is bad for flow.

it is?

[jcool]

It is, not for overall flow but for the individual blocks - read post #3

[b@llz0r]

It is, not for overall flow but for the individual blocks - read post #3

meh its a bit of a trade off... 3 blocks in parallel will have less head loss than three in series, the cpu blocks benefit the most from high flow, so by having your gpus in parallel your cpu will get more flow

id say 50c under load is a pretty good result, definitely not gona have temp related stability issues


the point to note from Honda's results is the even temps across all three cards

[Humminn55]

Interesting show, but I had thought it was already pretty established that cooling gpu's was not dependent upon high flow or massive radiator capacity.

On the other hand, I see your monitoring sofware you're using is all over the place in reference to your cpu's core temps. One has the highest core at 66C, one has it at 74C....odd, that.

FU who thinks otherwise.

But what makes you use comments like that? Completely uncalled for and really shows a lack of thinking, at the least.

[IanY]

My belief is that there is a difference in design philosophy between the German equipment and American equipment. The continental Europeans generally do not believe in excessive force, and use a more measured approach. Their water cooling generally uses smaller diameter tubing and somewhat less pumping power, keeping note that Laing pumps are German.

Not to say that there isn't a large contingency of Americans who want absolute silence, but I belong to those who worry a little less about silence than continuous load temps, not to say that HKs are not good at continuous gpu loads. Obviously, the HKs would work better in serial (as in lower continuous load temps), but I gather more HK owners than not do not worry about losing a little performance for elegance.

I'm American and I believe in the brute force approach. It works in my sort of setup, where the optimal flow rates are much higher than those of a German machine. Whether anyone agrees or not, I have four pumps right now in my single loop with four video cards. I need all the performance help I can get when the gpus are loaded fully 24/7. Run 3D Mark Vantage looping 24/7/365 and you'll understand. Therefore, I am a very poor candidate for these HK blocks or the parallel gpu cooling approach.

Regarding the state of the hobby, there was once when performance actually mattered. Now, gltiz and show takes center stage.

[Martinm210]

Wow, awesome setup. My single 8800GTX is whimpering in comparison. Nice test, love the 8 core utilization too!

Thanks for testing!
Martin

[b@llz0r]

<clip>

I was under the impression that 50c fully loaded was a great result for 285s which were at 80+ on air.

What temps do your 285s run under load?

[shazza]

OK ... let me see if I understand the bar chart ...

Starting on the left, it shows lower flow rates for the 3 different CPU water block setups. As you add pumps/tops, you get a higher flow rate. Then you get even higher when you switch from the 480GTX to the PA120.3. The config you are running is the third one ... and that's the one you used to get the CPU/GPU temps you posted. Is this right so far?

I also take it from your comment about flow being overrated, that you didn't test the 1st or 2nd setup (or at least didn't post temp results). So would you expect to get the same performance level on the GPU temps with the lower flow rates? (I think this is what SNiiPE was saying, just not sure).

NOTE -- I'm not questioning your testing or your conclusions, just trying to make sure I understand. I guess I'm confused with the fact that you've shown multiple flow rate setups, yet posted only one set of temps. Maybe the flow rate data was from another test, and you included it here for info?

[IanY]

I was under the impression that 50c fully loaded was a great result for 285s which were at 80+ on air.

What temps do your 285s run under load?

Not an apples and apples comparison.

I am operating with severe constraints. Ease the constraints and I'll show you better temps than that. Alternatively slap my constraints on him and he'll suffer a bit more on the temps.

I can assure you, ceteris paribus, that a serial setup American style with ultra high flow will perform better than a parallel setup German style on an apples to apples basis.

Is the German style parallel setup an acceptable compromise for convenience and appearances? That's up to the respective individual.

Does it work? Of course it does. It works well, just not as well as the American serial style ultra high flow setup, that's all.

I'm not into peeing contests and they bore me.

[b@llz0r]

I'm not into peeing contests and they bore me.

soooooo yours are over 50?

[IanY]

soooooo yours are over 50?

Yeah. tell him to put four in a row in a 50C case ambient (not an open bench) and then tell me what temps he has

Want me to show you 40C ? I'll set my own parameters... hell, I'll show you 35C my way.

Why don't I wait for January to come along, take the sucker outside and brag to you that my gpus load at 5C ?

[Zehnsucht]

Has there been done any serious testing conserning parallel vs serial on three GPU's with varying pump pressure/flow? It would be nice if you, Hondacity, made some testing parallel vs serial. I know this is a lot to ask, but right now one side is saying "yay it works" and the other "nay it doesn't work good enough".

[BDW88]

Yeah. tell him to put four in a row in a 50C case ambient (not an open bench) and then tell me what temps he has

Want me to show you 40C ? I'll set my own parameters... hell, I'll show you 35C my way.

Why don't I wait for January to come along, take the sucker outside and brag to you that my gpus load at 5C ?

So what real world temps would you estimate roughly since your running 4 atm? I'm planning on running the same setup (tri-285's+HK blocks) but with 2xddc's and 2x480's in series dedicated gpu loop

[JOCKTHEGLIDE]

Nice Simple To Understand Thx Again Hc.......

[SNiiPE_DoGG]

Ian, you cant possibly be telling me you think giving an FC GPU block an extra gallon of flow increases your performance very much.... considering that the performance of a parallel we a setup with .5gpm is on par with anyones decent series setup we are clearly on the flatplane of GPU fc block performance > .5gpm

[IanY]

So what real world temps would you estimate roughly since your running 4 atm? I'm planning on running the same setup (tri-285's+HK blocks) but with 2xddc's and 2x480's in series dedicated gpu loop

Never mind me. I'm adding a fifth video card in the next few days, without adding any more radiators. I'm pushing the upper boundaries of the thermal envelope.

With your planned setup, you should be under 48C, perhaps even 45C. It depends on what fans you use and at what voltage. Plus, all important, it depends on your case ambient temp, which I cannot predict. An air conditioned room would be most beneficial to load temps.