PDA

View Full Version : GPU Parallel vs Series



StAndrew
08-03-2009, 05:05 AM
I read over Honda's thread here:

http://www.xtremesystems.org/forums/showthread.php?t=223129&page=5

However, the data/pics are not coming up. Im planning out my next build and plan on running a TRI SLI setup with the Swiftech Obsidian block (9800's). My end loop will be FuZion 2.0 for CPU, 3 Obsidians, a 120 and 240 GTS rad, and MCP655 pump.

Im concerned over flow rate (specially with the GTS rads), however Ill be volt moding the GPUs and will also be concerned about their temps. Can someone offer any more incite of parallel vs series? Honda, can I get a link to those graphs/charts or just some numbers?

Im currently working on a plan to replace my 120 with a 240 but dont think that will happen (tight budget and crpy case). I decided a while ago to make up the lack of rads with more air flow, equipping each with an Ultra Kaze for push and a basic 25mm Silverstone 109CFM for pull. The end result was a huge decrease in temps across the board. I have no doubt that this will hold up a third card (though I also have no doubt a new rad will be even better).

SNiiPE_DoGG
08-03-2009, 05:20 AM
its good you are okay with loud fans because you will need them - my suggestion would be to get an EK x-top v2 for the D5 and it will give you a bit of extra flow in case your lacking. - the worst that can happen is you set it up in parallel and your gpu's are starved for flow so you would have to replumb, it aint so bad :)

veedubfreak
08-03-2009, 05:34 AM
First things first, you wont have enough rad for 3 285s. I run a separate loop for my tri-285s using a swiftech 320 and 655 pump and the rad barely keeps up. Temps are ok on the cards, but the water temps are atrocious.

SNiiPE_DoGG
08-03-2009, 05:45 AM
he doesnt have 285's.... he has 9800's

and those rads with kazes and push pull have so much more power than other rads, seriously he will be FINE

StAndrew
08-03-2009, 06:58 AM
I thought about getting a top. In fact, I think I may. Im sure the rads will be enough so Ill stick with them for now (yes they are loud but Im not aloud to use my speakers anymore -wife- so I use headphones). Im just really curious about parallel vs series flow. With what I read, below is my brain storm:

-Series: Higher resistance but more flow/min over each GPU at any given time. Should result in lower GPU temps do to the higher volume of water/time, however, by the second and third GPU, the water temp will be a bit higher, creating un-even cooling. This can be an issue when voltmoding and overclocking as clocks will be limited by the weakest/hottest GPU (I like to keep clocks in sync with my GPUs, though they can be unsync'ed with EVGA's Precision tool). Also, reduced flow rate can possibly raise CPU temps.

-Parallel: Lower restriction but less flow/min over each GPU. Each GPU should be cooled evenly, however in Honda's thread, there was talk about the middle block not getting the same cooling as the other two. I cant comment on this as I dont know the numbers as mentioned above (anyone help me here?). CPU temps should be relatively better than compared to series.

With that said, these blocks are relatively low in restriction in the first place. Im doing just fine with 2 MC60's so a third (the Obsidian and MCW60 have similar pressure drops) shouldnt make too much of an impact. Using the Flow Rate Estimator sheet my flow should be around 1.2-1.3GPM (rough estimate). Im not entirely sure the differences between parallel and series are so big as to make any noticeable difference. Am I wasting my time or are there real benefits with each setup?

SNiiPE_DoGG
08-03-2009, 07:05 AM
Well what you will partially see by setting up the GPU's in parallel is an overall reduction in loop restriction - the way the d5's PQ works it will be the ideal pump for this setup and I think you will see a nice boost from parallel vs series.

skinnee
08-03-2009, 07:41 AM
Actually, I'm not seeing the D5 and the DDC perform all that different in parallel configuration. I'm still running the PQ tests, but the system flow tests suprised me at how the DDC held its own and came close to the T3 than anything else (updated charts (http://www.xtremesystems.org/forums/showpost.php?p=3939970&postcount=24)).

I'm also trying to wrangle up 3 of the same GPU block to get the NO BS data on parallel gpu blocks...I probably won't make the lifetime of this thread, but I am actively working on it. :)

Hondacity
08-03-2009, 08:07 AM
-Parallel: Lower restriction but less flow/min over each GPU. Each GPU should be cooled evenly, however in Honda's thread, there was talk about the middle block not getting the same cooling as the other two. I cant comment on this as I dont know the numbers as mentioned above (anyone help me here?). CPU temps should be relatively better than compared to series.



the difference was more or less 1 degree of the unit called as celsius :D
but seriously that last stupid gpu was from bfg...its bios or its temp sensors or my mounting could have affected the difference...

i say go parallel if your blocks support it.

eth0s
08-03-2009, 11:40 AM
Hey I love to see that you are using my favorite radiators: the HW Labs GT Stealth! GTS are friggin' sweet (although not the best performing rads mind you). Even though I love your radiators, I was wondering if you couldn't fit a thicker rad in the front of that Armor case. I am assuming that you have the GTS 240 mounted up front in the HDD cage area, and the 120 GTS mounted on the rear 120mm exhaust fan hole. If so, I think you can get a higher performance rad in the front, like a Feser 240, or if you're more budget-constrained, then an XSPC RX 240. Swapping rads may give you a slight boost in cooling performance. Also, you should be able to fit a 360 rad in the front of that case, if you mount your hard drives in the rear HDD cage that is next to the PSU, unless you are rocking more than 3 hdd's and a single optical drive. I would also suggest that maybe you can find room to fit one more GTS 120 in there (like maybe in the bottom of the case pushing hot air out and down), which is the reason I love those GTS rads, you can almost always fit another GTS 120 in there, since it's only 50mm thick, even with a 25 mm fan attached to it. And there is always external mounting, it doesn't look as nice, but it does offer better performance compared to an internally mounted radiator.

As for parallel v. series, I would say go series, as you really need to balance the flow rate between the two parallel sub-loops to get the maximum benefit from a parallel setup (meaning also identical length tubing, identical blocks, identical Y-splitters, etc., and I can't actually think of a way to perfectly balance the flow on a 3-way split off the top of my head, without using a three-way splitter, which I have never seen, and would require a custom fabrication, I think). And to really equalize the pressure in the parallel sub-loops, are you thinking of adding pressure control valves? I think not, so just stick with series, your temps will not likely be affected either way. If you want to increase cooling performance, you really need to add more radiator, fooling around with flow options won't really do much. Although, I don't want to discourage you from an interesting project, if you are so inclined to break new ground, then of course you should pursue it. On the other hand, if you are just looking for a way to cool your OC'd 3 x 9800's SLI'd gpu's, then I would suggest investigating options to get more radiator into the loop first.

StAndrew
08-03-2009, 09:51 PM
I think the GTS actually outperforms those mentioned rads with the fans Im using. As for the setup, you are spot on (HD's already next to PS). Unfortunately, I have two DVD drives and two fan controllers so no 360 for now. Ill see how this setup works and maybe upgrade further down the road when the $$$ isnt so tight. Im sure the temps will be manageable at the least.

As for series, no splitters. Just putting the GPU blocks in series ;).

StAndrew
08-04-2009, 08:34 PM
Ok, I think Ive decided on Parallel. I have two blocks and need to find a third. Question is, how many sets of Swiftech SLI fittings will I need? I presume 2 sets, but I'd like to be sure.

zanzabar
08-04-2009, 08:43 PM
just to make sure, u do know that the obsidian blocks mix metal and instantly corrode, hence u cant get them new

and u would need 8 fittings with 4 sli, 2 caps and 2 real barbs

StAndrew
08-04-2009, 11:31 PM
Really? We're not talking about the Stealth here. Im almost 100% sure the Obsidian has two copper plates (GPU and Memory) and a delrin top.

Im also pretty sure that the popularity of this block (or lack of) is due to the card itself not being hugely popular. I think if this design was utilized with a new a more popular GPU, it could be a big seller.

Anyways, 8 fittings, so two kits. Thanks!

zanzabar
08-04-2009, 11:43 PM
the swiftech site says that its delryn but i could have sworn that it had an aluminum top

zeropluszero
08-05-2009, 12:37 AM
in a parallel loop(2 loops, sharing 1 reservior), with EK blocks would i see better temps for folding with my two EK 280 blocks in series or parallel?
still not really sure about this. i'll also be cooling my 780i with a beepee 780i block in the same loop, cooled with a feser 240. i think the radiator for the two cards and mobo should be fine, but i'm not sure i understand the physics of why the water would choose to go through the block rather than straight to the other block and then out the top again.

zanzabar
08-05-2009, 12:45 AM
watter wants to take the path of least resistance its the same as electricity for this example, so it will go through the part that has the least restriction so it goes down the pipe to the end then it goes through the block then more watter comes and its all there is to much pressure down at the bottom im going to go into the block in the middle then the same for the top. u can do this with electricity just make a 3 leg parallel curcut with some leds or christmas lights