Overclock.net banner

1080Ti SLI GPU Temps Differ by 25 C During Bench-marking

2K views 27 replies 15 participants last post by  mouacyk 
#1 ·
Racking my brain trying to figure this one out and since I've already taken everything apart three times troubleshooting, I'm hoping for some direction and explanation this time around. I'm running the Heaven 4.0 Benchmark.

Case - NZXT H440
Radiators - x2 Black Ice Nemesis 360GTS
Pump - EK-XRES 140 DDC 3.2 PWM Elite
Fans - x7 Thermaltake Riing Premium Plus PWM
EK waterblocks for the i7 7700k CPU and both 1080Ti cards
12mm OD Mayhems borosilicate glass tubing.

Rear fan and top rad is exhaust and front rad is intake.

The pictures speak a thousand words. I just want my cards to have relatively similar temps. I know there will be a difference of about 5C or so, but what is going on here?

I cleaned and re-applied the thermal paste twice. Also took the side panel off as this was trapping heat as well. This actually helped by a few degrees as last night I was actually crashing running Heaven when it hit 70C. The idle temps are about 4C apart from each other at around 27C and 31C. But under load, all hell breaks loose. It was at 39C and 62C at it's highest.



 
See less See more
2
#2 ·
You forgot to mention which card runs hotter. I assume it's the top one.

If that's the case, it's primarily because the pump isn't strong enough to push the liquid from the bottom card's intake to the upper card's intake. All the pressure is going through the bottom block and exiting out the top outlet while very little liquid gets to the top block, again, due to low pressure.

Tilt the computer 45 or so degrees backwards so that the rear I/O is facing your desk/ground, then run a temperature test. The top GPU temperatures should improve, and if that's the case, my theory is correct.
 
#4 ·
I would say the problem is you are running small tube (10mm id i assume) and then trying to tee off into cpu and also parallel sli bridge and aren't getting enough pressure. I would suggest a reconfig going pump - front rad - cpu - top rad - sli bridge - pump. Cool looking rig btw
thumb.gif
 
#6 ·
Without shadow of doubt I will bet my money on needless split in the loop.

Flow is what it is. You can split the flow across identical blocks equally, you can't split the flow across 3 blocks where one is different that remaining 2. One loop. It doesn't matter from which direction fluid will come into the blocks (flow dynamics) top or bottom, but I always strive to use gravity as an ally. Fluid up to the top and the let gravity do the rest.
 
#7 ·
Is your setup 3 parallel? Are you sure the flow through each is adequate? All it takes is for one to be a little more restrictive such as clogged up by some plastic tubing degrading and it won't get much flow at all.
You could make 3 separate with 3 pumps, making it 3 single lines. Or just make at least the GPUs serial but then their restriction will be much higher than the CPU block... There is a reason cooling is done serial not parallel often, the tolerances and flow restrictions vary. Plus is your pump even pushing enough flow for the parallel setup?
 
#8 ·
There are definitely multiple things to address in a loop this complex. This is honestly what drove to give this a go on my own. I actually first saw this design at the Razer store in San Francisco several months back. Once I saw it, I knew I had to try it our for myself. I just didn't want to pay the $6,000 to have it built for me and also wanted the challenge and satisfaction of building it myself.

https://www.maingear.com/custom/desktops/r1/index.php

That all being said, this setup has to work somehow. I highly doubt that a professional modder would design a loop that doesn't work and keep proper temps. I'm very tempted to take the hour drive up there, install MSI afterburner and Heaven, and test for myself. There must be something different in my setup that is elevating the temps.

Quote:
Originally Posted by bazz69 View Post

I would say the problem is you are running small tube (10mm id i assume) and then trying to tee off into cpu and also parallel sli bridge and aren't getting enough pressure. I would suggest a reconfig going pump - front rad - cpu - top rad - sli bridge - pump. Cool looking rig btw
thumb.gif
This crossed my mind as well. I'm using 12mm OD glass which has a thicker inner wall and could perhaps be restricting the flow whereas MAINGEAR is using PETG which has a larger inner diameter. As a last resort, I could always increase everything to 16mm OD glass, but that would involve spending more money on new fittings to accommodate the larger tubing

Quote:
Originally Posted by Alasmodified View Post

I think your pump may be weak and you have split the flow to the gpu and cpu, water will travel faster through the cpu than the gpu's (more restriction). I would just make a single loop rather than splitting it. Maybe feed the gpus firts then the cpu
I'm hoping the pump isn't weak. Looking in more detail at EK's site I saw the following:
https://www.ekwb.com/shop/ek-xres-140-ddc-3-2-pwm-elite-incl-pump
- Maximum head pressure: up to 5.2m
- Maximum flow rate: up to 1000 L/h

This seems like it would be sufficient, right? By the way, thank you for the very clear and detailed drawing. I'm a very visual learner, so this helps. I'm hoping I don't have to chop off half of my loop since the MAINGEAR version of this seems to work. As I mentioned, I won't know for sure until I drive there and test for myself.

Quote:
Originally Posted by Ypsylon View Post

Without shadow of doubt I will bet my money on needless split in the loop.

Flow is what it is. You can split the flow across identical blocks equally, you can't split the flow across 3 blocks where one is different that remaining 2. One loop. It doesn't matter from which direction fluid will come into the blocks (flow dynamics) top or bottom, but I always strive to use gravity as an ally. Fluid up to the top and the let gravity do the rest.
So a possibility is to reverse the whole loop? Basically route the inlet of the pump to the front rad and have everything go the other direction to let gravity help?

Quote:
Originally Posted by JackCY View Post

Is your setup 3 parallel? Are you sure the flow through each is adequate? All it takes is for one to be a little more restrictive such as clogged up by some plastic tubing degrading and it won't get much flow at all.
You could make 3 separate with 3 pumps, making it 3 single lines. Or just make at least the GPUs serial but then their restriction will be much higher than the CPU block... There is a reason cooling is done serial not parallel often, the tolerances and flow restrictions vary. Plus is your pump even pushing enough flow for the parallel setup?
Yes, this is a three parallel setup. The dual GPUs and the CPU since the loop is split before it enters the GPU. The technical specs from EK's website state that this pump is ideal for complex loops with multiple bends and angled fittings.

I'm really hoping there is a way to get this working since there is a working model of this at the Razer store. I also can't seem locate where to change my pump speed. I have it plugged into the AIO header on the Maximus IX Hero motherboard and went into BIOS, but it only shows me 4500 RPM I believe. No way to adjust it.
 
#9 ·
You don't have to reverse everything. You may but there is no requirement to do that. Just pointed out that just for the sake of easing on a single pump a bit its more logical that way - use fluid own weight to do the job instead pushing up via blocks and via front mounted rad. For a single pump 3 blocks that's about maximum you can get away with. Perhaps temp difference with D5 would be smaller as D5 offer higher flow rate, but then head pressure is lower on D5 than on DDC.

And no mate It's not a working triple parallel setup. Well technically it is when you look at it, but its seriously flawed. If you want to split the flow across multiple blocks or rads you can do that ONLY with identical blocks. Then every block will get equal share of the flow. Out of your 3 blocks CPU is least restrictive one. Overwhelming majority of the flow will move via CPU block ignoring GPUs completely or barely pushing anything through there. Its not a rocket science, but its also fluid dynamics basic principle. You have to eliminate the split on the CPU bit. As for going with 10mm or 12mm tubing. Flow rate increase/decrease will be minimal.

What Razer built is example of what not to do in watercooling PC. Sure it looks aggressive and GAMING-blah. But it doesn't change situation that you can still mess up things for xxxx$ (holy cow!!!). Go with dual loop, go with single loop decision is yours, just don't mix blocks of various restrictiveness for parallel configuration. It won't work properly. Ever. Either 3 identical blocks or no deal.
 
#10 ·
I think a couple of people have expressed the same concerned of the split loop. When I was building my loop I had the same predicament but I ended up making it one single loop instead of splitting it and dual pump. You have to remember that the computer you saw was mainly for show not performance, but really think about the physics involved with your loop. Your gpu will never get the cooling capability that the cpu is getting, there is simply more restriction. I think in the coming months I will make a video with dyes to show the restriction of different loop designs.

Here is my loop
 
#11 ·
I'm not an expert in this field. Started water cooling about a year ago so don't put much weight on my post.

What I would try is to confirm that indeed the water is not flowing through the problem card and I see two ways to do that: introduce some bubbles in the loop (though difficult with pastel to see, try to create turbulence in the reservoir by reducing the level at the lowest point without letting the pump run dry), then with a mirror check both blocks; If this doesn't work try putting your hand on each block to confirm that the water is indeed hotter in one than the other. If both are the same then it means that the heat isn't being transferred to the block ergo bad mount/block.
 
#12 ·
Hi there

Something is wrong on yours there, running 3 GPUs(GTX1080Ti and 2*GTX1080) in series and don't have such issues,temps are in same ballpark or around same figure(34-38°C)

Running same pump as you are running EK DDC 3.2 PWM Elite edition and personally have 360mm on top and 240mm on bottom and 120mm on exhaust side

Bottom fans are intake and case fans are intake, top radiator is exhaust and 120mm radiator is set to exhaust air, tried to run fans on top as intake and my water temp increased by 4-6°C

From pictures I see you are using a lot 90° fittings which can slow down flow but without the knowing the what yours flow is hard to say

Are you have any water temperature sensor or do you know what water temperature is it?

Here is my build







Hope this helps

Thanks, Jura
 
#13 ·
Many thanks for the detailed and informative responses so far. What happens if I switch from a parallel setup on the GPUs to serial? Does the rule of not being able to split the flow if the blocks are not identical still apply to serial connections or just parallel?

I understand that there will still be more restriction since there are more blocks to go through than just a direct path like the CPU has, but at least the two GPUs won't be fighting each other for adequate flow like they do with a parallel setup.
 
#14 ·
Also, when I got home from work, I decided to try Furmark and see what happened. Boy did that bottom card get toasty! Topped out 69C! Yep, you read that right, the bottom card. GPU2. When I touched it, it would have burned me if I left my finger there. Does this info change anything? Kinda figured it would be the top card getting less flow.


 
#15 ·
It doesn't really matter which one is suffering. You have to find out why it is overheating. It is because of low flow or another thing?

To get to the bottom of it I would do the following: mount a ball valve between the CPU inlet and the split. This way you can gradually restrict the flow to the CPU and push it to the gpus. You can also mount a soft tube to the cpu inlet from the split and pinch it to get the same effect.

I would keep at it until I would find the culprit.
 
#17 ·
There is some bad info being given in this thread.

EK GPU and CPU blocks are close enough in restriction to be considered the same. The slight differences will only cause single digit percentage difference in flow.

Quote:
Originally Posted by Ypsylon View Post

You don't have to reverse everything. You may but there is no requirement to do that. Just pointed out that just for the sake of easing on a single pump a bit its more logical that way - use fluid own weight to do the job instead pushing up via blocks and via front mounted rad. For a single pump 3 blocks that's about maximum you can get away with. Perhaps temp difference with D5 would be smaller as D5 offer higher flow rate, but then head pressure is lower on D5 than on DDC.

And no mate It's not a working triple parallel setup. Well technically it is when you look at it, but its seriously flawed. If you want to split the flow across multiple blocks or rads you can do that ONLY with identical blocks. Then every block will get equal share of the flow. Out of your 3 blocks CPU is least restrictive one. Overwhelming majority of the flow will move via CPU block ignoring GPUs completely or barely pushing anything through there. Its not a rocket science, but its also fluid dynamics basic principle. You have to eliminate the split on the CPU bit. As for going with 10mm or 12mm tubing. Flow rate increase/decrease will be minimal.

What Razer built is example of what not to do in watercooling PC. Sure it looks aggressive and GAMING-blah. But it doesn't change situation that you can still **** up things for xxxx$ (holy cow!!!). Go with dual loop, go with single loop decision is yours, just don't mix blocks of various restrictiveness for parallel configuration. It won't work properly. Ever. Either 3 identical blocks or no deal.
In closed loops there is no up and down. Moving the coolant in one direction or the other makes precisely zero difference. The pump does not lift fluid and fluid does not fall on the other side because it is all counterbalanced. What the pump loses in head pressure on the vertical rise sections it gains back as inlet pressure from all the vertical fall sections. A loop has zero net height change, so there is zero head pressure lost.

In parallel loops there is rarely significant difference in the flow rate of each branch.
Flow is determined by restriction and restriction increases with the square of flow. For one branch to have twice the flow of the other it must be only 1/4 the restriction. Even if one path is 1/2 the restriction it only gets 50% more flow.

Recent testing of EK Blocks shows them to be only small percentage difference in restriction. These days it is rare for even blocks of different brands to be twice as restrictive.

Having said that the gts rads being used are the worst choice for this loop config. A full parallel system relies on high flow rates through the rest of the system (pump and rads here) but the gts rads are very restrictive and will by themselves limit the Max flow rate of the system, which then has to be split three ways. With two GTS rads in series the system flow rate could not be any higher than 1.75GPM and is likely a bit less when accounting for all the tubing and fittings. Each block can only get 0.5GPM at most. Different rads would significantly bump the system flow rate that is possible.
The relatively narrow 9mm id tubing will also restrict the system flow rate in the series sections of the loop. For the very high flow rates desirable for parallel systems wider tube is better.
The high flow rate needed for a triple parallel system line is also not friendly to DDC pumps. The needed flow rates are where they over heat.

This loop isn't truly triple path parallel, it's dual parallel with one branch having it's own dual parallel section.
The coolant gets to the first T and has two paths, the CPU block and the parallel GPU blocks. The parallel GPU blocks are roughly 1/4 the restriction of the CPU block so they get about twice the flow rate. They then split that between them. So the effect is roughly the same without including the extra pipe and fittings.

As Costa has said above the problem is not the setup but likely a contact issue or pastel fluid clog. Clogging that may have been made more likely by the relatively low flow being imposed by the GTS rads.
 
#18 ·
Quote:
Originally Posted by Costas View Post

When running in parallel - The fact that one GPU is running way hotter indicates a restriction only in that GPU's path.

Even if flow is ridiculously low into and through both gpu's - both GPU's should always still have roughly the same temps.

Either one of the GPU blocks is blocked or is not mounted correctly onto the GPU card.
Hi guys, late reply since I was in Denver for a week on vacation. Got back today and decided to completely take everything apart and install the GPUs in my old loop. It was a ton of work, and it's sloppy since I'm just troubleshooting, but I believe the results will help narrow down the issue.

First some specs on the old rig.

XSPC Raystorm CPU Waterblock
Swiftech MCP35X-BK 12 VDC Pump
XSPC RX360 v2



So the pump is different, the radiator is a full-size and not as restrictive. The loop is also very basic and has wider tubes with plain water instead of pastel fluid. Guess what? Exact same temperature differences. So, I'm tired now and don't know if swapping the GPU blocks with each other will make a difference, or remounting it will solve the problem, but it appears that the intricate loop design, the DDC pump, the slim restrictive rads and the pastel coolant have nothing to do with the issue.



I do know that this old loop and everything in it was perfectly able to adequately cool two 680 GPUs in SLI within a few degrees of each other. I'm open to any other suggestions.
 
#19 ·
Ek block tops look to be clear acrylic? When you pulled the loop apart and rebuilt it did you examine the condition of the gpu blocks before you built your test loop?

where they both in good condition? any gunk in the fins or build up along the acrylic?

Either there is a blockage in one of the water blocks internal flow path or you have a mounting issue.

Edit: check sli bridge openings as well as sli bridge adaptors where they are connected to the blocks

edit 2: Your original setup looks absolutely wicked btw, love the triple parallel setup. EK is probably the only blocks you could do this on. very well put together, congrats on a job well done!
 
#20 ·
Quote:
Originally Posted by djfunz View Post

Hi guys, late reply since I was in Denver for a week on vacation. Got back today and decided to completely take everything apart and install the GPUs in my old loop. It was a ton of work, and it's sloppy since I'm just troubleshooting, but I believe the results will help narrow down the issue.
First some specs on the old rig.

XSPC Raystorm CPU Waterblock
Swiftech MCP35X-BK 12 VDC Pump
XSPC RX360 v2



So the pump is different, the radiator is a full-size and not as restrictive. The loop is also very basic and has wider tubes with plain water instead of pastel fluid. Guess what? Exact same temperature differences. So, I'm tired now and don't know if swapping the GPU blocks with each other will make a difference, or remounting it will solve the problem, but it appears that the intricate loop design, the DDC pump, the slim restrictive rads and the pastel coolant have nothing to do with the issue.

I do know that this old loop and everything in it was perfectly able to adequately cool two 680 GPUs in SLI within a few degrees of each other. I'm open to any other suggestions.
That's to be expected because its a problem with the block. Even with the relatively low flow in the other loop layout its still plenty for decent results when everything else is hunky dory. GPU blocks are quite tolerant of low flow rates and it would normally only make a couple of degrees difference compared to a 1Gpm flow that is the often recommended target.
 
#21 ·
Quote:
Originally Posted by Ashcroft View Post

That's to be expected because its a problem with the block. Even with the relatively low flow in the other loop layout its still plenty for decent results when everything else is hunky dory. GPU blocks are quite tolerant of low flow rates and it would normally only make a couple of degrees difference compared to a 1Gpm flow that is the often recommended target.
I've concluded it's the block as well. I flipped the orientation from top card to bottom card in the SLI setup and it's the same result. Just that card is overheating.



I even took the block off and mounted a third time just to be sure.



I've reached out to Duke at Performance PCs for an RMA. I'm hoping this can be resolved quickly as this has been a nightmare to say the least. I'll keep this thread updated until this is resolved.
 
#23 ·
I had a similar temperature differential between 2 SLI GPU about 20 months ago. My setup is in series though.

In my instance the Waterblock on the "hot" GPU was not making level contact with the GPU. When I dissambled the waterblock/GPU all the thermal grease was being squeezed to one side of the GPU. One side of the GPU had good contact with the waterblock and the other side had none.

I had to replace the waterblock.
 
#25 ·
Quote:
Originally Posted by djfunz View Post

I even took the block off and mounted a third time just to be sure.



I've reached out to Duke at Performance PCs for an RMA. I'm hoping this can be resolved quickly as this has been a nightmare to say the least. I'll keep this thread updated until this is resolved.
Could the thermal pad on the chokes be offsetting the height? The chokes/inductors typically don't need to transfer heat. If it is done, it's to prevent coil whine.
 
#26 ·
Shot on the dark form a watercooling noob:
My GPU block (EK-FC1080 GTX Ti TF6) came with two sets of screws with the same threading, but different lengths. The manual doesn't clearly state this, but the longer screws are for reattaching the stock backplate and the shorter ones are to be used with the rubber o-rings and no backplate. In either case, the shorter screws were to be used on the four points closest to either corner of the die. If your block is anything like mine, a few strategically (erroneously) placed long screws could be preventing proper contact between the die an the block.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top