Production Winter One -- 15.6L SFF case, 3090 Support, 3-slot GPUs, dual 280mm radiators, CFD Optimized Design

WinterCharm

Master of Cramming
Original poster
Jan 19, 2019
404
1,779
Just a point of data for anyone else who is watercooling with 280s. I was having a very hard time fitting 2 HWL GTS 280mms in there. The ends of the radiators were touching/pushing against the side panels (probably had something to do with the tolerances I have seen WinterCharm talk about). I was also not able to get the radiator mounting rails completely on the radiators due to the screw holes on one end of the radiator being slightly outside where the holes in the mounting rails ended. I had a Corsair 280mm rad from a previous build and was able to get that to fit with a few mms to spare. Not sure if this was just due to weird tolerances on the HWL rads, but just another data point of a 280mm rad that fits in the case.
My 280s fit and all 8 screws lined up, but they are touching front and back panels... afaik, the corsair (and bitspower) rad is an L series.

Despite the "Real" clearance in Winter One being 214mm and HWL's listed dimensions being 212mm, the GTS 280 has considerable variance, due to its very loose tolerances. It's quite annoying as this can sometimes lead to a bad fit. However, based on what you described with the screw holes on one side not lining up, @doot4runner it may be that you got a bad unit and should see if you can return that one / RMA it.

I know this is a silly question to ask the guy who designed this case, but do you have any concrete data of real-world application with loops in different configurations to be able to confidently say that one configuration performs better than others?

Yes. Beyond a reasonable doubt, based on working knowledge + really good models, and simulation data from the Beta Program. There is a real efficiency increase in cooling if you are using counter-current flow when you have the same air going through multiple radiators in a case (only applies to solid panel setups). It comes from maintaining a better ∆T between Air and Water through the entire loop.

In real-life ∆T is constantly changing at each point across the loop. To measure it we use Log-Mean ∆T (also referred to as LMTD)which measures the effective ∆T between the hot side in/out temps and the cold-side in/out temps.The math for LMTD can be found here but it's not worth getting into, because a picture works much better to demonstrate why you'd want to do this:

For any given inlet / outlet temps on the hot/cold side, there is a difference in LMTD, as you can see from the image below.

Operating the same loop with the same inlet outlet temps, you can see an effective difference of an additional 13ºC of ∆T.

In PC watercooling loops, at high enough flow speed, you are only looking at about a 4-8ºC drop in water T across 2 radiators, based on your heat load and flow rate. I went ahead and did some calculations based on values found in @Qzrx 's loop from the Beta Program.

SOLID PANEL BOTTOM >> TOP LOOP
Hot Water In: 46ºC
Warm Water Out: 42ºC
Cool Air in:
23ºC
Warm Air Out: 39ºC
(Air-Water) LMTD for countercurrent flow: 12ºC
(Air-Water) LMTD for co-currnet flow: 9.8ºC

PERF PANEL ALL EXHAUST LOOP
Hot Water In: 44
Hot Water out: 42
Cool Air In: 23
Cool Air Out: 32
(Air-Water) LMTD for
perpendicular flow: 15ºC

Or put it this way: you could potentially dump more heat or take that extra difference between the two temps and reduce your Coolant-CPU ∆T by that much (so about 2.2ºC which could be quite nice, in a sequential radiator setup, IF you have countercurrent flow).

----------------------

This data confirms most of what I found, independently and by another person testing his loop, and evaluating radiator performance in various configurations. The only mistake they made, IMO, is they did not separate the two radiators in configuration 6 (the closest to our countercurrent flow), which does restrict flow considerably more than separating those fan / radiator pairs, so the performance delta you'd expect to see is much larger than the data I have shows.
 
Last edited:

mxj1

Cable-Tie Ninja
Sep 13, 2020
176
434
Thanks for the detailed post.

I've read about the benefits of stacking radiators previously, but never found myself in a position to try it out. With that, I understand that the main point is that coolant should exit the stack at the radiator with the coldest air... but I can't see how stacking radiators applies to this scenario, unless solid panels allow the case to become air tight.

I'll admit, some of this may go over my head, so you'll have to be patient with me, but my days of pre-itx custom loops taught me a few things that are confusing to me with the recommendation on WinterOne loop order:

1. Flow rate is important and can be determined by using >1 temp sensor at different places in the loop and tuning for equilibrium.

2. Loop order doesn't matter. When flow rate is adequate, coolant temperature reaches equilibrium and would be within a margin of error at any two points in the loop. I confirmed this using calitemp sensors on my aquaero 6 lt (I've also used an infrared thermometer to measure radiator end tanks). the coolant exiting the radiators was typically ~0.1-0.4* cooler than the coolant entering the radiators, after the heat sources.

Let's say there's ~600w of heat to be dissipated by the loop. Once the coolant reaches equilibrium, that heat should be dissipated equally among the radiators, with the exception being that the top radiator has much warmer air than the bottom radiator. If we change the order of the components in the loop, there is still a 600w load on the system, so why would there be any change in performance?

In your test scenarios were either of you monitoring flow and coolant temperature at >1 location?
 
  • Like
Reactions: WinterCharm

WinterCharm

Master of Cramming
Original poster
Jan 19, 2019
404
1,779
but I can't see how stacking radiators applies to this scenario, unless solid panels allow the case to become air tight.

That's exactly it. This loop order being discussed is for Solid panels.


Loop order doesn't matter. When flow rate is adequate, coolant temperature reaches equilibrium and would be within a margin of error at any two points in the loop. I confirmed this using calitemp sensors on my aquaero 6 lt (I've also used an infrared thermometer to measure radiator end tanks). the coolant exiting the radiators was typically ~0.1-0.4* cooler than the coolant entering the radiators, after the heat sources.

If your loop isn't set up in a way that radiators stacked are coupled and airflow path is sequential through radiators, then It won't matter. This Is why the suggested loop order for perf panels is to just have parts in between each radiator, since each radiator gets a fresh supply of room air.

It's only in the solid panel setup, where the ∆T can be as high as 2-3C across a countercurrent flow loop. that it begins to matter.

Once the coolant reaches equilibrium, that heat should be dissipated equally among the radiators, with the exception being that the top radiator has much warmer air than the bottom radiator. If we change the order of the components in the loop, there is still a 600w load on the system, so why would there be any change in performance?

So that's the problem The top radiator can't dissipate as much heat as the bottom one, because it's receiving warmer air, and the amount of heat transfer depends on U*A*∆T

U is coolant flow. That's not changing between either radiator. A is area of the fins on the radiator. That doesn't change. But ∆T is changing, because the air entering the second radiator is *warmer*. Therefore, the second radiator will dump less heat **until it heats up a few degrees more*, and the coolant flow at that point will start to warm up until it has enough ∆T to dump half of the 600W, while the coolant flow will remain cool at the bottom radiator.

Your observations are not wrong for a radiator setup with independent airflow to each rad. Such a setup will keep coolant temp very very close to level throughout the entire loop. But once you make it so airflow is going from one radiator to another, the system changes significantly, and these temperature effects can be observed.
 

bomaaye

Caliper Novice
Dec 15, 2020
27
19
1. Flow rate is important and can be determined by using >1 temp sensor at different places in the loop and tuning for equilibrium.
So for water-cooling it pretty much boils down to heat transfer laws, which just makes it a lot easier to explain since its all math and physics. I'm not sure how far you want me to go into it so I'll try to keep it relatively simple to follow.

1) So flowrate is important, and increasing it also raises the total heat the water can soak up, up to a certain point. The simple form of the heat transfer formula is:

Q = m * Cp * dT

where Q is the amount of heat the liquid can soak( in Watts), m is the mass flowrate( kg/second), Cp is the heat capacity of the fluid ( assume water here for simplicity, so 4181J/kg. C ) and dT is the temperature difference (which you can do for different parts of the loop so you can better analyse it all).

Lets look at the CPU block as an example assuming the CPU dumps 200W of heat to the loop. If we look at the equation just in terms of the coolant entering and leaving the block (so dT = coolant Out - Coolant In)
Now say we want to get better heat removal by the coolant, we could do this in 2 ways, increase flowrate (m) or increase dT, Cp is constant. In this case we want to see how flowrate changes heat load i.e. cooling performance so I'll discuss this.

-Higher flowrate from the formula above suggests that dT would decrease since Q is 200W, and Cp is constant. Lower dT here would mean that the coolant exiting the CPU block is a bit cooler here, which means slightly better performance..

However, the one area we have missed here is that when you increase the flowrate of coolant, you're running the pump at a higher rpm, which adds a certain amount of heat into the loop. Faster rpm means more heat is dumped into the loop over time. Now, because of this and the CPU always dumping 200W of power into the loop, it becomes easier to see that as we increase flowrate, the coolant entering the CPU block will be slightly hotter which eats away at cooling performance. Referring back to dT = Coolant Out - Coolant In, we can see dT would decrease ever so slightly every time you increase flow.
Eventually you would reach a point where the heat generated from the pump pretty much "cancels out" the gains you get from increasing the flowrate, which is what you saw when you tested it with your setup! Hopefully this made some sort of sense, if more info is needed or if I wasnt clear let me know!
Let's say there's ~600w of heat to be dissipated by the loop. Once the coolant reaches equilibrium, that heat should be dissipated equally among the radiators, with the exception being that the top radiator has much warmer air than the bottom radiator. If we change the order of the components in the loop, there is still a 600w load on the system, so why would there be any change in performance?

In your test scenarios were either of you monitoring flow and coolant temperature at >1 location?
Kind of. The radiators operate based on the delta between the water inside, and the air coming through it. The top radiator is fed hotter air, so dT is lower, which means its cooling performance would be lower than the bottom rad( both identical radiators so their fin areas are identical) because the bottom rad gets colder air (dT is higher).
 

WinterCharm

Master of Cramming
Original poster
Jan 19, 2019
404
1,779
Update #19 is Live!!! - More cases ship later this week, Standoff Issue & Nylon Spacer Fix.

--------------------------------


Hello Everyone!

We’re in the middle of our busiest week ever, so apologies for the late update. We’ve got a good bit to talk about, so let’s get into it!

Quick Summary​

  • 33 cases ship later this week
  • More parts arrived Tuesday (awaiting QC)
  • 23 cases ship next week
  • more parts arrive next week, and the remainder of Batch 1 should be en route in 2 weeks time.
  • There was an issue with some motherboards and the standoffs A fix has been created. Fixes are being sent with all new case shipments, and given to anyone from shipment 1 that requests them
  • Website 2.0 is nearly complete (expect a switchover in 1-2 days!), and with it, Batch 2 countdown begins!

Standoff Issue​

Issue Summary:
We’ve noticed that a few motherboards may slip onto the standoffs, rather than sitting directly on top of them, causing a misalignment of the I/O bracket with the rear panel.

Details:
The mITX spec lists 4mm as the correct spec for mounting hole size on motherboards, and we used a 4.2mm standoff diameter, to maximize backplate compatibility, as Backplates have gotten larger on motherboards.
However, at the extreme edges of both tolerances, there's a chance that the motherboard hole is wide enough that the standoff can poke all the way through, causing a misalignment of the rear I/O. This issue did not surface in any of the Beta Testing units, and we were unaware of it until after the first set of cases had shipped.



Nylon Spacer Fix​

For cases that have shipped:
We're sending anyone who has this problem a set of height-matched Nylon spacers that fit around the standoffs, allowing the motherboard to sit at the correct height on the standoffs.

For cases that haven't shipped:
We're going to be including them in the pack of screws / fasteners that come inside the box. Most people won’t need them, but if you do they’ll be within reach. This issue affects Batches 1 & 2 of the Kickstarter, as all central spines for existing orders have been manufactured.

What if I encounter it later?
For those of you in the first shipment, if you don’t have the issue right now, but encounter it in the future, contact our support email and we’ll get you the spacers.

Permanent Fix
In future batches of Winter One, we will be widening the standoffs to 5-6 mm OD in order to permanently correct this issue.

 

Koxx5D

SFF Lingo Aficionado
Oct 26, 2020
110
98
The l series is also more restrictive than the gts, so keep that in mind when planning pump type.
On this subject (restriction), can you please tell us what it's in your settings? ( 2x 280 HWL "L" + Optimus + 90 angles) I am interested.
 

Koxx5D

SFF Lingo Aficionado
Oct 26, 2020
110
98
I have a stupid question to ask, so don't blame me too much; what would be the cooling performance if we used 2 XSPC TX240 with a push pull of A12x15? (Without taking into consideration the notion of noise) This with a case with solid panels.

I have the 2 280 GTS and I will not change, but I wonder about it :mad:

It becomes very very hard to wait for the W1 :)
 

mxj1

Cable-Tie Ninja
Sep 13, 2020
176
434
On this subject (restriction), can you please tell us what it's in your settings? ( 2x 280 HWL "L" + Optimus + 90 angles) I am interested.

My configuration is:

2x GTS 280
Optimus Sig V2
Heatkiller 3090 reference
Swiftech MCP35X w/ alphacool brass top
~16 90* angles/fittings
2 QDCs for Mora
High Flow Next

External:
Mora 420
Heatkiller tube 200 w/ D5 Next

2 pumps at 100% provide ~230 l/h, but noise is unbearable, and no additional performance is gained. I usually operate both pumps at ~40% which yields ~140 l/h

Push/pull on thin radiators will not be beneficial. That's better suited for thicker, and high(er) fpi. 2x gts and good static pressure fans is the way to go.
 
  • Like
Reactions: WinterCharm

Koxx5D

SFF Lingo Aficionado
Oct 26, 2020
110
98
My configuration is:

2x GTS 280
Optimus Sig V2
Heatkiller 3090 reference
Swiftech MCP35X w/ alphacool brass top
~16 90* angles/fittings
2 QDCs for Mora
High Flow Next

External:
Mora 420
Heatkiller tube 200 w/ D5 Next

2 pumps at 100% provide ~230 l/h, but noise is unbearable, and no additional performance is gained. I usually operate both pumps at ~40% which yields ~140 l/h

Push/pull on thin radiators will not be beneficial. That's better suited for thicker, and high(er) fpi. 2x gts and good static pressure fans is the way to go.
Thx :) even if it's a bit extreme for me; how do you get the pass through?
 
  • Like
Reactions: WinterCharm

WinterCharm

Master of Cramming
Original poster
Jan 19, 2019
404
1,779
I have a stupid question to ask, so don't blame me too much; what would be the cooling performance if we used 2 XSPC TX240 with a push pull of A12x15? (Without taking into consideration the notion of noise) This with a case with solid panels.

I have the 2 280 GTS and I will not change, but I wonder about it :mad:

It becomes very very hard to wait for the W1 :)
Push-pull performs worse. See data Here. On a slim radiator you are getting less flow restriction so a second fan doesn’t help much. And you’re giving up area to add a fan.

TX240 is only 240mm and slimmer. A GTS 280 is only 8% away from a 360mm radiator! It’s that much better.
 
  • Like
Reactions: Koxx5D

Qzrx

Buried under radiators
Silver Supporter
Bronze Supporter
Dec 29, 2019
83
210
Would the EK-AIO 280 D-RGB fit in this case? It’s stated to be 313mm long.
With some sanding and squeezing, possibly. The bigger question (har har har) is what are EK's manufacturing tolerances? At 313mm you could likely squeeze it in if you don't mind voiding your warranty with some sanding, but if the tolerances are anything like HWLabs' you could wind up with a 316mm radiator that isn't going to fit despite your best efforts.
 

Ricky

Average Stuffer
Jan 23, 2016
86
35
With some sanding and squeezing, possibly. The bigger question (har har har) is what are EK's manufacturing tolerances? At 313mm you could likely squeeze it in if you don't mind voiding your warranty with some sanding, but if the tolerances are anything like HWLabs' you could wind up with a 316mm radiator that isn't going to fit despite your best efforts.
Well that’s disappointing. Most 280mm AIO’s seem around 313-315mm.
 
Last edited:

Qzrx

Buried under radiators
Silver Supporter
Bronze Supporter
Dec 29, 2019
83
210
Well that’s disappointing. Most 280mm AIO’s seem around 313-315mm.
The supported length is 312mm but the actual dimensions are 314mm...like I said: case tolerances for W1 are _tight_ but radiator ones are usually not so it’s a coin flip on whether any individual 313-314mm radiator will actually fit.
 
  • Like
Reactions: Ricky

Ricky

Average Stuffer
Jan 23, 2016
86
35
The supported length is 312mm but the actual dimensions are 314mm...like I said: case tolerances for W1 are _tight_ but radiator ones are usually not so it’s a coin flip on whether any individual 313-314mm radiator will actually fit.
Definitely seems like a risky venture, I’d rather not take it.