Stalled H2O-Micro (sub 3 Liter Custom Watercooled Gaming Rig)

[QinX]

Hey all, just a quick Easter update!

Please read until the end, I need your opinions on something

I’m still working out the costs of the DIY kit, still figuring out ways to reduce costs.

In the mean time I received the anodised cases back and I transplanted the entire system over to it.
Here are some quick snapshots

Photo of the Front


Photo of the back


Photo of my attempt at a closeup to show the finish.


I personally really love the finish I’ve gone for, it’s not quite the silver finish that you find on most cases it has a little more grey mixed it.

Now as for progress on the DIY kit.

I’ve mentioned before that I am considering to support the Gigabyte GA-H81-TN over the original ASUS Q87T I used in my build. My reasoning is that besides the lack of 1 Ethernet port, who needs 2 anyway? And 2 of the 4 USB ports being 2.0 there is not a whole lot of difference between the boards.

Did I mention it costs almost half of the Q87T?

They both have
mSATA
Half-length mini-PCIe
PCIe 3.0 4x
4790K support
16GB SODIMM support
2 Fan header

You could even argue the GA-H81-TN has better thermals because the heatsink on the PCH is bigger.

Here is a side-by-side

Photo of Top of Motherboards


Photo of Motherboard I/O


Would you guys be okay with the GA-H81-TN?

Next is the thing I am very eager to hear you opinions about.
So right now I am running on a 20V 350W Gigantic external brick. I think it is bulky and you all should feel the same .

Recently ASUS has released the ASUS ROG G20 and it has 2 external bricks. Why? Most likely 2 smaller bricks are cheaper than 1 big brick. They made a bracket to hold them.

I want to switch to a 2 brick design for 2 reasons.
Reason 1: Costs, by going for 2 seperate bricks I can reduce the costs of these bricks
Supply and demand is the cause of this. More manufacturers and consumers buy lower power AC adapters so more manufacturers make them and prices go down because of that.

Reason 2: Space/Cost I can probably cut the HD-Plex 250W DC-DC board by going this route. I can use a power brick that supplies 12V to the GPU, thus giving me the option to make the system more versatile, maybe I can squeeze in a 2,5” or at least a 1,8” drive mount, allowing some form of HDD storage besides the mSATA drive. Also the DC board is expensive for what is does in this system, it is a glorified 20V to 12V adapter, so why not go from 230/115VAC to 12VDC directly?

PS: I will make sure I have a nice bracket to discretly mount the 2 power bricks.

Please let me know what you guys think about this.
I’ve made a strawpoll, but if you feel the need to voice your opinion please do.
http://strawpoll.me/4032713

That is all for now have a nice Easter if you celebrate!

 

[PlayfulPhoenix]

I love the finish too - it's not overly metallic, and it has a very nice matte texture with just a hint of graininess in the metal.
Anyway, I can't speak as someone who's be interested in buying the kit, but I think your reasoning for both the bricks and the motherboard change is sound.

 

[QinX]

I've been getting more and more question about the final price for the kit, so I guess I'll go ahead and release an estimate.

I'm still waiting on some small bits and pieces for final pricing, but the biggest wait is still the case. Also some other parts I haven't been able to calculate yet, because I still have to design them, such as the GPU power board.
If I'd have to give an estimate I would say somewhere between €350 and €450 would be end user pricing.

It will be an expensive kit, I realize that, but I think the pricing is justified looking at what the market has to offer and what you actually get.
What do you actually get for that kind of money?

Preliminary Box of Contents
1x 2mm anodized Aluminum Chassis(With Horizontal and Vertical brackets)
1x Dual 92mm Radiator
2x 92*92*15mm Fans
1x Copper/Acetal Socket 1150 CPU Waterblock
1x Copper/Acetal GTX970 GPU Waterblock
1x Acetal Reservoir
1x 12VDC Pump
1x 12VDC 180W AC Adapter for GPU
1x 19VDC 120W AC Adapter for System(CPU, Motherboard etc.)
1x AC Adapter mounting bracket
1x GPU Power Board
Miscellaneous accessories (Fittings, Tubing, Screws, Allen keys, Manual and all other things needed to build a system)

What else do you need to buy to build a working system?

Shopping list:
Motherboard: Gigabyte GA-H81TN, GA-B85TN, GA-H87TN, GA-Q87TN
CPU: Any S1150 Haswell CPU up to the i7-4790K
DRAM: SODIMM DDR3 or DDR3L, up to 16GB
SSD: Any mSATA SSD
GPU: GTX970 with short PCB and Single Slot I/O


Here in the Netherlands I can build the following system including the kit for around €1550:
CPU: Intel Core i5-4690 Boxed
Motherboard: Gigabyte GA-B85TN
GPU: Gainward GeForce GTX 970 4GB
RAM: Kingston KVR16S11K2/16 (2x8GB)
SSD: Crucial M550 mSATA 512GB
Wifi: Intel Dual Band Wireless-AC 7260 Plus Bluetooth
OS: Windows 8.1 64-Bit OEM

 

[QinX]

A while back I mentioned some results I had gotten with my setup but I've decided to do a quick run with some games included so here we go.

Firstly I ran an Idle test, nothing spectacular, I was streaming Youtube at the same time.

Ambient was 21C, with the radiator temperature being 30C so about 9C higher than ambient. Fans where at their minimum speed.
Power draw at the wall was about 48 Watts.

Next up we have Prime95 on 3 Cores with Furmark on 1 Core, this proved to be the most stressfull scenario I could create and is by no means even remotely comparable to a normal load.

Ambient was 21C, with the radiator temperature being 40C so about 19C higher than ambient, Fans where at full blast.
Power draw at the wall was about 285 Watts.
HWInfo64 results seem a little vague, de maximums for CPU temperature are hitting 99C, but the average is 88C, and looking at the current temperatures it jumps from 88C to 97C for 1 second and goes back down.
I this image you can also see how I had my probes for measuring temperatures arranged.

Next up I tested 2 games, Skyrim with +- 200 mods and Far Cry 3.

Ambient was 21C, with the radiator temperature being 47C so about 26C higher than ambient, fans where around 70%. Here can see that H2O-Micro can be small and quiet when playing games.
Power draw at the wall was about 233 Watts.
So with only 50 Watts less power consumed than the Prime95/Furmark run and the radiator being 5C hotter, temperatures are excellent on both the CPU(76C MAX) and GPU(73C MAX)



Ambient was 20C, with the radiator temperature being 46C so about 26C higher than ambient. Fans where at around 70% speed.
Power draw at the wall was about 207 Watts.
Again excellent temperatures and fan speeds while playing a game.


I've also received some samples this week for the kit to check out and test.
Among them a set of fans most of you will recognize instantly.

But first! I orderded some Anti-vandal switches, Both are the same model and have a white led but 1 is a Dot and the other a Ring, let me know which you prefer. I already have a preference, can you tell?

Switches in Off state


Switches in On state



I modeled the switch in CAD and pooped out a render.


And finally I was sent 2 Noctua NF-A9x14 PWM fans by Noctua. These are actually my first Noctua's and I'm looking forward to testing them. I will be comparing them to some other manufacturers but they haven't arrived yet.






Some of my models I download from a website called www.grabcad.com and someone on there already made some Noctua models, but not of the Noctua NF-A9x14 PWM.
So I went ahead and made a CAD model and gave it a lick of paint in 3dsmax. I have to say I'm pretty proud of this model, it's 99% accurate.


That was it for now.

 

[jeshikat]

I much prefer the ring LED as well

 

[PlayfulPhoenix]

Awesome update and testing And I concur, the ring looks a lot nicer to me.

 

[QinX]

Most people seem to prefer the LED Ring.
I also spent some time today rendering a short animation of the fans.

 

[QinX]

I've been reading up on the new AMD GPUs and the HBM tech. Anandtech had some interesting information regaring the GPU Package.

Finally, aftermarket enthusiasts may or may not enjoy one final benefit from the use of HBM. Because the DRAM and GPU are now on the same package, AMD is going to be capping the package with an integrated heat spreader (IHS) to compensate for any differences in height between the HBM stacks and GPU die, to protect the HBM stacks, and to supply the HBM stacks with sufficient cooling.

After reading this, it seems like the GPU waterblocks for the new AMD cards could be made in a very simple a cheap way.

If I were guessing we might go back to GPU core only waterblocks, perhaps with a little wider base to allow attachment of a heatpipe for VRM cooling, the same way that those Streacom cases do it. What do you guys think?

 

[EdZ]

Depending on where the VRMs are located, it may still be easier to make a slightly larger waterblock than trying to bend a heatpipe from the block over all the VRMs (which may need another custom heatspreader depending on how spread out they are).

I am a little worried about the on-package heatspreader. The HMB stacks are taller than the GPU die (only by a few mm at most) so it's going to need a pretty tight tolerance to have decent contact on the GPU without crushing the HMB stacks, unless they go for flood-the-gap-with-thermal-compound-and-hope-for-the-best.

 

[Phuncz]

unless they go for flood-the-gap-with-thermal-compound-and-hope-for-the-best.

Now you wouldn't be referring to Intel's abundant goop-splurging, would you ?

I hope they'll use high quality thermal pads, it's in their best interest to use proper contact to avoid the core throttling like a random plastic octocore smartphone.

 

[jeshikat]

Now you wouldn't be referring to Intel's abundant goop-splurging, would you ?

It's bad enough on a 84W TDP part, I can't imagine the temps you'd get on a 250W+ GPU if AMD just used cheap unsuitable paste like Intel did on Haswell.

 

[WiSK]

It's not that Intel uses "cheap" paste. Someone tested it on Anand and it performed adequately. The problem is that emulsion based paste just doesn't wet to the glass die like solder does. The liquid metal TIMs do wet to glass, so that's why there is such a striking improvement in temperature transference for people who delid their Ivy/Haswell CPUs.

 

[jeshikat]

Ah, didn't realize that. So it's just the wrong stuff for this application, like when Nvidia used the wrong solder on all those mobile GPUs that failed several years ago.

 

[WiSK]

I'm sure the reason they did it is more complex than we know. Ultimately down to a financial decision no doubt, but I would guess that it's not as simple as "paste is cheaper than solder". There might have been fabrication issues leading to a significantly lower yield. Or similar.

 

[EdZ]

One interesting theory - that seems to be borne out by testing - is that the TIM itself is fine, and application of the TIM is fine, but that the process of replacing the TIM is effective because the adhesive that attaches the IHS to the package is too thick preventing proper contact. i.e. It's a package assembly issue, the individual components are all OK. That's why the variable-height IHS for a HBM+GPU package is a worry, because it would be very easy for assembly tolerance (to prevent crushing the HBM modules) to result in poor contact even with a good choice of IHS design and TIM.

 

[WiSK]

The shim theory, it's part of the same thread I referenced before. It's quite convincing at first...

The guy "Idontcare" who was doing the shim testing never tried CLU because IIRC he'd had a bad experience with Indigo Extreme hardening and sticking his cooler to his IHS. He initially assumed that if you used gallium under the heatspreader, it would permanently stick the IHS to the chip. Even though it's known that gallium is not absorbed into glass. So his shim tests are with NT-H1 which is known as a good but not outstanding TIM.

Even 2 months after people at Xtremesystems were recommending CLU, IDC had still not tried it and claimed in bold "I am now firmly convinced the sole reason Ivy has elevated temperatures is because of the gap that exists between the CPU and the IHS". After another month, he finally tests CLU along with several other TIMs. He doesn't bother with the IHS any more and tests bare die. Surprisingly CLU gets almost exactly the same results as the NT-H1, despite a theoretical difference of factor 3-4 in thermal transference. Other pastes that perform better than NT-H1 in published reviews are disappointing in IDC's review. He also complains about the damage CLU did to his chip. Despite it, suddenly he is recommending CLU when people ask. Then I started to have doubts.

Well I've never done any testing like that myself, so I can't be sure about one theory over another

 

[EdZ]

Under the IHS, I don't think there's any 'glass'. It's the backside of the CPU die itself, which is crystalline elemental Silicon. Amorphous Silicon (which would be a Silicon glass) appears to be used mainly in Optoelectronics (mainly solar cells and photonic devices), and I cant find any references to0 anyone using it for fabricating regular ICs in production.
Most of the data I can find on Gallium diffusion into Silicon is at high temperatures (1000°C region), but I'm personally wary of putting Gallium directly on any metal.

 

[Phuncz]

Ah nice to know, thanks ! I'm learning so much from this forum alone, where's the donate button ?

 

[jeshikat]

I'm learning so much from this forum alone, where's the donate button ?

I may do it in the future. Right now the hosting costs are pretty minimal so I'm just paying out of pocket.
As the forum grows and hosting costs increase I may add a donate button so I can avoid having ads, I love how fast and smooth the forum runs without bloated advertisements slowing it down

 

[WiSK]

Under the IHS, I don't think there's any 'glass'. It's the backside of the CPU die itself, which is crystalline elemental Silicon. Amorphous Silicon (which would be a Silicon glass) appears to be used mainly in Optoelectronics (mainly solar cells and photonic devices), and I cant find any references to0 anyone using it for fabricating regular ICs in production.

Chips are sealed with either silicon nitride or silicon dioxide. IIRC the former is that kind of matte black material, the latter is pure glass.

Edit. I was meaning to find a source to corroborate, but don't have time, wife gone out I have the kids. Maybe I'm just making the assumption because once I had wiped off the TIM, whatever the die is sealed with looks and feels like glass. Perfectly reflective without more than a quick wipe with iso alcohol.


Edit2. Just realised how off topic this is in your case thread, sorry QinX!