|25th Jan 2013, 14:58||#1|
What's a Dremel?
Join Date: Jan 2013
Location: Debrecen, Hungary
Passively air-cooled GTX570 SLI - April, 2013
Hello and welcome from Hungary!
The project I would like to introduce to you in this thread is actually finished from the design point of view.
This is how it looks like:
However, from a technical point of view, there is a long way to go. The reason for me posting this on bit-tech.net forums is that some days ago, personnel from a Hungarian IT website, where I first posted my project in Hungarian, suggested me to turn to the members of this forum, if I wish to get valuable advice on the problem, because of which I cannot declare my build 100% ready.
I believe my English to be good, but not perfect, sorry for any mistakes!
This below is a short sum-up of the story:
We, together with my father, built this PC from January, 2012 to September, 2012. Being a hobby PC enthusiast, it was my long lasting obsession to build a PC, where stack-, or chimney-effect keeps a strong config at low temperatures. My father is an engineer, after many "leave me alone"s, he finally committed himself to helping me out with my "madness". The key elements of the design were in my mind already. At first, I had to find two long and extra quality riser cables. After weeks of searching the internet, I managed to obtain two 12 inches (30,5cms) PCIe GEN2 flexible riser cables. With this freedom in setup, we fitted CPU heatsinks to the GTX570 GPUs, together with custom made copper PCB heatsinks on each of the ASUS ENGTX570DCII cards. By doing this, we were able to insert the cards into my old PC case in a configuration that corresponded to our needs of passive cooling. However, it turned out that because of the huge heatsinks on the GF110 GPUs, we will need a minimum of 22cms long SLI bridge to connect the two cards in SLI. Again, after some weeks of searching through and out the internet, I realized that SLI bridges are only available up to a maximum of 14cms in length, so we decided to connect two ribbon bridges (one 14cms long and another 12cms long) by soldering, to make it a 26cms long piece. With this extra long bridge, we could finish our build in September, last year.
To prevent and/or interrupt overheating of the GPUs, my father suggested me to fit 2 fan phases to the case. The upper section with the three San Ace 109R1212H1011s constitutes an "emergency break". With these top notch (and very rare) fans, I am be able to interrupt overheating through manual (wirewound potentiometer) control if degrees are rising high. The bottom section with the PWM Thermalright fans was designed to just maintain some soft air movement in the case, if our planned full passive operation turns out to be a failure. The Thermalrigth fans are controlled through the PWM headers of the graphics cards.
Here are the specs:
- ASUS Maximus GeneZ/Gen3
- i7-2600s with Thermalright Silver Arrow SB-E
- 1TB EARX WD with Silentmaxx HDD silencer
- 64GB Samsung 830 SSD with Scythe Himuro mini
- 2X4GB Kingston Blue with Thermalright HR-07
- 2X ASUS ENGTX 570DCII with Thermalright Macho heatsinks applied to the GPU, 3,5 mm copper PCB heatsinks to the VRMs and RAM modules and Alpenföhn Ötzis to the PCB heatsink.
- 2X 12 coll (30,5 cm) flex GEN2 risers
- 26cm (10.2 inches), hand-soldered SLI bridge
- Seasonic X-760
- Chieftec Matrix (MX-01WD-D) from my old rig
- Phobya XT gap-pad / AC MX-2 thermal compound
- 3X SANYO DENKI San Ace 109R1212H1011 fans
- 2X TY-140 fans
- 25 Watt, 20 Ohm wirewound potentiometer, VU meter (to control and check voltage for the San Aces)
- 3X 110 mm, 66 degree sewage pipe elbows
- 2 square meters of thermal foil to keep, and thus, elevate heat within the case
Here comes the pics section with some comments:
This is how it started, with remnants of my old rig (ABIT IC7-G; NV6800; 3,06Ghz Northwood):
Holes cut (seems easier than it actually was to cut them):
San Ace 9G1212H1011s fitted. After some weeks, I thought 109Rs were no more available, so I bought these first.
26cms (10.2 inches) long, hand-soldered SLI bridge:
ASUS ENGTX570 full copper PCB heatsink (hand-cut with a fretsaw):
Two 12 inches long risers:
Custom wall stand:
Seasonic x-760 with custom consoles:
RAM modules with Thermalright HR-07:
Brand new San Ace 109R1212H1011s. Finally found them:
VGA under construction - part one:
VGA under construction - part two:
Storage block (hanged on 4 springs at the corners to eliminate HDD resonance):
VGAs ready (1.):
VGAs ready (2.):
VGAs ready (3.):
VGAs ready (4.):
Thermalright Silver Arrow SB-E:
Motherboard tray (made from 6mms thich aluminum):
Motherboard cover (plexiglass cut again with a fretsaw - took two days):
Motherboard block assembled:
All components ready for assembly:
Assembly - part one (40mm flexible aluminum heat tubes bringing fresh air to the lower components of the motherboard):
Assembly - part two:
Assembly - part three:
Final product (1.):
Final product (2.):
Final product (3.):
It was after finishing the building process and after having started testing that we found the problem, because of which I am now turning to you for some advice:
I call this the "blu(e)ish looming" or "blue shading" effect, meaning that when I enable SLI in nVIDIA Forceware (ver. 295.73 at present, tried newer Forcewares, too, but found the same problem), barely lit and beclouded, foggy areas on the screen in games, but also in 3DMARK06 to 3DMARK11 start to shade very-very light blue and some light blue curves and stripes (artifacts) appear on the screen. This happens only when running 3D games, irrespective of the game I choose. When running 2D applications (for ex. MS Word), there are no artifacts, even if SLI is enabled. Again, there are no artifacts when SLI is disabled, at all. Here are two pictures I took:
For more detailed mages, please, use these links (one and two). Both pics were taken with my Samsung EX-1 camera to try to let you see the nature of the problem. Monitor is a DELL U2711, tried it with other LCDs, too, but no matter what monitor I use, the problem still prevails. Please, look for those light blue stripes and curves on the screen! However, when disabling SLI in Forceware, this "looming" effect stops, the picture turns perfect with one card. In SLI, the GPUs are working properly, they are heating up and produce good SLI (almost doubled 3DMARK11) results as opposed to the one-card setup, but the blue shading remains a problem up until I disable SLI. From this situation, I came to a (mistaken ?) conclusion that our problem has to do something with the extra long SLI bridge that I hand-soldered, as without the cards communicating through the bridge, picture quality remains perfect. I have already tried to wrap my SLI bridge into aluminum foil that I grounded to the PC case, but the blue shading persists.
Some full passive test result (SLI enabled,case side open, none of the fans running):
After running 3DMark06 once:
GPU1: max usage %: 98, max temp: 68 C
GPU2: max usage %: 98, max temp: 66 C
After running 3DMark06 seven (7) consequtive times:
GPU1: max usage %: 98, max temp: 85 C
GPU2: max usage %: 99, max temp: 80 C
Should you have any question, or should you have any idea or information about
- the theoretical maximum for the length of an SLI bridge,
- how to remove those blue artifacts,
- how to obtain a 22cms (8.7 inches) long factory SLI bridge,
do not hesitate to post a reply!
I only have the time to check these forums once or twice a day, sorry for that in advance.
Thank you and best regards!
Today, I did the testing with an "off-the-shelf" 90mms long SLI bridge. SLI enabled, no artifacts at all .
A finally got my SLI bridge ready. 400mms long. Testing is expected to happen the forthcoming weekend. Here is a picture of the bridge:
Last edited by jippa; 10th Apr 2013 at 22:28.
|1st Feb 2013, 00:41||#2|
Join Date: Sep 2005
Welcome to the forums and Nice project.
Since I have a passion for passive cooling aswell, I really love the engineering in this. I've never worked with sli-setups and Im not familiar with the problems you are having. But just as you says, I would also belive it to be the SLI cable that cause the problem. There are a few things I would try.
First one would be shielding as you mention you have already tried, but make shure that the shielding create a good shield all the way to the tip of each connectors.
Another thing I would try is to redo the solderpoints incase you got a bad connection somwhere.
I also see that you have a big cluster of powercables verry close to the SLI cable. This might cause som interference. Maybe you could try to rout them away from the SLI cable and see if the problem changes. Maybe you could just hook up an external PSU incase you don't want to reroute everything just for testing.
The last thing I would try is probably the most timeconsuming one but the one I guess have the best possibility to solve the problem if it is caused by the cable. I would try to make a custom cable by only using the original connectors and replace the thin flexible PCB with thicker individual cables for each connector. Because you increase the length of the cable, there might be a problem with losses because of the cross-sectional area in the flex pcb being too small. You would need to increase this area with thicker cables to solve this.
As I said, Im not shure these suggestions would fix anything but thats what I would try.
I hope you get this problem solved.
Really nice to see a mod project that is mainly focused on the enginnering than cool design features.
Next time your father gives you a "leave me alone" moment, tell him that this will make you an engineer aswell. Many people see PC modding as "madness" but I think is's better than any school book available, so keep it up
|1st Feb 2013, 03:09||#4|
↓ Don't trust that face!
Join Date: May 2007
Holy COW! That's some fantastic work.
I'd go with what Gnu mentioned. (I've never touched SLI,) but it seems to mostly work for you. The only thing left is signal interference, either from the SLI bridge, or the PCIE extensions.
...Any relation to Jipa?
|1st Feb 2013, 08:34||#5|
Join Date: Oct 2008
Wow! Just Wow
Really like the fact that you did a lot of modifications that you would normally only see in scratchbuilds, not in existing cases!
The GPU cooling setup is awesome btw!
Last edited by Azariel; 1st Feb 2013 at 08:42.
|1st Feb 2013, 10:19||#6|
May contain traces of sanity.
Join Date: Apr 2011
Location: Lancashire, UK
Impressive! Love the work on the custom copper PCB heatsinks too. Very neat
Current Rig: Intel 3570K | Asus Z77 Sabertooth | 8GB Corsair Vengeance 1600 RAM | Samsung 240GB SSD | 2TB Seagate Barracuda 7200 | EVGA GTX680 SC Signature Editon | Asus Xonar D2X | Corsair HX850W | Fractal Design Define XL R2 (Black)
Cooled by: Aquaero 5 Pro | Alphacool NexXxos XP3 | Alphacool Monsta 240 | Laing D5
|1st Feb 2013, 16:28||#9|
Level 99 Retro Nerd
Join Date: Aug 2011
Dear lord. I've been here a while now and I've not seen something as crazy as this. Excellent work. Those temps look great too.
I have a passion for passive cooling. I hope to one day make a machine passively cooled with a minimum of fans. I like no noise =p
Main Rig: i3 3220 ~ GTX 670 4GB ~ 8GB Vengeance Ram ~ EVGA Supernova G2 750W ~ P8H77-m le ~ Samsung 830 128GB SSD ~ Steelseries GCv2 ~ Razer imperator ~
|2nd Feb 2013, 20:00||#11|
What's a Dremel?
Join Date: Jan 2013
Location: Debrecen, Hungary
Thank you, guys!
The ribbon bridge was tested with an ohm-meter millions of times, those solderpoints are ok.
I am ready to solder a hard-bridge, but it would be nice first to see some suggested technical specs for the cable type to be used with respect to the signal parameters transfered through the bridge. I need this signal to be kept intact.
...Any relation to Jipa? No.
|3rd Feb 2013, 04:05||#13|
Join Date: Jun 2010
Wow. Very interesting. Aircooling on another level for sure.
JAllen Labs GT1000- 3770K @ 4.5ghz, 32GB Dominators, ASUS 660GTX OC SLI, Samsung 840 Pro 256GB SSD, 4 drive RAID 0 and 3TB storage in a Silverstone Fortress GT03, Logitech G9X Gaming mouse on Razer eXact Mat, Custom built Fostex studio monitors, Yamaha amp, Nvidia Surround 3x23.5 AOC LED LCDs
|3rd Feb 2013, 12:29||#14|
MD of M.Z.A. Inc and Weasel Pimp
Join Date: Dec 2011
Location: Birmingham, England
This is utterly amazing.
i5 3570K@4.3, ASUS Maximus 5 Gene, EVGA GTX 970 Cheesecake, 4GB 1600MHz G.Skill RipjawsX CL6@2133 CL9, Enermax Revolution
XSPC Twin DDC Dual Bay Res, 2x Laing DDC-1T Plus, Alphacool NexXxoS XP³ Light - Acetal Edition, Alphacool NexXxoS NVXP GTX760 Full Cover Copper Block, Alphacool Heatmaster II, 2x 240mm Alphacool ST30 Full Copper Rads, 5x Xigmatec CLF Red 120mm Fans
|3rd Feb 2013, 21:45||#16|
Join Date: Feb 2011
Location: Estonia Pärnu
Well with the SLI problem of yours is common all most every SLI or Crossfire has the same problem it is call micro lag and this causes the artifacts.
|3rd Feb 2013, 21:52||#17|
Join Date: Nov 2012
Location: Zuich, Switzerland
This is really interesting. I just spend a lot of time planning a water cooling loop, pretty much as most people do in here, and you just do the most contrary thing
If it works, it's awesome. I really like it!
Going just from a physical approach, reaching a good passive air cooling is actually pretty hard, as air is a better isolation than cooling...
Are you going to paint hose murder huge air intakes on the top?
|3rd Feb 2013, 22:28||#19|
Resident Bulldozer Guru
Join Date: Feb 2011
Location: Northamptonshire Region.
Firstly; Holy crap, this is utterly ridiculous. I don't think I've seen anyone strapping HR-02 heatsinks to a GPU before.
Secondly; My first thought was noise on the ribbon cable. If it's as lazy a transmission method as I think; (As in; The ribbon Cable is acting simply as a frame-transmit system.) it may simply be the length causing it to act as an antenna for local radiowaves or similar.
Dozy Beast: AMD FX8120 | Thermalright Archon | Asus Crosshair V Formula | 16GB (4x4GB) Corsair Vengeance LP | Asus Xonar DX/XD | 2X SLI KFA2 550Ti White Edition | Silverstone SSTG1000W | Silverstone Raven RV-02B V1.7 | OCZ Agility 3 60GB | 2x WD 3TB RED RAID0 | Lite-on Blu-ray ROM
Dell U2913WM | Sennheiser HD 485s | Creative Gigaworks T40s | Logitech G13 | Microsoft Sidewinder X6 | Cyborg R.A.T 9
|3rd Feb 2013, 22:40||#20|
What's a Dremel?
Join Date: Jan 2013
Location: Debrecen, Hungary
At first, I would like to thank all of you the comments, again.
"Well with the SLI problem of yours is common all most every SLI or Crossfire has the same problem it is call micro lag and this causes the artifacts."
I would like to point out that when I am saying "artifacts", I am not talking about broken images, missing pixels or anything that you may usually call an "error". It is more like a typical visual manifestation of somekind of an interference. I neither consider this blueish looming a collateral damage to all SLI or Xfire configs, I have seen many of them, some had stability problems, but none of them had my stripes. However, since I have not seen micro lag in effect up to date, I am asking you if you can confirm that it is micro lag that I am dealing with?
Anyone seen micro lag in effect? Is this really something similar?
"Are you going to paint hose murder huge air intakes on the top?"
The top vent shafts have been painted red already. Their factory color is more like yellow.
|artifact, heatsink, passive cooling, sli bridge|