• Save 20% on ALL SFF Network merch, until Dec 31st! Use code SFF2024 at checkout. Click here!

!nverse: A Highly Versatile Steam-Box Design

Hahutzy

Airflow Optimizer
Sep 9, 2015
252
187
Hmm, I'm wondering if the brand of your unshielded PCIe riser cable is super amazing? Could you reveal which PCIe cable you have and then some of us can test it? I think I might test it on my second prototype.

I just searched ebay for 250mm PCI-E 16x cable and got the cheapest one. I ripped the first one at the PCB solder points while testing with the 750 Ti, and ordered a second one again. This one I tested (and still am using with) the 980 Ti. I ordered a third one just in case and it's just sitting on my shelf now.

I think iFreilicht is correct in that the EVGA 980 Ti has a backplate, and that shielded the EMI to/from the ribbon and the GPU.

Also, how much clearance does !nverse have between the casing aluminum to the ribbon, and then the ribbon to the backplate of the GPU? If its less than what Hassium has (~5mm), then there's something we can point at.
 
Last edited:
  • Like
Reactions: esplin2966

esplin2966

Cable-Tie Ninja
Original poster
Mar 2, 2015
169
113
I just searched ebay for 250mm PCI-E 16x cable and got the cheapest one. I ripped the first one at the PCB solder points while testing with the 750 Ti, and ordered a second one again. This one I tested (and still am using with) the 980 Ti. I ordered a third one just in case and it's just sitting on my shelf now.

I think iFreilicht is correct in that the EVGA 980 Ti has a backplate, and that shielded the EMI to/from the ribbon and the GPU.

Also, how much clearance does !nverse have between the casing aluminum to the ribbon, and then the ribbon to the backplate of the GPU? If its less than what Hassium has (~5mm), then there's something we can point at.

Can you post the link to the ebay site? I'm buying one haha.

In !nverse, the back of the GPU is essentially touching the riser cable, and the riser cable is essentially touching the case. I'm not sure if this distance is the issue though, since as I've mentioned, I have taken the GPU out of the case (still connected to Mobo with riser cable) and tried running the PC. It is able to boot, but still crashes during unigine valley.
 

Stevo_

Master of Cramming
Jul 2, 2015
449
304
Something missing in the discussion besides the EMI goofing things up, there are also reflections caused every time impedance changes in a transmission line. This means at every connector and if the board traces at both ends aren't matched well with the extender it's like a double whammy. There's a simple formula based on the impedance mismatch to determine how bad the reflection will be, but these reflections also end up bouncing back and forth between mismatches in a decay pattern sometimes adding sometimes subtracting from the signals in the conductor, can look really ugly on a scope. Bottom line, the quality of boards at both ends(connectors as well as characteristic impedance of the traces) can also affect the apparent performance of the riser or extender. This may account for some of the variability in results.
 

|||

King of Cable Management
Sep 26, 2015
775
759
On high-end motherboards (we're talking $1000+ rugged boards and such) they will alleviate some of the signal reflections by either back-drilling the through-hole soldered pins or by using blind vias where the solder connection ends at the plane of the motherboard where the traces are. The Asus Auto-Extreme process appears to be using blind vias, so I'm interested in seeing that being used more wide spread.
 

jeshikat

Jessica. Wayward SFF.n Founder
Silver Supporter
Feb 22, 2015
4,969
4,783
It's this quoted block:

There are a number of contributors to problems. The first would be that with the cables that you are using, they are not impedance matched, and may not be low loss dielectric. Thus, attenuation, cross talk and reflections will close the eyes, due to loss and noise. Reflections are the result of impedance discontinuities. Our assemblies are tuned to 85 ohms, cable and PCB. PCIe Gen3 is 85 ohms. 85 and 100 ohms work interchangeably, but if mixed, you lose some signal to reflection and shorten the maximum possible length. If your performance is on the edge, that's not a good thing. Our cable is tightly coupled and uses silver plated conductors. Lower loss for conductivity, and for radiative losses coupling to ground. And we use a low loss dielectric. Impedance matching is one thing, but dielectrics absorb some of the energy too. The only thing separating these assemblies from what is used in supercomputers is the connector, and we have no control over that. The connectors are geared at being cheap and good enough, by Intel and the industry. By and large, they are.

As for crosstalk, our cable exhibits less than -40 db of crosstalk. Crosstalk can occur at the termination, since the shielding has been removed. We design the PCB termination and stackup to minimize the effects. Again, the other cable is not shielded, so there could be considerable cross talk along it's length. Between attenuation, crosstalk, and reflections there may be no eye at the far end. So for dual GPU, I think it is simply a cable performance issue. The other cable assembly was probably designed with PCIe Gen I in mind. I have seen some that advertise Gen 2, but the longest that I recall there was around 6 inches. And the prices were considerably higher than Gen I.
 
  • Like
Reactions: iFreilicht

Stevo_

Master of Cramming
Jul 2, 2015
449
304
Used to use the book below as a bible more or less back when I designed boards for use in Mercury Systems(big area jobs size-wise), still have it around somewhere I believe. Good reference.

https://books.google.com/books/about/High_speed_Digital_Design.html?id=H5SsQgAACAAJ

Used to get boards made by Multiwire(now Hitachi apparently), they could even run coax in the board, impedance and run length control like no other. Of course money was no object to the "man" back then ;)

http://www.hitachi-chemical.com/products_pwb_05.htm
 
  • Like
Reactions: jØrd

esplin2966

Cable-Tie Ninja
Original poster
Mar 2, 2015
169
113
So is this still going on?

Yes it is. Not much to update because we're still waiting for our second prototype. There was a miscommunication with our manufacturer that resulted in multiple prototypes being made incorrectly, which is why it's been taking so long :(

I believe we got the current one right though, and it should be shipping out this week :)
 

esplin2966

Cable-Tie Ninja
Original poster
Mar 2, 2015
169
113
So do you think the Fury X or similar cards could be used in this case?

I don't think Fury X specifically fits, since the radiator is too thick. However, if there is a card out there with 27mm radiators and FEP tubing, it should fit fine. Keep in mind that the case fits 1 x 120mm radiator or 1 x 240mm radiator, not 2 x 120mm radiators.
 

EdZ

Virtual Realist
May 11, 2015
1,578
2,107
IIRC the Fuxy X uses pump-on-block, so you could just swap to a different radiator without needing to find space for a pump.
 

iFreilicht

FlexATX Authority
Feb 28, 2015
3,243
2,361
freilite.com
True. I don't know how complicated it would be to swap the tubing on it, so I just assumed you'd remove the whole AIO assembly, thought that would be much more costly.
You'd still need a reservouir, though, right? Or is it possible to fill a loop without having one?
 

CC Ricers

Shrink Ray Wielder
Bronze Supporter
Nov 1, 2015
2,234
2,557
I've seen a RVZ01 watercooled build with no reservoir, so it's definitely possible. You just have to take more care in bleeding the loop.
 

Ricky

SFF Lingo Aficionado
Jan 23, 2016
92
36
See, I just thought that with cards like the Fury X (and hopefully future HBM2 cards), that the length would be short enough that I could replace one of the two blue 120mm fans with the radiator. But I have no build experience, so I guess it's naíve thinking on my part.